[_0w8t]

The problem was precisely the Pepper, a pile of API that Google thought they could force on everybody, not PNaCl per see. If the initial NaCl implementation just exposed the pre-existing Web API into the native code sandbox without adding anything extra, the story can be rather different and a version of PNaCl could be standardized.

[roca]

That's true, although NaCl's design made that difficult because the native code sandbox had to be a different process so interacting with the page's DOM would have required IPC.

[_0w8t]

I am curious what made it necessary to run NaCl in a different process? I thought the main idea behind NaCl was to allow the same-process native sandboxes.

[roca]

According to https://static.googleusercontent.com/media/research.google.c... NaCl does not sandbox loads, relying on address-space separation to ensure secret data is not leaked. Obviously this only works with a single sandboxed application per address space. (And even then you'd have to be pretty careful!)

[_0w8t]

This is only for NaCl on ARM or AMD64. The original NaCl for x86 uses the segment registers for isolation allowing to restrict both loads and stores only to the permitted addresses. That, as far as I understand, does allow to embed into 32-bit process without compromising secretes.

So as a speculation in an alternative world where Google has not developed Pepper, but bridged web api into x86 NaCl, the latter designs for x64 and ARM would restrict loads only from the allowed address space.

[roca]

https://static.googleusercontent.com/media/research.google.c... shows that the original NaCl-x86 only allowed one sandboxed application per process. So to support pages with more than one plugin, IPC would have been needed.

Sure, (P)NaCl could have been implemented differently in a way that allowed multiple sandboxed applications per process, and then DOM access would have been easier and maybe Pepper wouldn't have been necessary, though there would have been slightly higher overhead I guess.

That's all history now.

[johncolanduoni]

Probably for the same reason Chrome runs V8 in a different process: planning for inevitable vulnerabilities in the implementation.

[bzbarsky]

Chrome doesn't run V8 in a different process from the page DOM. That's the difference: NaCl/PNaCl _does_ run in a different process from the page DOM, so interacting with the DOM gets complicated.