Context-awareness

[Xe]

It would be interesting to provide context-awareness to the exec.VM type. This would check the context Done every so often during execution, and if it is done then a new "took too long" error would be returned to the callee. This would make it possible for users to submit arbitrary untrusted webassembly code for a backend server to run and return the results without as much fear of users submitting code that blocks infinitely.

[ghost]

Great idea ! Was wondering about how to solve that corner case too

…

On Fri, 22 Jun 2018, 02:45 Christine Dodrill, ***@***.***> wrote: It would be interesting to provide context <https://godoc.org/context>-awareness to the exec.VM type. This would check the context Done every so often during execution, and if it is done then a new "took too long" error would be returned to the callee. This would make it possible for users to submit arbitrary untrusted webassembly code for a backend server to run and return the results without as much fear of users submitting code that blocks infinitely. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#62>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ATuCwqZsL6YWjlbZHjV5Q8M24PGZkSKxks5t_D4OgaJpZM4Uy-j9> .

[sbinet]

apologies for the belated answer.

yes, having a context aware exec.VM type would definitely be interesting.

now, the also interesting thing would be to decide until what "depth" we want to weave it through:

• down to each VM operation? (e.g. pushInt64, i32Load8s, ...)
• only at VM.ExecCode ?
• somewhere in the middle?
• piggy back on what the Go runtime does for preemption ? (ie: only check at the end of loops, function calls, ...)

[Xe]

• down to each VM operation? (e.g. pushInt64, i32Load8s, ...)

This would be beyond overkill and would be super performance impactful.

• only at VM.ExecCode ?

Probably check here too.

• somewhere in the middle?

I was thinking every n operations or on function/loop/block barriers.

[sbinet]

down to each VM operation? (e.g. pushInt64, i32Load8s, ...)

This would be beyond overkill and would be super performance impactful.

I agree, I was just being facetious :)

I was thinking every n operations or on function/loop/block barriers.

every n operation is probably simplest to explain and implement, while not being too intrusive.

[Xe]

How about every 65536 instructions? That constant is already used as the size of a memory page and it's a very nice, even number.

[sbinet]

SGTM