Re: [dpdk-dev] [PATCH v6] net/virtio-user: add support for server mode

["Yang, Zhiyong" <zhiyong.yang@intel.com>]

Hi Jianfeng,

> -----Original Message-----
> From: Tan, Jianfeng
> Sent: Friday, April 6, 2018 2:13 AM
> To: Yang, Zhiyong <zhiyong.yang@intel.com>; dev@dpdk.org
> Cc: maxime.coquelin@redhat.com; Bie, Tiwei <tiwei.bie@intel.com>; Wang,
> Dong1 <dong1.wang@intel.com>; Wang, Zhihong <zhihong.wang@intel.com>
> Subject: Re: [PATCH v6] net/virtio-user: add support for server mode
> 
> 
> 
> On 4/6/2018 8:18 AM, zhiyong.yang@intel.com wrote:
> > In a container environment if the vhost-user backend restarts, there's
> > no way for it to reconnect to virtio-user. To address this, support
> > for server mode is added. In this mode the socket file is created by
> > virtio- user, which the backend then connects to. This means that if
> > the backend restarts, it can reconnect to virtio-user and continue
> communications.
> >
> > With current implementation, LSC is enabled at virtio-user side to
> > support to accept the coming connection.
> >
> > Release note is updated in this patch.
> >
> > Signed-off-by: Zhiyong Yang <zhiyong.yang@intel.com>
> > ---
> >
> > Changes in V6:
> > 1. fix report wrong link stauts in server mode.
> > 2. fix some code style issues.
> >
> > Changes in V5:
> > 1. Support server mode virtio-user startup in non-blocking mode.
> > 2. rebase on top of dpdk-next-virtio.
> >
> > Changes in V4:
> > 1. Don't create new pthread any more and use librte_eal interrupt thread.
> > 2. virtio-user doesn't work in blocking mode any more for the first
> connection.
> > Client mode vhost-user startups firstly, then server mode virtio-user
> > creates socket file and startups. Keep consistency with usage of
> > client mode virtio-user.
> >
> > Changes in V3:
> > 1. use EAL epoll mechanism instead of vhost events. Cancel to export
> > vhost event APIs.
> > 2. rebase the code on top of dpdk-next-virtio
> >
> > Changes in V2:
> > 1. split two patches 1/5 and 2/5 from v1 patchset to fix some existing
> > issues which is not strongly related to support for server mode 2.
> > move fdset related functions to librte_eal from librte_vhost exposed
> > as new APIs.
> > 3. release note is added in the patch 5/5.
> > 4. squash data structure change patch into 4/5 according to Maxime's
> suggestion.
> >
> >   doc/guides/rel_notes/release_18_05.rst           |   6 ++
> >   drivers/net/virtio/virtio_user/vhost_user.c      |  45 ++++++++--
> >   drivers/net/virtio/virtio_user/virtio_user_dev.c |  40 +++++++--
> >   drivers/net/virtio/virtio_user/virtio_user_dev.h |   3 +
> >   drivers/net/virtio/virtio_user_ethdev.c          | 101
> ++++++++++++++++++++---
> >   5 files changed, 171 insertions(+), 24 deletions(-)
> >
> > diff --git a/doc/guides/rel_notes/release_18_05.rst
> > b/doc/guides/rel_notes/release_18_05.rst
> > index 9cc77f893..f8897b2e9 100644
> > --- a/doc/guides/rel_notes/release_18_05.rst
> > +++ b/doc/guides/rel_notes/release_18_05.rst
> > @@ -58,6 +58,12 @@ New Features
> >     * Added support for NVGRE, VXLAN and GENEVE filters in flow API.
> >     * Added support for DROP action in flow API.
> >
> > +* **Added support for virtio-user server mode.**
> > +  In a container environment if the vhost-user backend restarts,
> > +there's no way
> > +  for it to reconnect to virtio-user. To address this, support for
> > +server mode
> > +  is added. In this mode the socket file is created by virtio-user,
> > +which the
> > +  backend connects to. This means that if the backend restarts, it
> > +can reconnect
> > +  to virtio-user and continue communications.
> >
> >   API Changes
> >   -----------
> > diff --git a/drivers/net/virtio/virtio_user/vhost_user.c
> > b/drivers/net/virtio/virtio_user/vhost_user.c
> > index 91c6449bb..a6df97a00 100644
> > --- a/drivers/net/virtio/virtio_user/vhost_user.c
> > +++ b/drivers/net/virtio/virtio_user/vhost_user.c
> > @@ -378,6 +378,30 @@ vhost_user_sock(struct virtio_user_dev *dev,
> >   	return 0;
> >   }
> >
> > +#define MAX_VIRTIO_USER_BACKLOG 1
> > +static int
> > +virtio_user_start_server(struct virtio_user_dev *dev, struct
> > +sockaddr_un *un) {
> > +	int ret;
> > +	int flag;
> > +	int fd = dev->listenfd;
> > +
> > +	ret = bind(fd, (struct sockaddr *)un, sizeof(*un));
> > +	if (ret < 0) {
> > +		PMD_DRV_LOG(ERR, "failed to bind to %s: %s; remove it and
> try again\n",
> > +			    dev->path, strerror(errno));
> > +		return -1;
> > +	}
> > +	ret = listen(fd, MAX_VIRTIO_USER_BACKLOG);
> > +	if (ret < 0)
> > +		return -1;
> > +
> > +	flag = fcntl(fd, F_GETFL);
> > +	fcntl(fd, F_SETFL, flag | O_NONBLOCK);
> > +
> > +	return 0;
> > +}
> > +
> >   /**
> >    * Set up environment to talk with a vhost user backend.
> >    *
> > @@ -405,13 +429,24 @@ vhost_user_setup(struct virtio_user_dev *dev)
> >   	memset(&un, 0, sizeof(un));
> >   	un.sun_family = AF_UNIX;
> >   	snprintf(un.sun_path, sizeof(un.sun_path), "%s", dev->path);
> > -	if (connect(fd, (struct sockaddr *)&un, sizeof(un)) < 0) {
> > -		PMD_DRV_LOG(ERR, "connect error, %s", strerror(errno));
> > -		close(fd);
> > -		return -1;
> > +
> > +	if (dev->is_server) {
> > +		dev->listenfd = fd;
> > +		if (virtio_user_start_server(dev, &un) < 0) {
> > +			PMD_DRV_LOG(ERR, "virtio-user startup fails in
> server mode");
> > +			close(fd);
> > +			return -1;
> > +		}
> > +		dev->vhostfd = -1;
> > +	} else {
> > +		if (connect(fd, (struct sockaddr *)&un, sizeof(un)) < 0) {
> > +			PMD_DRV_LOG(ERR, "connect error, %s",
> strerror(errno));
> > +			close(fd);
> > +			return -1;
> > +		}
> > +		dev->vhostfd = fd;
> >   	}
> >
> > -	dev->vhostfd = fd;
> >   	return 0;
> >   }
> >
> > diff --git a/drivers/net/virtio/virtio_user/virtio_user_dev.c
> > b/drivers/net/virtio/virtio_user/virtio_user_dev.c
> > index f90fee9e5..3b776282c 100644
> > --- a/drivers/net/virtio/virtio_user/virtio_user_dev.c
> > +++ b/drivers/net/virtio/virtio_user/virtio_user_dev.c
> > @@ -99,6 +99,9 @@ virtio_user_start_device(struct virtio_user_dev *dev)
> >   	uint64_t features;
> >   	int ret;
> >
> > +	if (dev->vhostfd < 0)
> > +		return -1;
> 
> This breaks virtio-user with vhost-kernel which does not need vhostfd, and
> will be always -1.
> 

