net/virtio-user: support kernel vhost
authorJianfeng Tan <jianfeng.tan@intel.com>
Fri, 13 Jan 2017 12:18:38 +0000 (12:18 +0000)
committerYuanhan Liu <yuanhan.liu@linux.intel.com>
Tue, 17 Jan 2017 08:24:56 +0000 (09:24 +0100)
This patch add support vhost kernel as the backend for virtio_user.
Three main hook functions are added:
  - vhost_kernel_setup() to open char device, each vq pair needs one
    vhostfd;
  - vhost_kernel_ioctl() to communicate control messages with vhost
    kernel module;
  - vhost_kernel_enable_queue_pair() to open tap device and set it
    as the backend of corresonding vhost fd (that is to say, vq pair).

Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
doc/guides/rel_notes/release_17_02.rst
drivers/net/virtio/Makefile
drivers/net/virtio/virtio_user/vhost.h
drivers/net/virtio/virtio_user/vhost_kernel.c [new file with mode: 0644]
drivers/net/virtio/virtio_user/vhost_kernel_tap.c [new file with mode: 0644]
drivers/net/virtio/virtio_user/vhost_kernel_tap.h [new file with mode: 0644]
drivers/net/virtio/virtio_user/virtio_user_dev.c
drivers/net/virtio/virtio_user/virtio_user_dev.h

index dd5b3f8..a230724 100644 (file)
@@ -57,6 +57,27 @@ New Features
   Six new APIs have been added to the ixgbe PMD for MACsec offload support.
   The declarations for the APIs can be found in ``rte_pmd_ixgbe.h``.
 
+* **virtio-user with vhost-kernel as another exceptional path.**
+
+  Previously, we upstreamed a virtual device, virtio-user with vhost-user
+  as the backend, as a way for IPC (Inter-Process Communication) and user
+  space container networking.
+
+  Virtio-user with vhost-kernel as the backend is a solution for exceptional
+  path, such as KNI, which exchanges packets with kernel networking stack.
+  This solution is very promising in:
+
+  * maintenance: vhost and vhost-net (kernel) is upstreamed and extensively
+    used kernel module.
+  * features: vhost-net is born to be a networking solution, which has
+    lots of networking related features, like multi-queue, TSO, multi-seg
+    mbuf, etc.
+  * performance: similar to KNI, this solution would use one or more
+    kthreads to send/receive packets from user space DPDK applications,
+    which has little impact on user space polling thread (except that
+    it might enter into kernel space to wake up those kthreads if
+    necessary).
+
 
 Resolved Issues
 ---------------
index 97972a6..8b5b5d6 100644 (file)
@@ -60,6 +60,8 @@ endif
 
 ifeq ($(CONFIG_RTE_VIRTIO_USER),y)
 SRCS-$(CONFIG_RTE_LIBRTE_VIRTIO_PMD) += virtio_user/vhost_user.c
+SRCS-$(CONFIG_RTE_LIBRTE_VIRTIO_PMD) += virtio_user/vhost_kernel.c
+SRCS-$(CONFIG_RTE_LIBRTE_VIRTIO_PMD) += virtio_user/vhost_kernel_tap.c
 SRCS-$(CONFIG_RTE_LIBRTE_VIRTIO_PMD) += virtio_user/virtio_user_dev.c
 SRCS-$(CONFIG_RTE_LIBRTE_VIRTIO_PMD) += virtio_user_ethdev.c
 endif
index 515e4fc..5c983bd 100644 (file)
@@ -118,4 +118,6 @@ struct virtio_user_backend_ops {
 };
 
