Harden validation of remote debugging offset tables

[pablogsal]

_remote_debugging should perform stricter validation of externally read debug-offset metadata before using it to size memory reads or interpret remote process layouts. The current attach path verifies version compatibility, but it should also reject malformed or inconsistent offset and size tables up front, including optional asyncio metadata when it is loaded, so invalid data is handled cleanly rather than being consumed by the unwinder.

Linked PRs

• gh-148178: Validate async debug offsets read from remote process #148179
• gh-148178: Validate remote debug offset tables on load #148187

[gpshead]

My own claude also independently found this (very much a theme at the moment in time as all sorts of people are realizing what the tools they've got can do) so I had one variant of a PR waiting for this being a public issue. attached as a draft in case you find it useful.

[pablogsal]

My own claude also independently found this (very much a theme at the moment in time as all sorts of people are realizing what the tools they've got can do) so I had one variant of a PR waiting for this being a public issue. attached as a draft in case you find it useful.

Thanks a lot @gpshead! I answer in the PR but I am afraid the fix requires much more than just checking for the async debug as we need to validate all the offsets somehow and also ensure there aren't other vectors for the same thing :(

[gpshead]

yeah i figured this probably went beyond just that very targeted one...

[pablogsal]

Alright I will prepare a PR. At the very least we need to:

• Validate all fields in the async debug and regular offsets fields.
• Ensure that if someone adds a new field we fail at compile time if a validation is not added
• Sanitize other vectors for the same problem.
• Do this in a way that doesn't have runtime performance impact