On Fri, Nov 22, 2024 at 9:36 AM Abdiel Janulgue <abdiel.janulgue@xxxxxxxxx> wrote: > > On 22/11/2024 09:55, Alice Ryhl wrote: > > On Thu, Nov 21, 2024 at 9:18 PM Jann Horn <jannh@xxxxxxxxxx> wrote: > >> > >> On Wed, Nov 20, 2024 at 11:56 PM Abdiel Janulgue > >> <abdiel.janulgue@xxxxxxxxx> wrote: > >>> On 19/11/2024 19:07, Jann Horn wrote: > >>>>> + pub fn page_slice_to_page<'a>(page: &PageSlice) -> Result<&'a Self> > >>>> > >>>> Sorry, can you explain to me what the semantics of this are? Does this > >>>> create a Page reference that is not lifetime-bound to the PageSlice? > >>> > >>> This creates a Page reference that is tied to the lifetime of the `C > >>> struct page` behind the PageSlice buffer. Basically, it's just a cast > >>> from the struct page pointer and does not own that resource. > >> > >> How is the Page reference tied to the lifetime of the C "struct page"? > >> > >> I asked some Rust experts to explain to me what this method signature > >> expands to, and they added the following to the Rust docs: > >> > >> https://github.com/rust-lang/reference/blob/master/src/lifetime-elision.md > >> ``` > >> fn other_args1<'a>(arg: &str) -> &'a str; // elided > >> fn other_args2<'a, 'b>(arg: &'b str) -> &'a str; // expanded > >> ``` > >> > >> Basically, my understanding is that since you are explicitly > >> specifying that the result should have lifetime 'a, but you are not > >> specifying the lifetime of the parameter, the parameter is given a > >> separate, unrelated lifetime by the compiler? Am I misunderstanding > >> how this works, or is that a typo in the method signature? > > > > No, you are correct. The signature is wrong and lets the caller pick > > any lifetime they want, with no relation to the lifetime of the > > underlying `struct page`. > > But that could be put in the invariant that the PageSlice buffer must > last at least the lifetime `'a`? > > > > > From a C perspective, what are the lifetime requirements of vmalloc_to_page? > > > > If I'm not mistaken, that should be the lifetime of the vmalloc'd buffer > right? It seems to me that the signature should look like this: fn vmalloc_to_page(vec: &VVec<PageSlice>, i: usize) -> &Page This way, by providing the VVec, you can only use it with memory that really comes from a vmalloc call. Alice
