Re: [PATCH] [RFC] mm: mmap: Allow mmap(MAP_STACK) to map growable stack

[Helge Deller]

On 9/12/24 17:43, Liam R. Howlett wrote:
> * Helge Deller <deller@xxxxxx> [240911 22:10]:
> > On 9/12/24 03:32, Liam R. Howlett wrote:
> > > * Helge Deller <deller@xxxxxx> [240911 20:51]:
> > > > On 9/12/24 01:05, Liam R. Howlett wrote:
> > > > > * Yang Shi <shy828301@xxxxxxxxx> [240911 18:16]:
> > > > > > On Wed, Sep 11, 2024 at 12:49 PM Liam R. Howlett
> > > > > > <Liam.Howlett@xxxxxxxxxx> wrote:
> > > > > > > * Helge Deller <deller@xxxxxxxxxx> [240911 15:20]:
> > > > > > > > This is a RFC to change the behaviour of mmap(MAP_STACK) to be
> > > > > > > > sufficient to map memory for usage as stack on all architectures.
> > > > > > > > Currently MAP_STACK is a no-op on Linux, and instead MAP_GROWSDOWN
> > > > > > > > has to be used.
> > > > > > > > To clarify, here is the relevant info from the mmap() man page:
> > > > > > > >
> > > > > > > > MAP_GROWSDOWN
> > > > > > > >       This flag is used for stacks. It indicates to the kernel virtual
> > > > > > > >       memory system that the mapping should extend downward in memory.  The
> > > > > > > >       return address is one page lower than the memory area that is
> > > > > > > >       actually created in the process's virtual address space.  Touching an
> > > > > > > >       address in the "guard" page below the mapping will cause the mapping
> > > > > > > >       to grow by a page. This growth can be repeated until the mapping
> > > > > > > >       grows to within a page of the high end of the next lower mapping,
> > > > > > > >       at which point touching the "guard" page will result in a SIGSEGV
> > > > > > > >       signal.
> > > > > > > >
> > > > > > > > MAP_STACK (since Linux 2.6.27)
> > > > > > > >       Allocate the mapping at an address suitable for a process or thread
> > > > > > > >       stack.
> > > > > > > >
> > > > > > > >       This flag is currently a no-op on Linux. However, by employing this
> > > > > > > >       flag, applications can ensure that they transparently obtain support
> > > > > > > >       if the flag is implemented in the future. Thus, it is used in the
> > > > > > > >       glibc threading implementation to allow for the fact that
> > > > > > > >       some architectures may (later) require special treatment for
> > > > > > > >       stack allocations. A further reason to employ this flag is
> > > > > > > >       portability: MAP_STACK exists (and has an effect) on some
> > > > > > > >       other systems (e.g., some of the BSDs).
> > > > > > > >
> > > > > > > > The reason to suggest this change is, that on the parisc architecture the
> > > > > > > > stack grows upwards. As such, using solely the MAP_GROWSDOWN flag will not
> > > > > > > > work. Note that there exists no MAP_GROWSUP flag.
> > > > > > > > By changing the behaviour of MAP_STACK to mark the memory area with the
> > > > > > > > VM_STACK bit (which is VM_GROWSUP or VM_GROWSDOWN depending on the
> > > > > > > > architecture) the MAP_STACK flag does exactly what people would expect on
> > > > > > > > all platforms.
> > > > > > > >
> > > > > > > > This change should have no negative side-effect, as all code which
> > > > > > > > used mmap(MAP_GROWSDOWN | MAP_STACK) still work as before.
> > > > > > > >
> > > > > > > > Signed-off-by: Helge Deller <deller@xxxxxx>
> > > > > > > >
> > > > > > > > diff --git a/include/linux/mman.h b/include/linux/mman.h
> > > > > > > > index bcb201ab7a41..66bc72a0cb19 100644
> > > > > > > > --- a/include/linux/mman.h
> > > > > > > > +++ b/include/linux/mman.h
> > > > > > > > @@ -156,6 +156,7 @@ calc_vm_flag_bits(unsigned long flags)
> > > > > > > >          return _calc_vm_trans(flags, MAP_GROWSDOWN,  VM_GROWSDOWN ) |
> > > > > > > >                 _calc_vm_trans(flags, MAP_LOCKED,     VM_LOCKED    ) |
> > > > > > > >                 _calc_vm_trans(flags, MAP_SYNC,       VM_SYNC      ) |
> > > > > > > > +            _calc_vm_trans(flags, MAP_STACK,      VM_STACK     ) |
> > > > > > >
> > > > > > > Right now MAP_STACK can be used to set VM_NOHUGEPAGE, but this will
> > > > > > > change the user interface to create a vma that will grow.  I'm not
> > > > > > > entirely sure this is okay?
> > > > > >
> > > > > > AFAICT, I don't see this is a problem. Currently huge page also skips
> > > > > > the VMAs with VM_GROWS* flags set. See vma_is_temporary_stack().
> > > > > > __thp_vma_allowable_orders() returns 0 if the vma is a temporary
> > > > > > stack.
> > > > >
> > > > > If someone is using MAP_STACK to avoid having a huge page, they will
> > > > > also get a mapping that grows - which is different than what happens
> > > > > today.
> > > > >
> > > > > I'm not saying that's right, but someone could be abusing the existing
> > > > > flag and this will change the behaviour.
> > > >
> > > > Wouldn't a plain mmap() followed by madvise(MADV_NOHUGEPAGE) do exactly that?
> > > > Why abusing MAP_STACK for that?
> > >
> > > I can think of two answers:
> > > 1. An error that has worked without issues so far
> > > 2. One less system call
> > >
> > > I'm not saying this really is a blocker, but the change is not without
> > > risk as it does change behaviour the user could see.
> > >
> > > Interestingly enough, the man page is incorrect as it is written because
> > > the flag is not strictly a no-op; it ensures no huge pages.  So the
> > > feature of applying VM_NOHUGEPAGE with the use of MAP_STACK is not
> > > documented today.
> >
> > Yes.
> >
> > > What happens to call that use the mmap(MAP_GROWSDOWN | MAP_STACK) on
> > > parisc today?
> >
> > Today, without my patch, on parisc the area is then flagged VM_GROWSDOWN only.
> > In that case, stack expansion will not work, as it checks for VM_GROWSUP.
> >
> > > How does this change with your VM_STACK change?  Wouldn't this result
> > > in failed mappings?
> > > VM_GROWSDOWN | VM_GROWSUP would fail in do_mmap(), and these would be> set if you map MAP_STACK to VM_STACK which is defined as VM_GROWSUP?
> >
> > Right, with my change, the area will be tagged VM_GROWSUP and VM_GROWSDOWN.
> > Due to VM_GROWSUP stack expansion works.
> > The mapping doesn't fail in do_mmap(), because stacks are not file-mapped
> > or shared or droppable. They should be mapped with MAP_PRIVATE, right?
>
> Oh my, yes.  So now you will have a stack that can expand in either
> direction, but it's okay because one direction isn't checked.  I sure
> hope the rest of the code is correctly #ifdef'ed for this.