Ok, I will modify the code as followings
	if (is_vhost_user_by_type(dev->path) && dev->vhostfd < 0)
		return -1; 

(is_vhost_user_by_type(dev->path))
> > +
> >   	/* Do not check return as already done in init, or reset in stop */
> >   	dev->ops->send_request(dev, VHOST_USER_SET_OWNER, NULL);
> >
> > @@ -254,6 +257,8 @@ virtio_user_fill_intr_handle(struct virtio_user_dev
> *dev)
> >   	eth_dev->intr_handle->fd = -1;
> >   	if (dev->vhostfd >= 0)
> >   		eth_dev->intr_handle->fd = dev->vhostfd;
> > +	else if (dev->is_server)
> > +		eth_dev->intr_handle->fd = dev->listenfd;
> >
> >   	return 0;
> >   }
> > @@ -267,7 +272,7 @@ virtio_user_dev_setup(struct virtio_user_dev *dev)
> >   	dev->vhostfds = NULL;
> >   	dev->tapfds = NULL;
> >
> > -	if (is_vhost_user_by_type(dev->path)) {
> > +	if (dev->is_server || is_vhost_user_by_type(dev->path)) {
> 
> I think we still fail to pick out an invalidated case: specify "server"
> parameter for a vhost-kernel path.

I will add the checking at the starting of the function.

If (dev->is_server && access(dev->path) == 0)
	Return -1;
How about the change ?

> 
> >   		dev->ops = &ops_user;
> >   	} else {
> >   		dev->ops = &ops_kernel;
> > @@ -337,16 +342,25 @@ virtio_user_dev_init(struct virtio_user_dev *dev,
> char *path, int queues,
> >   		return -1;
> >   	}
> >
> > -	if (dev->ops->send_request(dev, VHOST_USER_SET_OWNER, NULL)
> < 0) {
> > -		PMD_INIT_LOG(ERR, "set_owner fails: %s", strerror(errno));
> > -		return -1;
> > -	}
> > +	if (dev->vhostfd >= 0) {
> > +		if (dev->ops->send_request(dev,
> VHOST_USER_SET_OWNER,
> > +					   NULL) < 0) {
> > +			PMD_INIT_LOG(ERR, "set_owner fails: %s",
> > +				     strerror(errno));
> > +			return -1;
> > +		}
> >
> > -	if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES,
> > -			    &dev->device_features) < 0) {
> > -		PMD_INIT_LOG(ERR, "get_features failed: %s",
> strerror(errno));
> > -		return -1;
> > +		if (dev->ops->send_request(dev,
> VHOST_USER_GET_FEATURES,
> > +					   &dev->device_features) < 0) {
> > +			PMD_INIT_LOG(ERR, "get_features failed: %s",
> > +				     strerror(errno));
> > +			return -1;
> > +		}
> > +	} else {
> > +		/* Just pretend vhost-user can support all these features */
> 
> Although I don't think we need to fix this, please also note that this could be
> problematic that if some feature is negotiated but not supported by the
> vhost-user which comes later.
>

Good description. I will add it in next version.
 
Thanks for your those comments

Zhiyong