Re: [PATCH] mm/vmalloc: fix data race in show_numa_info()

[Eric Dumazet <edumazet@google.com>]

On Mon, May 5, 2025 at 10:20 AM Jeongjun Park <aha310510@gmail.com> wrote:
>
> The following data-race was found in show_numa_info():
>
> ==================================================================
> BUG: KCSAN: data-race in vmalloc_info_show / vmalloc_info_show
>
> r
> value changed: 0x0000008f -> 0x00000000
>
> Reported by Kernel Concurrency Sanitizer on:
> CPU: 1 UID: 0 PID: 8287 Comm: syz.0.411 Not tainted 6.15.0-rc4-00256-g95d3481af6dc-dirty #1 PREEMPT(voluntary)
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1 04/01/2014
> ==================================================================
>
> There is a read/write data-race in counter[]. This seems to be happening
> because only read memory barriers are currently applied, so we need to
> modify the write operation to counters[] to be handled atomically.
>
> Fixes: a47a126ad5ea ("vmallocinfo: add NUMA information")
> Signed-off-by: Jeongjun Park <aha310510@gmail.com>
> ---
>  mm/vmalloc.c | 9 +++++----
>  1 file changed, 5 insertions(+), 4 deletions(-)
>
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index 3ed720a787ec..d93fa535bc21 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -4917,7 +4917,8 @@ bool vmalloc_dump_obj(void *object)
>  static void show_numa_info(struct seq_file *m, struct vm_struct *v)
>  {
>         if (IS_ENABLED(CONFIG_NUMA)) {
> -               unsigned int nr, *counters = m->private;
> +               atomic_t *counters = m->private;
> +               unsigned int nr;
>                 unsigned int step = 1U << vm_area_page_order(v);
>
>                 if (!counters)
> @@ -4931,10 +4932,10 @@ static void show_numa_info(struct seq_file *m, struct vm_struct *v)
>                 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
>
>                 for (nr = 0; nr < v->nr_pages; nr += step)
> -                       counters[page_to_nid(v->pages[nr])] += step;
> +                       atomic_add(step, &counters[page_to_nid(v->pages[nr])]);
>                 for_each_node_state(nr, N_HIGH_MEMORY)
> -                       if (counters[nr])
> -                               seq_printf(m, " N%u=%u", nr, counters[nr]);
> +                       if (atomic_read(&counters[nr]))
> +                               seq_printf(m, " N%u=%u", nr, atomic_read(&counters[nr]));
>         }
>  }
>
> --

This patch looks bogus to me.

The race is about using m->private for storage, while the same file
can be read from multiple threads.

Using atomic_t is going to silence syzbot, but the bug is still there.