[PATCH] mmap: Fix do_brk_flags() modifying obviously incorrect VMAs

[PATCH] mmap: Fix do_brk_flags() modifying obviously incorrect VMAs

[Liam Howlett]

Add more sanity checks to the VMA that do_brk_flags() will expand.
Ensure the VMA matches basic merge requirements within the function
before calling can_vma_merge_after().

Drop the duplicate checks from vm_brk_flags() since they will be
enforced later.

Fixes: 2e7ce7d354f2 ("mm/mmap: change do_brk_flags() to expand existing VMA and add do_brk_munmap()")
Suggested-by: Jann Horn <jannh@google.com>
Signed-off-by: Liam R. Howlett <Liam.Howlett@oracle.com>
---
 mm/mmap.c | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)

diff --git a/mm/mmap.c b/mm/mmap.c
index a5eb2f175da0..41a2c42593e8 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -2946,9 +2946,9 @@ static int do_brk_flags(struct ma_state *mas, struct vm_area_struct *vma,
 	 * Expand the existing vma if possible; Note that singular lists do not
 	 * occur after forking, so the expand will only happen on new VMAs.
 	 */
-	if (vma &&
-	    (!vma->anon_vma || list_is_singular(&vma->anon_vma_chain)) &&
-	    ((vma->vm_flags & ~VM_SOFTDIRTY) == flags)) {
+	if (vma && vma->vm_end == addr && !vma_policy(vma) && vma->anon_vma &&
+	    can_vma_merge_after(vma, flags, NULL, NULL,
+				addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) {
 		mas_set_range(mas, vma->vm_start, addr + len - 1);
 		if (mas_preallocate(mas, vma, GFP_KERNEL))
 			return -ENOMEM;
@@ -3035,11 +3035,6 @@ int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags)
 		goto munmap_failed;
 
 	vma = mas_prev(&mas, 0);
-	if (!vma || vma->vm_end != addr || vma_policy(vma) ||
-	    !can_vma_merge_after(vma, flags, NULL, NULL,
-				 addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL))
-		vma = NULL;
-
 	ret = do_brk_flags(&mas, vma, addr, len, flags);
 	populate = ((mm->def_flags & VM_LOCKED) != 0);
 	mmap_write_unlock(mm);
-- 
2.35.1

Re: [PATCH] mmap: Fix do_brk_flags() modifying obviously incorrect VMAs

[Jann Horn]

On Mon, Dec 5, 2022 at 5:53 PM Liam Howlett <liam.howlett@oracle.com> wrote:
> Add more sanity checks to the VMA that do_brk_flags() will expand.
> Ensure the VMA matches basic merge requirements within the function
> before calling can_vma_merge_after().
>
> Drop the duplicate checks from vm_brk_flags() since they will be
> enforced later.

Looks good to me, with one note:

> Fixes: 2e7ce7d354f2 ("mm/mmap: change do_brk_flags() to expand existing VMA and add do_brk_munmap()")
> Suggested-by: Jann Horn <jannh@google.com>
> Signed-off-by: Liam R. Howlett <Liam.Howlett@oracle.com>
> ---
>  mm/mmap.c | 11 +++--------
>  1 file changed, 3 insertions(+), 8 deletions(-)
>
> diff --git a/mm/mmap.c b/mm/mmap.c
> index a5eb2f175da0..41a2c42593e8 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -2946,9 +2946,9 @@ static int do_brk_flags(struct ma_state *mas, struct vm_area_struct *vma,
>          * Expand the existing vma if possible; Note that singular lists do not
>          * occur after forking, so the expand will only happen on new VMAs.
>          */
> -       if (vma &&
> -           (!vma->anon_vma || list_is_singular(&vma->anon_vma_chain)) &&
> -           ((vma->vm_flags & ~VM_SOFTDIRTY) == flags)) {
> +       if (vma && vma->vm_end == addr && !vma_policy(vma) && vma->anon_vma &&

Why the "vma->anon_vma" check here? The old code was checking that the
existing VMA is not attached to more than one anon_vma; but the new
code instead checks that the existing VMA is attached to at least one
anon_vma, and then is_mergeable_anon_vma() checks that the VMA is not
attached to more than one anon_vma, so in effect the VMA has to be
attached to exactly one anon_vma. Is that intentional?

If not, maybe delete the "vma->anon_vma &&" - can_vma_merge_after()
already does the equivalent check of the old "(!vma->anon_vma ||
list_is_singular(&vma->anon_vma_chain))".

