From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga02.intel.com (mga02.intel.com [134.134.136.20]) by dpdk.org (Postfix) with ESMTP id 833591CD59 for ; Thu, 5 Apr 2018 20:13:25 +0200 (CEST) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from orsmga004.jf.intel.com ([10.7.209.38]) by orsmga101.jf.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 05 Apr 2018 11:13:24 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.48,411,1517904000"; d="scan'208";a="188970461" Received: from tanjianf-mobl.ccr.corp.intel.com (HELO [10.255.27.107]) ([10.255.27.107]) by orsmga004.jf.intel.com with ESMTP; 05 Apr 2018 11:13:22 -0700 To: zhiyong.yang@intel.com, dev@dpdk.org References: <20180404171753.43422-1-zhiyong.yang@intel.com> <20180406001855.54062-1-zhiyong.yang@intel.com> Cc: maxime.coquelin@redhat.com, tiwei.bie@intel.com, dong1.wang@intel.com, zhihong.wang@intel.com From: "Tan, Jianfeng" Message-ID: <61fda2c8-588d-7afe-bee5-44d73c04a470@intel.com> Date: Fri, 6 Apr 2018 02:13:21 +0800 User-Agent: Mozilla/5.0 (Windows NT 6.3; WOW64; rv:45.0) Gecko/20100101 Thunderbird/45.8.0 MIME-Version: 1.0 In-Reply-To: <20180406001855.54062-1-zhiyong.yang@intel.com> Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 7bit Subject: Re: [dpdk-dev] [PATCH v6] net/virtio-user: add support for server mode X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 05 Apr 2018 18:13:27 -0000 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 > --- > > 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. > + > /* 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. > 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. > + dev->device_features = VIRTIO_USER_SUPPORTED_FEATURES; > } > + > if (dev->mac_specified) > dev->device_features |= (1ull << VIRTIO_NET_F_MAC); > > @@ -388,6 +402,11 @@ virtio_user_dev_uninit(struct virtio_user_dev *dev) > > close(dev->vhostfd); > > + if (dev->is_server && dev->listenfd >= 0) { > + close(dev->listenfd); > + dev->listenfd = -1; > + } > + > if (dev->vhostfds) { > for (i = 0; i < dev->max_queue_pairs; ++i) > close(dev->vhostfds[i]); > @@ -396,6 +415,9 @@ virtio_user_dev_uninit(struct virtio_user_dev *dev) > } > > free(dev->ifname); > + > + if (dev->is_server) > + unlink(dev->path); > } > > static uint8_t > diff --git a/drivers/net/virtio/virtio_user/virtio_user_dev.h b/drivers/net/virtio/virtio_user/virtio_user_dev.h > index 5f8755771..ade727e46 100644 > --- a/drivers/net/virtio/virtio_user/virtio_user_dev.h > +++ b/drivers/net/virtio/virtio_user/virtio_user_dev.h > @@ -6,6 +6,7 @@ > #define _VIRTIO_USER_DEV_H > > #include > +#include > #include "../virtio_pci.h" > #include "../virtio_ring.h" > #include "vhost.h" > @@ -13,6 +14,8 @@ > struct virtio_user_dev { > /* for vhost_user backend */ > int vhostfd; > + int listenfd; /* listening fd */ > + bool is_server; /* server or client mode */ > > /* for vhost_kernel backend */ > char *ifname; > diff --git a/drivers/net/virtio/virtio_user_ethdev.c b/drivers/net/virtio/virtio_user_ethdev.c > index 263649006..4e7b3c34f 100644 > --- a/drivers/net/virtio/virtio_user_ethdev.c > +++ b/drivers/net/virtio/virtio_user_ethdev.c > @@ -24,15 +24,73 @@ > #define virtio_user_get_dev(hw) \ > ((struct virtio_user_dev *)(hw)->virtio_user_dev) > > +static int > +virtio_user_server_reconnect(struct virtio_user_dev *dev) > +{ > + int ret; > + int flag; > + int connectfd; > + struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id]; > + > + connectfd = accept(dev->listenfd, NULL, NULL); > + if (connectfd < 0) > + return -1; > + > + dev->vhostfd = connectfd; > + flag = fcntl(connectfd, F_GETFD); > + fcntl(connectfd, F_SETFL, flag | O_NONBLOCK); > + > + ret = virtio_user_start_device(dev); > + if (ret < 0) > + return -1; > + > + if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) { > + if (rte_intr_disable(eth_dev->intr_handle) < 0) { > + PMD_DRV_LOG(ERR, "interrupt disable failed"); > + return -1; > + } > + rte_intr_callback_unregister(eth_dev->intr_handle, > + virtio_interrupt_handler, > + eth_dev); > + eth_dev->intr_handle->fd = connectfd; > + rte_intr_callback_register(eth_dev->intr_handle, > + virtio_interrupt_handler, eth_dev); > + > + if (rte_intr_enable(eth_dev->intr_handle) < 0) { > + PMD_DRV_LOG(ERR, "interrupt