|
|
|
# Security
|
|
|
|
|
|
|
|
One of WebAssembly (and Wasmtime's) main goals is to execute untrusted code in
|
|
|
|
a safe manner inside of a sandbox. WebAssembly is inherently sandboxed by design
|
|
|
|
(must import all functionality, etc). This document is intended to cover the
|
|
|
|
various sandboxing implementation strategies that Wasmtime has as they are
|
|
|
|
developed.
|
|
|
|
|
|
|
|
At this time Wasmtime implements what's necessary for the WebAssembly
|
|
|
|
specification, for example memory isolation between instances. Additionally the
|
|
|
|
safe Rust API is intended to mitigate accidental bugs in hosts.
|
|
|
|
|
|
|
|
Different sandboxing implementation techniques will also come with different
|
|
|
|
tradeoffs in terms of performance and feature limitations, and Wasmtime plans to
|
|
|
|
offer users choices of which tradeoffs they want to make.
|
|
|
|
|
|
|
|
## WebAssembly Core
|
|
|
|
|
|
|
|
The core WebAssembly spec has several features which create a unique sandboxed
|
|
|
|
environment:
|
|
|
|
|
|
|
|
- The callstack is inaccessible. Unlike most native execution environments,
|
|
|
|
return addresses from calls and spilled registers are not stored in memory
|
|
|
|
accessible to applications. They are stored in memory that only the
|
|
|
|
implementation has access to, which makes traditional stack-smashing attacks
|
|
|
|
targeting return addresses impossible.
|
|
|
|
|
|
|
|
- Pointers, in source languages which have them, are compiled to offsets
|
|
|
|
into linear memory, so implementations details such as virtual addresses
|
|
|
|
are hidden from applications. And all accesses within linear memory are
|
|
|
|
checked to ensure they stay in bounds.
|
|
|
|
|
|
|
|
- All control transfers—direct and indirect branches, as well as direct and
|
|
|
|
indirect calls—are to known and type-checked destinations, so it's not
|
|
|
|
possible to accidentally call into the middle of a function or branch
|
|
|
|
outside of a function.
|
|
|
|
|
|
|
|
- All interaction with the outside world is done through imports and exports.
|
|
|
|
There is no raw access to system calls or other forms of I/O; the only
|
|
|
|
thing a WebAssembly instance can do is what is available through interfaces
|
|
|
|
it has been explicitly linked with.
|
|
|
|
|
|
|
|
- There is no undefined behavior. Even where the WebAssembly spec permits
|
|
|
|
multiple possible behaviors, it doesn't permit arbitrary behavior.
|
|
|
|
|
|
|
|
## Defense-in-depth
|
|
|
|
|
|
|
|
While WebAssembly is designed to be sandboxed bugs or issues inevitably arise so
|
|
|
|
Wasmtime also implements a number of mitigations which are not required for
|
|
|
|
correct execution of WebAssembly but can help mitigate issues if bugs are found:
|
|
|
|
|
|
|
|
* Linear memories by default are preceded with a 2GB guard region. WebAssembly
|
|
|
|
has no means of ever accessing this memory but this can protect against
|
|
|
|
accidental sign-extension bugs in Cranelift where if an offset is accidentally
|
|
|
|
interpreted as a signed 32-bit offset instead of an unsigned offset it could
|
|
|
|
access memory before the addressable memory for WebAssembly.
|
|
|
|
|
|
|
|
* Wasmtime uses explicit checks to determine if a WebAssembly function should be
|
|
|
|
considered to stack overflow, but it still uses guard pages on all native
|
|
|
|
thread stacks. These guard pages are never intended to be hit and will abort
|
|
|
|
the program if they're hit. Hitting a guard page within WebAssembly indicates
|
|
|
|
a bug in host configuration or a bug in Cranelift itself.
|
|
|
|
|
|
|
|
* Where it can Wasmtime will zero memory used by a WebAssembly instance after
|
|
|
|
it's finished. This is not necessary unless the memory is actually reused for
|
|
|
|
instantiation elsewhere but this is done to prevent accidental leakage of
|
|
|
|
information between instances in the face of other bugs. This applies to
|
|
|
|
linear memories, tables, and the memory used to store instance information
|
|
|
|
itself.
