diff mbox series

vhost: fix host notifier configuration error flow

Message ID 1591890450-63055-1-git-send-email-matan@mellanox.com (mailing list archive)
State Rejected, archived
Delegated to: Maxime Coquelin
Headers show
Series vhost: fix host notifier configuration error flow | expand

Checks

Context Check Description
ci/Intel-compilation success Compilation OK
ci/iol-testing success Testing PASS
ci/iol-mellanox-Performance success Performance Testing PASS
ci/iol-nxp-Performance success Performance Testing PASS
ci/iol-intel-Performance success Performance Testing PASS
ci/travis-robot success Travis build: passed
ci/iol-broadcom-Performance success Performance Testing PASS
ci/checkpatch success coding style OK

Commit Message

Matan Azrad June 11, 2020, 3:47 p.m. UTC
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@dpdk.org

Signed-off-by: Matan Azrad <matan@mellanox.com>
---
 lib/librte_vhost/vhost_user.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

Comments

Xia, Chenbo June 12, 2020, 7:25 a.m. UTC | #1
Hi Matan,

> -----Original Message-----
> From: dev <dev-bounces@dpdk.org> On Behalf Of Matan Azrad
> Sent: Thursday, June 11, 2020 11:48 PM
> To: Maxime Coquelin <maxime.coquelin@redhat.com>
> Cc: dev@dpdk.org; Wang, Xiao W <xiao.w.wang@intel.com>; Tiwei Bie
> <tiwei.bie@intel.com>; stable@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@dpdk.org
> 
> Signed-off-by: Matan Azrad <matan@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
Matan Azrad June 12, 2020, 11:05 a.m. UTC | #2
Hi Xia
Thanks for the fast review.
Please see inline.

From: Xia, Chenbo:
> Hi Matan,
> 
> > -----Original Message-----
> > From: dev <dev-bounces@dpdk.org> On Behalf Of Matan Azrad
> > Sent: Thursday, June 11, 2020 11:48 PM
> > To: Maxime Coquelin <maxime.coquelin@redhat.com>
> > Cc: dev@dpdk.org; Wang, Xiao W <xiao.w.wang@intel.com>; Tiwei Bie
> > <tiwei.bie@intel.com>; stable@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@dpdk.org
> >
> > Signed-off-by: Matan Azrad <matan@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.

Yes, looks like I missed here something.

> 
> 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.

I got stuck in this function while receiving massage back from QEMU(4.2.0).
Looks like QEMU printed "Failed to read from slave. ".
In this case looks like QEMU doesn't reply and therefore the dpdk thread was stuck.
Any idea?




> Thanks!
> Chenbo
> 
> >
> >  	ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
> >  	if (ret <= 0) {
> > --
> > 1.8.3.1
Xia, Chenbo June 14, 2020, 1:11 p.m. UTC | #3
Hi Matan,

> -----Original Message-----
> From: Matan Azrad <matan@mellanox.com>
> Sent: Friday, June 12, 2020 7:06 PM
> To: Xia, Chenbo <chenbo.xia@intel.com>; Maxime Coquelin
> <maxime.coquelin@redhat.com>
> Cc: dev@dpdk.org; Wang, Xiao W <xiao.w.wang@intel.com>; Tiwei Bie
> <tiwei.bie@intel.com>; stable@dpdk.org
> Subject: RE: [dpdk-dev] [PATCH] vhost: fix host notifier configuration error flow
> 
> 
> Hi Xia
> Thanks for the fast review.
> Please see inline.
> 
> From: Xia, Chenbo:
> > Hi Matan,
> >
> > > -----Original Message-----
> > > From: dev <dev-bounces@dpdk.org> On Behalf Of Matan Azrad
> > > Sent: Thursday, June 11, 2020 11:48 PM
> > > To: Maxime Coquelin <maxime.coquelin@redhat.com>
> > > Cc: dev@dpdk.org; Wang, Xiao W <xiao.w.wang@intel.com>; Tiwei Bie
> > > <tiwei.bie@intel.com>; stable@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@dpdk.org
> > >
> > > Signed-off-by: Matan Azrad <matan@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.
> 
> Yes, looks like I missed here something.
> 
> >
> > 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.
> 
> I got stuck in this function while receiving massage back from QEMU(4.2.0).
> Looks like QEMU printed "Failed to read from slave. ".
> In this case looks like QEMU doesn't reply and therefore the dpdk thread was
> stuck.
> Any idea?

I see in QEMU code that the error is caused by recv size not match, could you check on the slave channel send/recv?
I check in dpdk code that normally all the slave msg should have size larger than VHOST_USER_HDR_SIZE. There is one
possibility that dpdk close its socket for some reason so that recvmsg returns 0. But I am sure. It's better to check on
send/recv of slave chanel :)

BRs,
Chenbo

> 
> 
> 
> 
> > Thanks!
> > Chenbo
> >
> > >
> > >  	ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
> > >  	if (ret <= 0) {
> > > --
> > > 1.8.3.1
diff mbox series

Patch

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;
+		}
+	}
 
 	ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
 	if (ret <= 0) {