update_flags does not add flags to parent page tables

[ChocolateLoverRaj]

Steps to reproduce:

• Create a mapping without USER_ACCESSIBLE
• Update the flags to include USER_ACCESSIBLE

Expected:
The page can be accessed by user mode.

Actual:
The page can't be accessed by user mode because even though the L1 entry has USER_ACCESSIBLE, the L2, L3, L4 entries may not be user accessible.

[Freax13]

Note that this isn't unique to USER_ACCESSIBLE, the same should be true for other bits e.g. WRITE and NO_EXECUTE.

I think we should be careful changing about update_flags's behaviors around this, it's possible that some users depend on the current behavior, so I'd rather not change it. That said I don't think we currently offer a good solution to this, so I'm open to suggestions :)

[ChocolateLoverRaj]

One idea is to set all the flags for all L4-L2 entries and only use L1 flags to control permissions.

[Freax13]

I don't think we should change the behavior of existing methods in such a drastic way. There might be people that depend on the current behavior.

[ChocolateLoverRaj]

What about if update_flags was left as is and there was a update_flags_2 function with the new behavior?

[Freax13]

I'm open to that idea (although I don't like that method name).

Another idea might be to improve TranslateResult to also contain the page table flags for the upper levels. With this information, a user can call set_flags_p{2,3,4}_entry as required. Changing TranslateResult would be a breaking change though...

[phil-opp]

In the Mapper::map_to function, we automatically set the PRESENT, WRITABLE and USER_ACCESSIBLE flags for the parent table if required. We also have Mapper::map_to_with_table_flags with an explicit parent_table_flags argument. It probably makes sense to provide this functionality also for flag updates.

I'm also against changing the behavior of update_flags directly since this would be a very subtle breaking change, but a new method with a different name sounds good to me. Maybe something like update_flags_recursive or update_flags_with_parents?

For completeness, we could also add a method like update_table_flags(page, flags, parent_table_flags) that allows you to specify the flags for parent table entries manually.

---

Including the parent table flags in TranslateResult sounds like a good idea too! So maybe we just do both?

[ChocolateLoverRaj]

I like the idea of a new function called update_flags_with_parents. I don't think I will be using the parent table flags so that change doesn't make a difference for my use.