Yajun Wu <yaj...@nvidia.com> writes:
> vhost backend sends host notification for every VQ. If backend creates
> VQs in parallel, the VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG may
> arrive to QEMU in different order than incremental queue index order.
>
> For example VQ 1's message arrive earlier than VQ 0's:
> After alloc VhostUserHostNotifier for VQ 1. GPtrArray becomes
>
> [ nil, VQ1 pointer ]
>
> After alloc VhostUserHostNotifier for VQ 0. GPtrArray becomes
>
> [ VQ0 pointer, nil, VQ1 pointer ]
>
> This is wrong. fetch_notifier will return NULL for VQ 1 in
> vhost_user_get_vring_base, causes host notifier miss removal(leak).
>
> The fix is to remove current element from GPtrArray, make the right
> position for element to insert.
>
> Fixes: 503e355465 ("virtio/vhost-user: dynamically assign
> VhostUserHostNotifiers")
> Signed-off-by: Yajun Wu <yaj...@nvidia.com>
> Acked-by: Parav Pandit <pa...@nvidia.com>
>
> ---
> hw/virtio/vhost-user.c | 5 +++++
> 1 file changed, 5 insertions(+)
>
> diff --git a/hw/virtio/vhost-user.c b/hw/virtio/vhost-user.c
> index 03415b6c95..d256ce589b 100644
> --- a/hw/virtio/vhost-user.c
> +++ b/hw/virtio/vhost-user.c
> @@ -1543,6 +1543,11 @@ static VhostUserHostNotifier
> *fetch_or_create_notifier(VhostUserState *u,
>
> n = g_ptr_array_index(u->notifiers, idx);
> if (!n) {
> + /*
> + * In case notification arrive out-of-order,
> + * make room for current index.
> + */
> + g_ptr_array_remove_index(u->notifiers, idx);
> n = g_new0(VhostUserHostNotifier, 1);
> n->idx = idx;
> g_ptr_array_insert(u->notifiers, idx, n);
Ahh reading the description of g_ptr_array_insert() it doesn't
explicitly say it shuffles the data currently at index up but I can see
from the code it does. I guess you can imply it from the other function
descriptions.
Reviewed-by: Alex Bennée <alex.ben...@linaro.org>
--
Alex Bennée