enable failed"); > + return -1; > + } > + } > + PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!"); > + return 0; > +} > + > static void > virtio_user_delayed_handler(void *param) > { > struct virtio_hw *hw = (struct virtio_hw *)param; > - struct rte_eth_dev *dev = &rte_eth_devices[hw->port_id]; > + struct rte_eth_dev *eth_dev = &rte_eth_devices[hw->port_id]; > + struct virtio_user_dev *dev = virtio_user_get_dev(hw); > > - rte_intr_callback_unregister(dev->intr_handle, > - virtio_interrupt_handler, > - dev); > + if (rte_intr_disable(eth_dev->intr_handle) < 0) { > + PMD_DRV_LOG(ERR, "interrupt disable failed"); > + return; > + } > + rte_intr_callback_unregister(eth_dev->intr_handle, > + virtio_interrupt_handler, eth_dev); > + if (dev->is_server) { > + if (dev->vhostfd >= 0) { > + close(dev->vhostfd); > + dev->vhostfd = -1; > + } > + eth_dev->intr_handle->fd = dev->listenfd; > + rte_intr_callback_register(eth_dev->intr_handle, > + virtio_interrupt_handler, eth_dev); > + if (rte_intr_enable(eth_dev->intr_handle) < 0) { > + PMD_DRV_LOG(ERR, "interrupt enable failed"); > + return; > + } > + } > } > > static void > @@ -67,12 +125,10 @@ virtio_user_read_dev_config(struct virtio_hw *hw, size_t offset, > dev->status &= (~VIRTIO_NET_S_LINK_UP); > PMD_DRV_LOG(ERR, "virtio-user port %u is down", > hw->port_id); > - /* Only client mode is available now. Once the > - * connection is broken, it can never be up > - * again. Besides, this function could be called > - * in the process of interrupt handling, > - * callback cannot be unregistered here, set an > - * alarm to do it. > + > + /* This function could be called in the process > + * of interrupt handling, callback cannot be > + * unregistered here, set an alarm to do it. > */ > rte_eal_alarm_set(1, > virtio_user_delayed_handler, > @@ -85,7 +141,12 @@ virtio_user_read_dev_config(struct virtio_hw *hw, size_t offset, > PMD_DRV_LOG(ERR, "error clearing O_NONBLOCK flag"); > return; > } > + } else if (dev->is_server) { > + dev->status &= (~VIRTIO_NET_S_LINK_UP); > + if (virtio_user_server_reconnect(dev) >= 0) > + dev->status |= VIRTIO_NET_S_LINK_UP; > } > + > *(uint16_t *)dst = dev->status; > } > > @@ -278,12 +339,15 @@ static const char *valid_args[] = { > VIRTIO_USER_ARG_QUEUE_SIZE, > #define VIRTIO_USER_ARG_INTERFACE_NAME "iface" > VIRTIO_USER_ARG_INTERFACE_NAME, > +#define VIRTIO_USER_ARG_SERVER_MODE "server" > + VIRTIO_USER_ARG_SERVER_MODE, > NULL > }; > > #define VIRTIO_USER_DEF_CQ_EN 0 > #define VIRTIO_USER_DEF_Q_NUM 1 > #define VIRTIO_USER_DEF_Q_SZ 256 > +#define VIRTIO_USER_DEF_SERVER_MODE 0 > > static int > get_string_arg(const char *key __rte_unused, > @@ -378,6 +442,7 @@ virtio_user_pmd_probe(struct rte_vdev_device *dev) > uint64_t queues = VIRTIO_USER_DEF_Q_NUM; > uint64_t cq = VIRTIO_USER_DEF_CQ_EN; > uint64_t queue_size = VIRTIO_USER_DEF_Q_SZ; > + uint64_t server_mode = VIRTIO_USER_DEF_SERVER_MODE; > char *path = NULL; > char *ifname = NULL; > char *mac_addr = NULL; > @@ -445,6 +510,15 @@ virtio_user_pmd_probe(struct rte_vdev_device *dev) > } > } > > + if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_SERVER_MODE) == 1) { > + if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_SERVER_MODE, > + &get_integer_arg, &server_mode) < 0) { > + PMD_INIT_LOG(ERR, "error to parse %s", > + VIRTIO_USER_ARG_SERVER_MODE); > + goto end; > + } > + } > + > if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_CQ_NUM) == 1) { > if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_CQ_NUM, > &get_integer_arg, &cq) < 0) { > @@ -469,6 +543,8 @@ virtio_user_pmd_probe(struct rte_vdev_device *dev) > } > > if (rte_eal_process_type() == RTE_PROC_PRIMARY) { > + struct virtio_user_dev *vu_dev; > + > eth_dev = virtio_user_eth_dev_alloc(dev); > if (!eth_dev) { > PMD_INIT_LOG(ERR, "virtio_user fails to alloc device"); > @@ -476,6 +552,11 @@ virtio_user_pmd_probe(struct rte_vdev_device *dev) > } > > hw = eth_dev->data->dev_private; > + vu_dev = virtio_user_get_dev(hw); > + if (server_mode == 1) > + vu_dev->is_server = true; > + else > + vu_dev->is_server = false; > if (virtio_user_dev_init(hw->virtio_user_dev, path, queues, cq, > queue_size, mac_addr, &ifname) < 0) { > PMD_INIT_LOG(ERR, "virtio_user_dev_init fails");