Re: [PATCH] Fixes: null pointer dereference in pfnmap_lockdep_assert

[Peter Xu]

On Fri, Oct 04, 2024 at 03:43:12PM +0530, Manas wrote:
> Hi Peter, thanks for reviewing.
> 
> On 03.10.2024 12:41, Peter Xu wrote:
> > On Thu, Oct 03, 2024 at 09:31:06PM +0530, Manas via B4 Relay wrote:
> > > From: Manas <manas18244@xxxxxxxxxxx>
> > > 
> > > syzbot has pointed to a possible null pointer dereference in
> > > pfnmap_lockdep_assert. vm_file member of vm_area_struct is being
> > > dereferenced without any checks.
> > > 
> > > This fix returns if vm_file member in vm_area_struct is NULL.
> > > 
> > > Reported-by: syzbot+093d096417e7038a689b@xxxxxxxxxxxxxxxxxxxxxxxxx
> > > Closes: https://syzkaller.appspot.com/bug?extid=093d096417e7038a689b
> > > ---
> > > This bug[1] triggers a general protection fault in follow_pfnmap_start
> > > function. An assertion pfnmap_lockdep_assert inside this function
> > > dereferences vm_file member of vm_area_struct. And panic gets triggered
> > > when vm_file is NULL.
> > > 
> > > This patch returns from the assertion pfnmap_lockdep_assert if vm_file
> > > is found to be NULL.
> > > 
> > > [1] https://syzkaller.appspot.com/bug?extid=093d096417e7038a689b
> > > 
> > > Signed-off-by: Manas <manas18244@xxxxxxxxxxx>
> > 
> > Thanks for the patch!
> > 
> > > ---
> > >  mm/memory.c | 3 +++
> > >  1 file changed, 3 insertions(+)
> > > 
> > > diff --git a/mm/memory.c b/mm/memory.c
> > > index 2366578015ad..b152a95e543f 100644
> > > --- a/mm/memory.c
> > > +++ b/mm/memory.c
> > > @@ -6346,6 +6346,9 @@ static inline void pfnmap_args_setup(struct follow_pfnmap_args *args,
> > >  static inline void pfnmap_lockdep_assert(struct vm_area_struct *vma)
> > >  {
> > >  #ifdef CONFIG_LOCKDEP
> > > +	if (!vma->vm_file)
> > > +		return;
> > > +
> > 
> > Hmm I guess I wasn't careful enough here as I was mostly only thinking
> > about file mappings, but I just notice we have other pfnmaps like the vvar
> > mappings.. the mapping var can also already be reused later when available.
> > 
> > Logically even if !vm_file we can still check against mmap write lock.  So
> > would it be better to do this instead:
> > 
> >        struct address_space *mapping = vma->vm_file && vma->vm_file->f_mapping;
> > 
> This will lead to `-Wint-conversion` error in the assignment. We can either do a
> cast like the following:
> 
>         struct address_space *mapping = (struct address_space *)(vma->vm_file && vma->vm_file->f_mapping);
> 
> But I am not sure if it is the canonical way of doing it. It will also lead to
> warning about pointer from integer casting.
> 
> Or will a conditional like this work here?
> 
>         struct address_space *mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;

Sorry, that was a pretty stupid mistake of mine, just to show what I meant
without any compilation tests.  Yes this one.

Thanks,

> >        if (mapping)
> >                lockdep_assert(lockdep_is_held(&mapping->i_mmap_rwsem) ||
> >                               lockdep_is_held(&vma->vm_mm->mmap_lock));
> >        else
> >                lockdep_assert(lockdep_is_held(&vma->vm_mm->mmap_lock));
> > 
> > ?
> > 
> 
> > >  	struct address_space *mapping = vma->vm_file->f_mapping;
> > > 
> > >  	if (mapping)
> > > 
> > > ---
> > > base-commit: 9852d85ec9d492ebef56dc5f229416c925758edc
> > > change-id: 20241003-fix-null-deref-6bfa0337efc3
> > > 
> > > Best regards,
> > > --
> > > Manas <manas18244@xxxxxxxxxxx>
> > > 
> > > 
> > > 
> > 
> > -- 
> > Peter Xu
> > 
> 
> -- 
> Manas
> 

-- 
Peter Xu