Re: mm: BUG in expand_downwards

From: Dmitry Vyukov
Date: Fri Jan 29 2016 - 05:39:52 EST


On Thu, Jan 28, 2016 at 4:47 PM, Andrea Arcangeli <aarcange@xxxxxxxxxx> wrote:
> Hello,
>
> On Wed, Jan 27, 2016 at 10:11:44PM +0100, Dmitry Vyukov wrote:
>> Sorry, I meant only the second once. The mm bug.
>> I guess you need at least CONFIG_DEBUG_VM. Run it in a tight parallel
>> loop with CPU oversubscription (e.g. 32 parallel processes on 2 cores)
>> for at least an hour.
>
> Does this help for the mm bug?

Yes, it seems to fix the issue.
I will also run fuzzer with this patch and report if I see it again.

> From 0cc410ae59800444ca929e3dc48e4f1580a95be6 Mon Sep 17 00:00:00 2001
> From: Andrea Arcangeli <aarcange@xxxxxxxxxx>
> Date: Thu, 28 Jan 2016 16:34:44 +0100
> Subject: [PATCH 1/1] mm: validate_mm browse_rb SMP race condition
>
> The mmap_sem for reading in validate_mm called from expand_stack is
> not enough to prevent the argumented rbtree rb_subtree_gap information
> to change from under us because expand_stack may be running from other
> threads concurrently which will hold the mmap_sem for reading too.
>
> The argumented rbtree is updated with vma_gap_update under the
> page_table_lock so use it in browse_rb() too to avoid false positives.
>
> Reported-by: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
> Signed-off-by: Andrea Arcangeli <aarcange@xxxxxxxxxx>
> ---
> mm/mmap.c | 7 +++++--
> 1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/mm/mmap.c b/mm/mmap.c
> index f384def..8389e03 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -389,8 +389,9 @@ static long vma_compute_subtree_gap(struct vm_area_struct *vma)
> }
>
> #ifdef CONFIG_DEBUG_VM_RB
> -static int browse_rb(struct rb_root *root)
> +static int browse_rb(struct mm_struct *mm)
> {
> + struct rb_root *root = &mm->mm_rb;
> int i = 0, j, bug = 0;
> struct rb_node *nd, *pn = NULL;
> unsigned long prev = 0, pend = 0;
> @@ -413,12 +414,14 @@ static int browse_rb(struct rb_root *root)
> vma->vm_start, vma->vm_end);
> bug = 1;
> }
> + spin_lock(&mm->page_table_lock);
> if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
> pr_emerg("free gap %lx, correct %lx\n",
> vma->rb_subtree_gap,
> vma_compute_subtree_gap(vma));
> bug = 1;
> }
> + spin_unlock(&mm->page_table_lock);
> i++;
> pn = nd;
> prev = vma->vm_start;
> @@ -474,7 +477,7 @@ static void validate_mm(struct mm_struct *mm)
> mm->highest_vm_end, highest_address);
> bug = 1;
> }
> - i = browse_rb(&mm->mm_rb);
> + i = browse_rb(mm);
> if (i != mm->map_count) {
> if (i != -1)
> pr_emerg("map_count %d rb %d\n", mm->map_count, i);