Limit memory usage by wasm code

[WhaleKit]

Currently there doesn't seem to be api to stop wasm code from hogging up ram, such as by growing tables and memories

I'm ready to start implementing it, but i'd like to discuss the API

• Something like "store.set_max_ram" might suffice, but for things that aren't in linear memory, ram consumption could differ between tinywasm versions
• we could have set_max_ram and also have user assign "cost" to each category (how much per reference in table, per call frame, per variable of each type on stack/in locals)
• we could have user callback that receives how much in every category is already allocated (in tables, linear memories, call frames, etc) before approving/denying new allocation

[WhaleKit]

On a related to "protecting from wasm code" note: I kind of just assumed without thinking that safely running potentially malicious code is a goal of tinywasm because it's a wasm interpreter
But actually it's a taller ask to make an interpreter that can also run even malicious code safely. So, is it a goal of the project in the first place?

[WhaleKit]

oh, aside from memories and tables, although tinywasm's stack has MAX_CALL_STACK_SIZE of 1024 - when host code calls wasm code, it needs to make new wasm stack. So, in presence of any imported host functions that call back into wasm, function frames with a lot of local and stack variables are also a concern for ram usage

[SeaRoll]

I am also interested in if this feature can be added 👀 running this on embedded devices and making sure only app crashes is way better than host dying too

[explodingcamera]

I am also interested in if this feature can be added 👀 running this on embedded devices and making sure only app crashes is way better than host dying too

Hi! v0.9.0 has added a lot more configurable limits and the option to trap on allocation failures + a new LinearMemory trait you can use to customize memory allocation in your own code, however this is still something I'd like to improve in the API (the trait will probably change a bit in the next version).

[SeaRoll]

I am also interested in if this feature can be added 👀 running this on embedded devices and making sure only app crashes is way better than host dying too

Hi! v0.9.0 has added a lot more configurable limits and the option to trap on allocation failures + a new LinearMemory trait you can use to customize memory allocation in your own code, however this is still something I'd like to improve in the API (the trait will probably change a bit in the next version).

Thanks! will take a look and see how it feels