On 22.12.21 02:12, Peng Hao wrote:
> There is a check for vm->sbm.sb_states before, and it should check
> it here as well.
>
> Signed-off-by: Peng Hao <flyingpeng@xxxxxxxxxxx>
> ---
> drivers/virtio/virtio_mem.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/virtio/virtio_mem.c b/drivers/virtio/virtio_mem.c
> index 96e5a8782769..b6b7c489c8b6 100644
> --- a/drivers/virtio/virtio_mem.c
> +++ b/drivers/virtio/virtio_mem.c
> @@ -592,7 +592,7 @@ static int virtio_mem_sbm_sb_states_prepare_next_mb(struct virtio_mem *vm)
> return -ENOMEM;
>
> mutex_lock(&vm->hotplug_mutex);
> - if (new_bitmap)
> + if (vm->sbm.sb_states)
> memcpy(new_bitmap, vm->sbm.sb_states, old_pages * PAGE_SIZE);
>
> old_bitmap = vm->sbm.sb_states;
Right, on the first iteration (vm->sbm.sb_states == NULL) we would copy
from NULL.
I wonder why this never failed so far. I guess that's because "the
behavior is undefined" if a NULL pointer is passed.
I assume the memcpy implementation that we've been using so far simply
skips the operation if they detect a NULL pointer, although according to
the standard the behavior is undefined:
"
AFAIK, most implementations allow null pointers for no-op calls to
memcpy() but gcc issues a warning when it detects at compile time
that a null pointer is passed as the first or second argument to
memcpy().
" [1]
Fixes: 5f1f79bbc9e2 ("virtio-mem: Paravirtualized memory hotplug")
Cc: stable@xxxxxxxxxxxxxxx # v5.8+
Thanks!
[1]
https://mail-archives.apache.org/mod_mbox/stdcxx-dev/200804.mbox/%3CCFFDD219128FD94FB4F92B99F52D0A49010A36F4@xxxxxxxxxxxxxxxxxxxxxxxxxxxxx%3E
--
Thanks,
David / dhildenb