> +           can_vma_merge_after(vma, flags, NULL, NULL,
> +                               addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) {
>                 mas_set_range(mas, vma->vm_start, addr + len - 1);
>                 if (mas_preallocate(mas, vma, GFP_KERNEL))
>                         return -ENOMEM;
> @@ -3035,11 +3035,6 @@ int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags)
>                 goto munmap_failed;
>
>         vma = mas_prev(&mas, 0);
> -       if (!vma || vma->vm_end != addr || vma_policy(vma) ||
> -           !can_vma_merge_after(vma, flags, NULL, NULL,
> -                                addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL))
> -               vma = NULL;
> -
>         ret = do_brk_flags(&mas, vma, addr, len, flags);
>         populate = ((mm->def_flags & VM_LOCKED) != 0);
>         mmap_write_unlock(mm);
> --
> 2.35.1

Re: [PATCH] mmap: Fix do_brk_flags() modifying obviously incorrect VMAs

[Liam Howlett]

* Jann Horn <jannh@google.com> [221205 12:16]:
> On Mon, Dec 5, 2022 at 5:53 PM Liam Howlett <liam.howlett@oracle.com> wrote:
> > Add more sanity checks to the VMA that do_brk_flags() will expand.
> > Ensure the VMA matches basic merge requirements within the function
> > before calling can_vma_merge_after().
> >
> > Drop the duplicate checks from vm_brk_flags() since they will be
> > enforced later.
> 
> Looks good to me, with one note:
> 
> > Fixes: 2e7ce7d354f2 ("mm/mmap: change do_brk_flags() to expand existing VMA and add do_brk_munmap()")
> > Suggested-by: Jann Horn <jannh@google.com>
> > Signed-off-by: Liam R. Howlett <Liam.Howlett@oracle.com>
> > ---
> >  mm/mmap.c | 11 +++--------
> >  1 file changed, 3 insertions(+), 8 deletions(-)
> >
> > diff --git a/mm/mmap.c b/mm/mmap.c
> > index a5eb2f175da0..41a2c42593e8 100644
> > --- a/mm/mmap.c
> > +++ b/mm/mmap.c
> > @@ -2946,9 +2946,9 @@ static int do_brk_flags(struct ma_state *mas, struct vm_area_struct *vma,
> >          * Expand the existing vma if possible; Note that singular lists do not
> >          * occur after forking, so the expand will only happen on new VMAs.
> >          */
> > -       if (vma &&
> > -           (!vma->anon_vma || list_is_singular(&vma->anon_vma_chain)) &&
> > -           ((vma->vm_flags & ~VM_SOFTDIRTY) == flags)) {
> > +       if (vma && vma->vm_end == addr && !vma_policy(vma) && vma->anon_vma &&
> 
> Why the "vma->anon_vma" check here? The old code was checking that the
> existing VMA is not attached to more than one anon_vma; but the new
> code instead checks that the existing VMA is attached to at least one
> anon_vma, and then is_mergeable_anon_vma() checks that the VMA is not
> attached to more than one anon_vma, so in effect the VMA has to be
> attached to exactly one anon_vma. Is that intentional?

That was not intentional.

> 
> If not, maybe delete the "vma->anon_vma &&" - can_vma_merge_after()
> already does the equivalent check of the old "(!vma->anon_vma ||
> list_is_singular(&vma->anon_vma_chain))".

Yes, I will do that and send a v2.

> 
> > +           can_vma_merge_after(vma, flags, NULL, NULL,
> > +                               addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) {
> >                 mas_set_range(mas, vma->vm_start, addr + len - 1);
> >                 if (mas_preallocate(mas, vma, GFP_KERNEL))
> >                         return -ENOMEM;
> > @@ -3035,11 +3035,6 @@ int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags)
> >                 goto munmap_failed;
> >
> >         vma = mas_prev(&mas, 0);
> > -       if (!vma || vma->vm_end != addr || vma_policy(vma) ||
> > -           !can_vma_merge_after(vma, flags, NULL, NULL,
> > -                                addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL))
> > -               vma = NULL;
> > -
> >         ret = do_brk_flags(&mas, vma, addr, len, flags);
> >         populate = ((mm->def_flags & VM_LOCKED) != 0);
> >         mmap_write_unlock(mm);
> > --
> > 2.35.1