As part of our work with the Open Technology Fund, we recently worked with the Tor Project to see how Tor Browser stands up in terms of modern exploit mitigations, and what could be done to make it harder to develop exploits for.
Tor Project has published a blog post that summarizes the report from their point of view and links to a number of issues on their bugtracker and other documentation.
This project was more of a research engagement than a security assessment — a lot of this engagement was identifying features that should be placed on the slider, and making recommendations for where they should land. But we looked at a lot of other more general hardening items too. We checked the status of DEP and ASLR on Windows and Mac, and found an interesting lack of exception handling on the Windows build, due to the MinGW build process (this throws SafeSEH and SEHOP out the window). We also went through, with the cooperation of the Mozilla Security team, and categorized over a hundred past security vulnerabilities in Firefox into feature category and bug type (Use-After-Free wins the latter overwhelmingly.) We analyzed a few public and private exploits, and also investigated enabling assertions in certain classes in Firefox. We have a skeleton patch for the latter, but it’s more a proof of concept than something we think they should use. One of the other major items was looking at replacing Firefox’s memory allocator (jemalloc) with a more hardened allocator (PartitionAlloc from Chrome). Fortunately, Mozilla makes this pretty easy, so most of the work is in adapting PartitionAlloc and making full use of its partition features. There are several other parts to the report, including looking at protocol handlers, media formats, and making regression tests for DOM object exposure.
We had a ton of fun working on this project and we’d like to thank Mike Perry at Tor for working with us so closely, OTF for sponsoring the work, and all the people inside iSEC and the security community we talked about this project with who gave us ideas — especially Chris Evans from Google (the author of PartitionAlloc). The report clocks in at about 30 pages, but with the appendices (which have patch files), it balloons up to a whopping 150 pages. You can find the report, and all the patch files in our publications repository.