 struct virtio_user_backend_ops ops_user;
+struct virtio_user_backend_ops ops_kernel;
+
 #endif
diff --git a/drivers/net/virtio/virtio_user/vhost_kernel.c b/drivers/net/virtio/virtio_user/vhost_kernel.c
new file mode 100644 (file)
index 0000000..d3cbdbd
--- /dev/null
@@ -0,0 +1,311 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2016 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <unistd.h>
+
+#include <rte_memory.h>
+#include <rte_eal_memconfig.h>
+
+#include "vhost.h"
+#include "virtio_user_dev.h"
+#include "vhost_kernel_tap.h"
+
+struct vhost_memory_kernel {
+       uint32_t nregions;
+       uint32_t padding;
+       struct vhost_memory_region regions[0];
+};
+
+/* vhost kernel ioctls */
+#define VHOST_VIRTIO 0xAF
+#define VHOST_GET_FEATURES _IOR(VHOST_VIRTIO, 0x00, __u64)
+#define VHOST_SET_FEATURES _IOW(VHOST_VIRTIO, 0x00, __u64)
+#define VHOST_SET_OWNER _IO(VHOST_VIRTIO, 0x01)
+#define VHOST_RESET_OWNER _IO(VHOST_VIRTIO, 0x02)
+#define VHOST_SET_MEM_TABLE _IOW(VHOST_VIRTIO, 0x03, struct vhost_memory_kernel)
+#define VHOST_SET_LOG_BASE _IOW(VHOST_VIRTIO, 0x04, __u64)
+#define VHOST_SET_LOG_FD _IOW(VHOST_VIRTIO, 0x07, int)
+#define VHOST_SET_VRING_NUM _IOW(VHOST_VIRTIO, 0x10, struct vhost_vring_state)
+#define VHOST_SET_VRING_ADDR _IOW(VHOST_VIRTIO, 0x11, struct vhost_vring_addr)
+#define VHOST_SET_VRING_BASE _IOW(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
+#define VHOST_GET_VRING_BASE _IOWR(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
+#define VHOST_SET_VRING_KICK _IOW(VHOST_VIRTIO, 0x20, struct vhost_vring_file)
+#define VHOST_SET_VRING_CALL _IOW(VHOST_VIRTIO, 0x21, struct vhost_vring_file)
+#define VHOST_SET_VRING_ERR _IOW(VHOST_VIRTIO, 0x22, struct vhost_vring_file)
+#define VHOST_NET_SET_BACKEND _IOW(VHOST_VIRTIO, 0x30, struct vhost_vring_file)
+
+static uint64_t max_regions = 64;
+
+static void
+get_vhost_kernel_max_regions(void)
+{
+       int fd;
+       char buf[20] = {'\0'};
+
+       fd = open("/sys/module/vhost/parameters/max_mem_regions", O_RDONLY);
+       if (fd < 0)
+               return;
+
+       if (read(fd, buf, sizeof(buf) - 1) > 0)
+               max_regions = strtoull(buf, NULL, 10);
+
+       close(fd);
+}
+
+static uint64_t vhost_req_user_to_kernel[] = {
+       [VHOST_USER_SET_OWNER] = VHOST_SET_OWNER,
+       [VHOST_USER_RESET_OWNER] = VHOST_RESET_OWNER,
+       [VHOST_USER_SET_FEATURES] = VHOST_SET_FEATURES,
+       [VHOST_USER_GET_FEATURES] = VHOST_GET_FEATURES,
+       [VHOST_USER_SET_VRING_CALL] = VHOST_SET_VRING_CALL,
+       [VHOST_USER_SET_VRING_NUM] = VHOST_SET_VRING_NUM,
+       [VHOST_USER_SET_VRING_BASE] = VHOST_SET_VRING_BASE,
+       [VHOST_USER_GET_VRING_BASE] = VHOST_GET_VRING_BASE,
+       [VHOST_USER_SET_VRING_ADDR] = VHOST_SET_VRING_ADDR,
+       [VHOST_USER_SET_VRING_KICK] = VHOST_SET_VRING_KICK,
+       [VHOST_USER_SET_MEM_TABLE] = VHOST_SET_MEM_TABLE,
+};
+
+/* By default, vhost kernel module allows 64 regions, but DPDK allows
+ * 256 segments. As a relief, below function merges those virtually
+ * adjacent memsegs into one region.
+ */
+static struct vhost_memory_kernel *
+prepare_vhost_memory_kernel(void)
+{
+       uint32_t i, j, k = 0;
+       struct rte_memseg *seg;
+       struct vhost_memory_region *mr;
+       struct vhost_memory_kernel *vm;
+
+       vm = malloc(sizeof(struct vhost_memory_kernel) +
+                   max_regions *
+                   sizeof(struct vhost_memory_region));
+
+       for (i = 0; i < RTE_MAX_MEMSEG; ++i) {
+               seg = &rte_eal_get_configuration()->mem_config->memseg[i];
+               if (!seg->addr)
+                       break;
+
+               int new_region = 1;
+
+               for (j = 0; j < k; ++j) {
+                       mr = &vm->regions[j];
+
+                       if (mr->userspace_addr + mr->memory_size ==
+                           (uint64_t)(uintptr_t)seg->addr) {
+                               mr->memory_size += seg->len;
+                               new_region = 0;
+                               break;
+                       }
+
+                       if ((uint64_t)(uintptr_t)seg->addr + seg->len ==
+                           mr->userspace_addr) {
+                               mr->guest_phys_addr =
+                                       (uint64_t)(uintptr_t)seg->addr;
+                               mr->userspace_addr =
+                                       (uint64_t)(uintptr_t)seg->addr;
+                               mr->memory_size += seg->len;
+                               new_region = 0;
