[dpdk-stable] [dpdk-dev] [PATCH] vhost: fix host notifier configuration error flow
Xia, Chenbo
chenbo.xia at intel.com
Fri Jun 12 09:25:57 CEST 2020
Hi Matan,
> -----Original Message-----
> From: dev <dev-bounces at dpdk.org> On Behalf Of Matan Azrad
> Sent: Thursday, June 11, 2020 11:48 PM
> To: Maxime Coquelin <maxime.coquelin at redhat.com>
> Cc: dev at dpdk.org; Wang, Xiao W <xiao.w.wang at intel.com>; Tiwei Bie
> <tiwei.bie at intel.com>; stable at dpdk.org
> Subject: [dpdk-dev] [PATCH] vhost: fix host notifier configuration error flow
>
> A vDPA driver can configure its device FD to be notified directly by the guest
> memory mapping using `rte_vhost_host_notifier_ctrl` API.
>
> The driver request is managed by the dpdk vhost management and is forwarded
> to the QEMU, the vhost massage includes reply request in order to be sure that
> the memory mapping was done correctly by the QEMU.
>
> When QEMU finishes the configuration, it marks that its replay is valid in the
> slave FD using VHOST_USER_REPLY_MASK flag.
> The flag is set only in success and when the slave FD includes the reply data.
>
> The vhost library didn't validate the above flag before accessing to the slave FD,
> it leaded to the thread to be blocked on recvmsg call forever in case the QEMU
> has some problems in the notifier configuration.
>
> Handle VHOST_USER_REPLY_MASK flag to validate that slave FD includes a reply
> data.
>
> Fixes: d90cf7d111ac ("vhost: support host notifier")
> Cc: stable at dpdk.org
>
> Signed-off-by: Matan Azrad <matan at mellanox.com>
> ---
> lib/librte_vhost/vhost_user.c | 10 ++++++++--
> 1 file changed, 8 insertions(+), 2 deletions(-)
>
> diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c index
> 84bebad..aa19d15 100644
> --- a/lib/librte_vhost/vhost_user.c
> +++ b/lib/librte_vhost/vhost_user.c
> @@ -2833,8 +2833,14 @@ static int process_slave_message_reply(struct
> virtio_net *dev,
> struct VhostUserMsg msg_reply;
> int ret;
>
> - if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
> - return 0;
> + if (!(msg->flags & VHOST_USER_REPLY_MASK)) {
> + if (msg->flags & VHOST_USER_NEED_REPLY) {
> + ret = -1;
> + goto out;
> + } else {
> + return 0;
> + }
> + }
Based on your commit log, I think you want to check the reply msg sent from qemu but msg is the request sent from vhost-user.
Also, could you clarify the problem based on that? Because I see in QEMU v5.0.0 that if vhost request has VHOST_USER_NEED_REPLY_MASK,
qemu will set the reply mask with VHOST_USER_REPLY_MASK and without VHOST_USER_NEED_REPLY_MASK no matter the handle
is correct. Do I miss something? Please correct me.
Thanks!
Chenbo
>
> ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
> if (ret <= 0) {
> --
> 1.8.3.1
More information about the stable
mailing list