Or you can read the code. I didn't mention yet that NaCl is an open-source project. One interesting aspect of this is that it includes a compiler for building NaCl modules and this compiler is a variant of the gcc compiler. I'll be interested in seeing whether support for other languages follows, although some languages are of dubious value in an environment such as this. You'd only want languages that are meant to create high-performance native code.
Google says in the white paper that it has tested the inner sandbox heavily and believes it to be reliable and robust. The design of it makes this credible to me. I was struck by how, even though NaCl is x86-specific, NaCl modules themselves are very RISC-like. For instance, all data and instructions have to be word-aligned. This is probably for the convenience and performance of NaCl.
There is also an "outer-sandbox" in NaCl. It checks all system calls made from NaCl against a short whitelist of allowed ones and blocks everything else. Currently, only the Linux outer-sandbox is up and running; Mac and Windows are "works in progress."
Calls into and out of the run-time are performed through structures called springboards (calling into NaCl module code) and trampolines (calling out from modules). These interfaces are dangerous ones from a security standpoint and likely attack targets. Google included them in a list of NaCl components that needed extra scrutiny.
Google reports that, as a general matter, legacy Linux libraries port over with little trouble to NaCl. The company reports on some tests it did porting an H.264 decoder, a physics simulation and Quake. Performance tests on Quake showed it on par with a stand-alone Linux executable.
Google decided to make NaCl x86-specific so that it could optimize the sandbox better, both for security and for performance. Certainly it's true that almost every desktop and notebook computer runs x86, but not all browsing devices do, and I'm specifically thinking of mobiles. NaCl is also 32-bit only. These restrictions could turn out to be disappointing in the future.
Speaking as a programmer, I think NaCl is cool. Will people find a need for it? Is it a solution in search of a problem? Even though I like the idea of sandboxing, I'm not sure that the trade-offs NaCl makes are the right ones for the market. Perhaps a richer non-native code environment, such as Java or .NET, is a better answer for most of the potential programmers for NaCl. Surely there are programs that need native code performance; do they need to run in a browser, or even in a sandbox?
I applaud Google for all this research work. Whether it ends up as part of Google OS or just educating programmers on the best ways to secure code, the company is asking questions that need to be asked.
Security CenterEditor Larry Seltzer has worked in and written about the computer industry since 1983.
For insights on security coverage around the Web, take a look at eWEEK.com Security Center Editor Larry Seltzer's blog Cheap Hack.