Only one direction will be handled (depending on the architecture),
even if both (VM_GROWSUP and VM_GROWSDOWN) are set.
So it shouldn't be a problem to have both directions set.

> > Another option is to introduce an alias, e.g.:
> > #define MAP_STACK_EXPANDABLE	MAP_GROWSDOWN
> > and then
>
> I don't like either of these options.
>
> I guess you could also detect the MAP_STACK and MAP_GROWSDOWN and change
> the flags for parisc, which I also don't like, but since parisc is the
> only arch using this it's hard to justify a change that may cause issues
> in other archs.

Well, I still don't actually see real issues with my proposal.
That's why I proposed to change MAP_STACK generally.
Just adapting it for parisc was my initial approach which I sent
to the parisc mailing list prior to this patch:
https://patchwork.kernel.org/project/linux-parisc/patch/Zt3yJUiczUBmEC3a@p100/

I'd like to wait for some possible further feedback.
In the end I might end up just changing it for parisc.

> A quick grep shows we set VM_STACK_DEFAULT_FLAGS in x86 and powerpc,
> which could affect what happens here with your change.  I am concerned
> about the bleeding of other flags through this change.

Shouldn't be a problem. The VM_STACK in VM_STACK_FLAGS is already VM_GROWSUP
or VM_GROWSDOWN.
include/linux/mm.h:#define VM_STACK_FLAGS       (VM_STACK | VM_STACK_DEFAULT_FLAGS | VM_ACCOUNT)

Helge