+                               break;
+                       }
+               }
+
+               if (new_region == 0)
+                       continue;
+
+               mr = &vm->regions[k++];
+               /* use vaddr here! */
+               mr->guest_phys_addr = (uint64_t)(uintptr_t)seg->addr;
+               mr->userspace_addr = (uint64_t)(uintptr_t)seg->addr;
+               mr->memory_size = seg->len;
+               mr->mmap_offset = 0;
+
+               if (k >= max_regions) {
+                       free(vm);
+                       return NULL;
+               }
+       }
+
+       vm->nregions = k;
+       vm->padding = 0;
+       return vm;
+}
+
+static int
+vhost_kernel_ioctl(struct virtio_user_dev *dev,
+                  enum vhost_user_request req,
+                  void *arg)
+{
+       int ret = -1;
+       unsigned int i;
+       uint64_t req_kernel;
+       struct vhost_memory_kernel *vm = NULL;
+
+       PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
+
+       req_kernel = vhost_req_user_to_kernel[req];
+
+       if (req_kernel == VHOST_SET_MEM_TABLE) {
+               vm = prepare_vhost_memory_kernel();
+               if (!vm)
+                       return -1;
+               arg = (void *)vm;
+       }
+
+       /* We don't need memory protection here */
+       if (req_kernel == VHOST_SET_FEATURES)
+               *(uint64_t *)arg &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
+
+       for (i = 0; i < dev->max_queue_pairs; ++i) {
+               if (dev->vhostfds[i] < 0)
+                       continue;
+
+               ret = ioctl(dev->vhostfds[i], req_kernel, arg);
+               if (ret < 0)
+                       break;
+       }
+
+       if (vm)
+               free(vm);
+
+       if (ret < 0)
+               PMD_DRV_LOG(ERR, "%s failed: %s",
+                           vhost_msg_strings[req], strerror(errno));
+
+       return ret;
+}
+
+/**
+ * Set up environment to talk with a vhost kernel backend.
+ *
+ * @return
+ *   - (-1) if fail to set up;
+ *   - (>=0) if successful.
+ */
+static int
+vhost_kernel_setup(struct virtio_user_dev *dev)
+{
+       int vhostfd;
+       uint32_t i;
+
+       get_vhost_kernel_max_regions();
+
+       for (i = 0; i < dev->max_queue_pairs; ++i) {
+               vhostfd = open(dev->path, O_RDWR);
+               if (vhostfd < 0) {
+                       PMD_DRV_LOG(ERR, "fail to open %s, %s",
+                                   dev->path, strerror(errno));
+                       return -1;
+               }
+
+               dev->vhostfds[i] = vhostfd;
+       }
+
+       return 0;
+}
+
+static int
+vhost_kernel_set_backend(int vhostfd, int tapfd)
+{
+       struct vhost_vring_file f;
+
+       f.fd = tapfd;
+       f.index = 0;
+       if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
+               PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
+                               strerror(errno));
+               return -1;
+       }
+
+       f.index = 1;
+       if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
+               PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
+                               strerror(errno));
+               return -1;
+       }
+
+       return 0;
+}
+
+static int
+vhost_kernel_enable_queue_pair(struct virtio_user_dev *dev,
+                              uint16_t pair_idx,
+                              int enable)
+{
+       int hdr_size;
+       int vhostfd;
+       int tapfd;
+
+       vhostfd = dev->vhostfds[pair_idx];
+
+       if (!enable) {
+               if (dev->tapfds[pair_idx]) {
+                       close(dev->tapfds[pair_idx]);
+                       dev->tapfds[pair_idx] = -1;
+               }
+               return vhost_kernel_set_backend(vhostfd, -1);
+       } else if (dev->tapfds[pair_idx] >= 0) {
+               return 0;
+       }
+
+       if ((dev->features & (1ULL << VIRTIO_NET_F_MRG_RXBUF)) ||
+           (dev->features & (1ULL << VIRTIO_F_VERSION_1)))
+               hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
+       else
+               hdr_size = sizeof(struct virtio_net_hdr);
+
+       tapfd = vhost_kernel_open_tap(&dev->ifname, hdr_size);
+       if (tapfd < 0) {
+               PMD_DRV_LOG(ERR, "fail to open tap for vhost kernel");
+               return -1;
+       }
+
+       if (vhost_kernel_set_backend(vhostfd, tapfd) < 0) {
+               PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
+               close(tapfd);
