You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 

4.8 KiB

What is considered a security vulnerability?

If you are still unsure whether an issue you are filing is a security vulnerability or not after reading this page, always err on the side of caution and report it as a security vulnerability!

Bugs must affect a tier 1 platform or feature to be considered a security vulnerability.

The security of the host and integrity of the sandbox when executing Wasm is paramount. Anything that undermines the Wasm execution sandbox is a security vulnerability.

On the other hand, execution that diverges from Wasm semantics (such as computing incorrect values) are not considered security vulnerabilities so long as they remain confined within the sandbox. This has a couple repercussions that are worth highlighting:

  • Even though it is safe from the host's point of view, an incorrectly computed value could lead to classic memory unsafety bugs from the Wasm guest's point of view, such as corruption of its malloc's free list or reading past the end of a source-level array.

  • Wasmtime embedders should never blindly trust values from the guest — no matter how trusted the guest program is, even if it was written by the embedders themselves — and should always validate these values before performing unsafe operations on behalf of the guest.

Denials of service when executing Wasm (either originating inside compiled Wasm code or Wasmtime's runtime subroutines) are considered security vulnerabilities. For example, if you configure Wasmtime to run Wasm guests with the async fuel mechanism, and then executing the Wasm goes into an infinite loop that never yields, that is considered a security vulnerability.

Denials of service when compiling Wasm, however, are not considered security vulnerabilities. For example, an infinite loop during register allocation is not a security vulnerability.

Any kind of memory unsafety (e.g. use-after-free bugs, out-of-bounds memory accesses, etc...) in the host is always a security vulnerability.

Cheat Sheet: Is this bug considered a security vulnerability?

Type of bug At Wasm Compile Time At Wasm Execution Time
Sandbox escape - Yes
    Uncaught out-of-bounds memory access
- Yes
    Uncaught out-of-bounds table access
- Yes
    Failure to uphold Wasm's control-flow integrity
- Yes
    File system access outside of the WASI file system's mapped directories
- Yes
    Use of a WASI resource without having been given the associated WASI capability
- Yes
    Etc...
- Yes
Divergence from Wasm semantics (without escaping the sandbox) - No
    Computing incorrect value
- No
    Raising errant trap
- No
    Etc...
- No
Memory unsafety Yes Yes
    Use-after-free
Yes Yes
    Out-of-bounds memory access
Yes Yes
    Use of uninitialized memory
Yes Yes
    Etc...
Yes Yes
Denial of service No Yes
    Panic
No Yes
    Process abort
No Yes
    Uninterruptible infinite loops
No Yes
    User-controlled memory exhaustion
No Yes
    Uncontrolled recursion over user-supplied input
No Yes
    Etc...
No Yes

Note that we still want to fix every bug mentioned above even if it is not a security vulnerability! We appreciate when issues are filed for non-vulnerability bugs, particularly when they come with test cases and steps to reproduce!