Re: [PATCH v3 16/16] mm/mmap: Move may_expand_vm() check in mmap_region()

[Lorenzo Stoakes <lorenzo.stoakes@oracle.com>]

On Tue, Jul 09, 2024 at 10:42:41AM GMT, Liam R. Howlett wrote:
> Cc'ing Dave Hansen on this.
>
> * Liam R. Howlett <Liam.Howlett@oracle.com> [240708 16:43]:
> > * Lorenzo Stoakes <lorenzo.stoakes@oracle.com> [240708 08:53]:
> > > On Thu, Jul 04, 2024 at 02:27:18PM GMT, Liam R. Howlett wrote:
> > > > From: "Liam R. Howlett" <Liam.Howlett@Oracle.com>
> > > >
> > > > The MAP_FIXED page count is available after the vms_gather_munmap_vmas()
> > > > call, so use it instead of looping over the vmas twice.
> > >
> > > Predictably indeed you removed the thing I commented on in the last patch
> > > ;) but at least this time I predicted it! ;)
> > >
> > > >
> > > > Signed-off-by: Liam R. Howlett <Liam.Howlett@Oracle.com>
> > > > ---
> > > >  mm/mmap.c | 36 ++++--------------------------------
> > > >  1 file changed, 4 insertions(+), 32 deletions(-)
> > > >
> > > > diff --git a/mm/mmap.c b/mm/mmap.c
> > > > index b2de26683903..62edaabf3987 100644
> > > > --- a/mm/mmap.c
> > > > +++ b/mm/mmap.c
>
> ...
>
> > > >  static void __vma_link_file(struct vm_area_struct *vma,
> > > >  			    struct address_space *mapping)
> > > >  {
> > > > @@ -2946,17 +2925,6 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
> > > >  	pgoff_t vm_pgoff;
> > > >  	int error = -ENOMEM;
> > > >  	VMA_ITERATOR(vmi, mm, addr);
> > > > -	unsigned long nr_pages, nr_accounted;
> > > > -
> > > > -	nr_pages = count_vma_pages_range(mm, addr, end, &nr_accounted);
> > > > -
> > > > -	/* Check against address space limit. */
> > > > -	/*
> > > > -	 * MAP_FIXED may remove pages of mappings that intersects with requested
> > > > -	 * mapping. Account for the pages it would unmap.
> > > > -	 */
> > > > -	if (!may_expand_vm(mm, vm_flags, pglen - nr_pages))
> > > > -		return -ENOMEM;
> > > >
> > > >  	if (unlikely(!can_modify_mm(mm, addr, end)))
> > > >  		return -EPERM;
> > > > @@ -2987,6 +2955,10 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
> > > >  			vma_iter_next_range(&vmi);
> > > >  	}
> > > >
> > > > +	/* Check against address space limit. */
> > > > +	if (!may_expand_vm(mm, vm_flags, pglen - vms.nr_pages))
> > > > +		goto abort_munmap;
> > > > +
> > >
> > > I know you can literally only do this after the vms_gather_munmap_vmas(),
> > > but this does change where we check this, so for instance we do
> > > arch_unmap() without having checked may_expand_vm().
> > >
> > > However I assume this is fine?
> >
> > Thanks for pointing this out.
> >
> > The functionality here has changed
> > --- from ---
> > may_expand_vm() check
> > can_modify_mm() check
> > arch_unmap()
> > vms_gather_munmap_vmas()
> > ...
> >
> > --- to ---
> > can_modify_mm() check
> > arch_unmap()
> > vms_gather_munmap_vmas()
> > may_expand_vm() check
> > ...
> >
> > vms_gather_munmap_vmas() does nothing but figures out what to do later,
> > but could use memory and can fail.
> >
> > The user implications are:
> >
> > 1. The return type on the error may change to -EPERM from -ENOMEM, if
> > you are not allowed to expand and are trying to overwrite mseal()'ed
> > VMAs. That seems so very rare that I'm not sure it's worth mentioning.
> >
> >
> > 2. arch_unmap() called prior to may_expand_vm().
> > powerpc uses this to set mm->context.vdso = NULL if mm->context.vdso is
> > within the unmap range.  User implication of this means that an
> > application my set the vdso to NULL prior to hitting the -ENOMEM case in
> > may_expand_vm() due to the address space limit.
> >
> > Assuming the removal of the vdso does not cause the application to seg
> > fault, then the user visible change is that any vdso call after a failed
> > mmap(MAP_FIXED) call would result in a seg fault.  The only reason it
> > would fail is if the mapping process was attempting to map a large
> > enough area over the vdso (which is accounted and in the vma tree,
> > afaict) and ran out of memory. Note that this situation could arise
> > already since we could run out of memory (not accounting) after the
> > arch_unmap() call within the kernel.
> >
> > The code today can suffer the same fate, but not by the accounting
> > failure.  It can happen due to failure to allocate a new vma,
> > do_vmi_munmap() failure after the arch_unmap() call, or any of the other
> > failure scenarios later in the mmap_region() function.
> >
> > At the very least, this requires an expanded change log.
>
> After doing a deep dive into the vdso issue, I think it would be best to
> remove the arch_unmap() call completely in a later patch set by changing
> the two areas highlighted by Dave in patch 5a28fc94c914 "x86/mpx,
> mm/core: Fix recursive munmap() corruption" back in 2019 in regards to
> the powerpc pointer use.  But that's for later work.

Our replies coincided but yes absolutely. But should be a separate patch,
agreed.

>
> In the above mentioned patch, the arch_unmap() was moved to an earlier
> time to avoid removing the same vma twice from the rbtree.  Since the
> mpx code no longer removes the vma and powerpc never removed the vma, it
> seems safe to reorder the calls as such:
>
> can_modify_mm() check
> vms_gather_munmap_vmas()
> may_expand_vm() check
> arch_unmap()
>
> This seems very much fine because:
> - powerpc is the only platform doing _anything_ in arch_unmap().
> - powerpc used to work with the arch_unmap() call  after the vma was
>   completely dropped.
> - The vma isn't even dropped by this point and so all proposed changes
>   will be completely undone in the rare case of may_expand_vm() failure.
> - The arch_unmap() call doesn't need to be that early anymore anyways
>   (mpx was dropped by Dave in 2020 git id ccaaaf6fe5a5).
>
> I will make the order change in v4 of the patch series in its own patch.
>

Great!

> Thanks,
> Liam