|
|
|
|
|
|
|
|
* The choice of implementation language, Rust, for Wasmtime is also a
|
|
|
|
defense in protecting the authors for Wasmtime from themselves in addition to
|
|
|
|
protecting embedders from themselves. Rust helps catch mistakes when writing
|
|
|
|
Wasmtime itself at compile time. Rust additionally enables Wasmtime developers
|
|
|
|
to create an API that means that embedders can't get it wrong. For example
|
|
|
|
it's guaranteed that Wasmtime won't segfault when using its public API,
|
|
|
|
empowering embedders with confidence that even if the embedding has bugs all
|
|
|
|
of the security guarantees of WebAssembly are still upheld.
|
|
|
|
|
|
|
|
* Wasmtime is in the [process of implementing control-flow-integrity
|
|
|
|
mechanisms][cfi-rfc] to leverage hardware state for futher guaranteeing that
|
|
|
|
WebAssembly stays within its sandbox. In the event of a bug in Cranelift this
|
|
|
|
can help mitigate the impact of where control flow can go to.
|
|
|
|
|
|
|
|
[cfi-rfc]: https://github.com/bytecodealliance/rfcs/blob/main/accepted/cfi-improvements-with-pauth-and-bti.md
|
|
|
|
|
|
|
|
## Filesystem Access
|
|
|
|
|
|
|
|
Wasmtime implements the WASI APIs for filesystem access, which follow a
|
|
|
|
capability-based security model, which ensures that applications can only
|
|
|
|
access files and directories they've been given access to. WASI's security
|
|
|
|
model keeps users safe today, and also helps us prepare for shared-nothing
|
|
|
|
linking and nanoprocesses in the future.
|
|
|
|
|
|
|
|
Wasmtime developers are intimately engaged with the WASI standards process,
|
|
|
|
libraries, and tooling development, all along the way too.
|
|
|
|
|
|
|
|
## Terminal Output
|
|
|
|
|
|
|
|
If untrusted code is allowed to print text which is displayed to a terminal, it may
|
|
|
|
emit ANSI-style escape sequences and other control sequences which, depending on
|
|
|
|
the terminal the user is using and how it is configured, can have side effects
|
|
|
|
including writing to files, executing commands, injecting text into the stream
|
|
|
|
as if the user had typed it, or reading the output of previous commands. ANSI-style
|
|
|
|
escape sequences can also confuse or mislead users, making other vulnerabilities
|
|
|
|
easier to exploit.
|
|
|
|
|
|
|
|
Our first priority is to protect users, so Wasmtime now filters writes to output
|
|
|
|
streams when they are connected to a terminal to translate escape sequences into
|
|
|
|
inert replacement sequences.
|
|
|
|
|
|
|
|
Some applications need ANSI-style escape sequences, such as terminal-based
|
|
|
|
editors and programs that use colors, so we are also developing a proposal for
|
|
|
|
the WASI Subgroup for safe and portable ANSI-style escape sequence support, which
|
|
|
|
we hope to post more about soon.
|
|
|
|
|
|
|
|
## Spectre
|
|
|
|
|
|
|
|
Wasmtime implements a few forms of basic spectre mitigations at this time:
|
|
|
|
|
|
|
|
* Bounds checks when accessing entries in a function table (e.g. the
|
|
|
|
`call_indirect` instruction) are mitigated.
|
|
|
|
|
|
|
|
* The `br_table` instruction is mitigated to ensure that speculation goes to a
|
|
|
|
deterministic location.
|
|
|
|
|
|
|
|
* Wasmtime's default configuration for linear memory means that bounds checks
|
|
|
|
will not be present for memory accesses due to the reliance on page faults to
|
|
|
|
instead detect out-of-bounds accesses. When Wasmtime is configured with
|
|
|
|
"dynamic" memories, however, Cranelift will insert spectre mitigation for the
|
|
|
|
bounds checks performed for all memory accesses.
|
|
|
|
|
|
|
|
Mitigating Spectre continues to be a subject of ongoing research, and Wasmtime
|
|
|
|
will likely grow more mitigations in the future as well.
|