+               return -1;
+       }
+
+       dev->tapfds[pair_idx] = tapfd;
+       return 0;
+}
+
+struct virtio_user_backend_ops ops_kernel = {
+       .setup = vhost_kernel_setup,
+       .send_request = vhost_kernel_ioctl,
+       .enable_qp = vhost_kernel_enable_queue_pair
+};
diff --git a/drivers/net/virtio/virtio_user/vhost_kernel_tap.c b/drivers/net/virtio/virtio_user/vhost_kernel_tap.c
new file mode 100644 (file)
index 0000000..797713b
--- /dev/null
@@ -0,0 +1,116 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2016 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <unistd.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <net/if.h>
+#include <errno.h>
+#include <string.h>
+#include <limits.h>
+
+#include "vhost_kernel_tap.h"
+#include "../virtio_logs.h"
+
+int
+vhost_kernel_open_tap(char **p_ifname, int hdr_size)
+{
+       unsigned int tap_features;
+       int sndbuf = INT_MAX;
+       struct ifreq ifr;
+       int tapfd;
+
+       /* TODO:
+        * 1. verify we can get/set vnet_hdr_len, tap_probe_vnet_hdr_len
+        * 2. get number of memory regions from vhost module parameter
+        * max_mem_regions, supported in newer version linux kernel
+        */
+       tapfd = open(PATH_NET_TUN, O_RDWR);
+       if (tapfd < 0) {
+               PMD_DRV_LOG(ERR, "fail to open %s: %s",
+                           PATH_NET_TUN, strerror(errno));
+               return -1;
+       }
+
+       /* Construct ifr */
+       memset(&ifr, 0, sizeof(ifr));
+       ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
+
+       if (ioctl(tapfd, TUNGETFEATURES, &tap_features) == -1) {
+               PMD_DRV_LOG(ERR, "TUNGETFEATURES failed: %s", strerror(errno));
+               goto error;
+       }
+       if (tap_features & IFF_ONE_QUEUE)
+               ifr.ifr_flags |= IFF_ONE_QUEUE;
+
+       /* Let tap instead of vhost-net handle vnet header, as the latter does
+        * not support offloading. And in this case, we should not set feature
+        * bit VHOST_NET_F_VIRTIO_NET_HDR.
+        */
+       if (tap_features & IFF_VNET_HDR) {
+               ifr.ifr_flags |= IFF_VNET_HDR;
+       } else {
+               PMD_DRV_LOG(ERR, "TAP does not support IFF_VNET_HDR");
+               goto error;
+       }
+
+       if (*p_ifname)
+               strncpy(ifr.ifr_name, *p_ifname, IFNAMSIZ);
+       else
+               strncpy(ifr.ifr_name, "tap%d", IFNAMSIZ);
+       if (ioctl(tapfd, TUNSETIFF, (void *)&ifr) == -1) {
+               PMD_DRV_LOG(ERR, "TUNSETIFF failed: %s", strerror(errno));
+               goto error;
+       }
+
+       fcntl(tapfd, F_SETFL, O_NONBLOCK);
+
+       if (ioctl(tapfd, TUNSETVNETHDRSZ, &hdr_size) < 0) {
+               PMD_DRV_LOG(ERR, "TUNSETVNETHDRSZ failed: %s", strerror(errno));
+               goto error;
+       }
+
+       if (ioctl(tapfd, TUNSETSNDBUF, &sndbuf) < 0) {
+               PMD_DRV_LOG(ERR, "TUNSETSNDBUF failed: %s", strerror(errno));
+               goto error;
+       }
+
+       if (!(*p_ifname))
+               *p_ifname = strdup(ifr.ifr_name);
+
+       return tapfd;
+error:
+       close(tapfd);
+       return -1;
+}
diff --git a/drivers/net/virtio/virtio_user/vhost_kernel_tap.h b/drivers/net/virtio/virtio_user/vhost_kernel_tap.h
new file mode 100644 (file)
index 0000000..1fc58ff
--- /dev/null
@@ -0,0 +1,60 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2016 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <sys/ioctl.h>
+
+/* TUN ioctls */
+#define TUNSETIFF     _IOW('T', 202, int)
+#define TUNGETFEATURES _IOR('T', 207, unsigned int)
+#define TUNSETOFFLOAD  _IOW('T', 208, unsigned int)
+#define TUNGETIFF      _IOR('T', 210, unsigned int)
+#define TUNSETSNDBUF   _IOW('T', 212, int)
+#define TUNGETVNETHDRSZ _IOR('T', 215, int)
+#define TUNSETVNETHDRSZ _IOW('T', 216, int)
+#define TUNSETQUEUE  _IOW('T', 217, int)
+#define TUNSETVNETLE _IOW('T', 220, int)
+#define TUNSETVNETBE _IOW('T', 222, int)
+
+/* TUNSETIFF ifr flags */
+#define IFF_TAP          0x0002
+#define IFF_NO_PI        0x1000
+#define IFF_ONE_QUEUE    0x2000
+#define IFF_VNET_HDR     0x4000
+#define IFF_MULTI_QUEUE  0x0100
+#define IFF_ATTACH_QUEUE 0x0200
+#define IFF_DETACH_QUEUE 0x0400
+
+/* Constants */
+#define PATH_NET_TUN   "/dev/net/tun"
+
+int vhost_kernel_open_tap(char **p_ifname, int hdr_size);
index 32039a1..ae0824c 100644 (file)
@@ -192,6 +192,9 @@ int virtio_user_stop_device(struct virtio_user_dev *dev)
        for (i = 0; i < dev->max_queue_pairs; ++i)
                dev->ops->enable_qp(dev, i, 0);
 
