multiboot2: fix memory issue in boxed_dst_tag

[phip1611]

This fixes multiple memory issues (wrong allocation and wrong fat pointer creation) that were introduced in #133 with 544e1ca.

I found the issues by running miri.

[phip1611]

@YtvwlD There were two memory issues in your boxed_dst_tag function. Not blaming you, just FYI :)

[YtvwlD]

Interesting. What's the difference?

Are you sure rounding up works for packed structs?

[phip1611]

Interesting. What's the difference?

1. Box::from_raw(ptr_meta::from_raw_parts_mut(ptr as *mut (), content.len()))
   By using content.len(), an error was introduced that may add more items to the slice than there are in content. Hence, access to possibly invalid memory.

   1. The value passed as size hint to ptr_meta::from_raw_parts_mut must be divided by the size of each item in the slice (see https://github.com/rust-osdev/multiboot2/blob/main/multiboot2/src/memory_map.rs#L70)
   2. It must be taken into account that for example for the ElfSectionsTag, content does contain more bytes than just for the dynamically sized slice part.

I totally confirm that it is confusing that the size hint is really just the amount of elements in the dynamically sized slice but not the struct itself.

2. There were allocation problems: wrong allocation size and wrong allocation alignment

This one wasn't obvious and I only discovered so because miri told me.

Are you sure rounding up works for packed structs?

Very well question. Thanks for the pointer :) I'm not sure.. can you give me an example please where you think it might fail?

[YtvwlD]

Ah, I think I caused that when moving the Box cast into the function instead of keeping in the news. (Because in a generic tag the content is just bytes.)

[phip1611]

Yeah, I thought so. It's a minefield, could have happened to me as well :)