+       free(dev->ifname);
+       dev->ifname = NULL;
+
        return 0;
 }
 
@@ -230,7 +233,7 @@ is_vhost_user_by_type(const char *path)
 static int
 virtio_user_dev_setup(struct virtio_user_dev *dev)
 {
-       uint32_t i;
+       uint32_t i, q;
 
        dev->vhostfd = -1;
        for (i = 0; i < VIRTIO_MAX_VIRTQUEUES * 2 + 1; ++i) {
@@ -238,12 +241,28 @@ virtio_user_dev_setup(struct virtio_user_dev *dev)
                dev->callfds[i] = -1;
        }
 
+       dev->vhostfds = NULL;
+       dev->tapfds = NULL;
+
        if (is_vhost_user_by_type(dev->path)) {
                dev->ops = &ops_user;
-               return dev->ops->setup(dev);
+       } else {
+               dev->ops = &ops_kernel;
+
+               dev->vhostfds = malloc(dev->max_queue_pairs * sizeof(int));
+               dev->tapfds = malloc(dev->max_queue_pairs * sizeof(int));
+               if (!dev->vhostfds || !dev->tapfds) {
+                       PMD_INIT_LOG(ERR, "Failed to malloc");
+                       return -1;
+               }
+
+               for (q = 0; q < dev->max_queue_pairs; ++q) {
+                       dev->vhostfds[q] = -1;
+                       dev->tapfds[q] = -1;
+               }
        }
 
-       return -1;
+       return dev->ops->setup(dev);
 }
 
 int
@@ -295,7 +314,18 @@ virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
 void
 virtio_user_dev_uninit(struct virtio_user_dev *dev)
 {
+       uint32_t i;
+
+       virtio_user_stop_device(dev);
+
        close(dev->vhostfd);
+
+       if (dev->vhostfds) {
+               for (i = 0; i < dev->max_queue_pairs; ++i)
+                       close(dev->vhostfds[i]);
+               free(dev->vhostfds);
+               free(dev->tapfds);
+       }
 }
 
 static uint8_t
index 9f2f82e..0d39f40 100644 (file)
@@ -43,6 +43,11 @@ struct virtio_user_dev {
        /* for vhost_user backend */
        int             vhostfd;
 
+       /* for vhost_kernel backend */
+       char            *ifname;
+       int             *vhostfds;
+       int             *tapfds;
+
        /* for both vhost_user and vhost_kernel */
        int             callfds[VIRTIO_MAX_VIRTQUEUES * 2 + 1];
        int             kickfds[VIRTIO_MAX_VIRTQUEUES * 2 + 1];