#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
+#include <errno.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_kernel_data {
+ int *vhostfds;
+ int *tapfds;
+};
+
struct vhost_memory_kernel {
uint32_t nregions;
uint32_t padding;
#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)
+/* with below features, vhost kernel does not need to do the checksum and TSO,
+ * these info will be passed to virtio_user through virtio net header.
+ */
+#define VHOST_KERNEL_GUEST_OFFLOADS_MASK \
+ ((1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
+ (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
+ (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
+ (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
+ (1ULL << VIRTIO_NET_F_GUEST_UFO))
+
+/* with below features, when flows from virtio_user to vhost kernel
+ * (1) if flows goes up through the kernel networking stack, it does not need
+ * to verify checksum, which can save CPU cycles;
+ * (2) if flows goes through a Linux bridge and outside from an interface
+ * (kernel driver), checksum and TSO will be done by GSO in kernel or even
+ * offloaded into real physical device.
+ */
+#define VHOST_KERNEL_HOST_OFFLOADS_MASK \
+ ((1ULL << VIRTIO_NET_F_HOST_TSO4) | \
+ (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
+ (1ULL << VIRTIO_NET_F_CSUM))
+
static uint64_t max_regions = 64;
static void
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,
-};
+static int
+vhost_kernel_ioctl(int fd, uint64_t request, void *arg)
+{
+ int ret;
+
+ ret = ioctl(fd, request, arg);
+ if (ret) {
+ PMD_DRV_LOG(ERR, "Vhost-kernel ioctl %"PRIu64" failed (%s)",
+ request, strerror(errno));
+ return -1;
+ }
+
+ return 0;
+}
+
+static int
+vhost_kernel_set_owner(struct virtio_user_dev *dev)
+{
+ int ret;
+ uint32_t i;
+ struct vhost_kernel_data *data = dev->backend_data;
+
+ for (i = 0; i < dev->max_queue_pairs; ++i) {
+ if (data->vhostfds[i] < 0)
+ continue;
+
+ ret = vhost_kernel_ioctl(data->vhostfds[i], VHOST_SET_OWNER, NULL);
+ if (ret < 0)
+ return ret;
+ }
+
+ return 0;
+}
+
+static int
+vhost_kernel_get_features(struct virtio_user_dev *dev, uint64_t *features)
+{
+ struct vhost_kernel_data *data = dev->backend_data;
+ unsigned int tap_flags;
+ int ret;
+
+ ret = vhost_kernel_ioctl(data->vhostfds[0], VHOST_GET_FEATURES, features);
+ if (ret < 0) {
+ PMD_DRV_LOG(ERR, "Failed to get features");
+ return -1;
+ }
+
+ ret = tap_get_flags(data->tapfds[0], &tap_flags);
+ if (ret < 0) {
+ PMD_DRV_LOG(ERR, "Failed to get TAP features");
+ return -1;
+ }
+
+ /* with tap as the backend, all these features are supported
+ * but not claimed by vhost-net, so we add them back when
+ * reporting to upper layer.
+ */
+ if (tap_flags & IFF_VNET_HDR) {
+ *features |= VHOST_KERNEL_GUEST_OFFLOADS_MASK;
+ *features |= VHOST_KERNEL_HOST_OFFLOADS_MASK;
+ }
+
+ /* vhost_kernel will not declare this feature, but it does
+ * support multi-queue.
+ */
+ if (tap_flags & IFF_MULTI_QUEUE)
+ *features |= (1ull << VIRTIO_NET_F_MQ);
+
+ return 0;
+}
+
+static int
+vhost_kernel_set_features(struct virtio_user_dev *dev, uint64_t features)
+{
+ struct vhost_kernel_data *data = dev->backend_data;
+ uint32_t i;
+ int ret;
+
+ /* We don't need memory protection here */
+ features &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
+ /* VHOST kernel does not know about below flags */
+ features &= ~VHOST_KERNEL_GUEST_OFFLOADS_MASK;
+ features &= ~VHOST_KERNEL_HOST_OFFLOADS_MASK;
+ features &= ~(1ULL << VIRTIO_NET_F_MQ);
+
+ for (i = 0; i < dev->max_queue_pairs; ++i) {
+ if (data->vhostfds[i] < 0)
+ continue;
+
+ ret = vhost_kernel_ioctl(data->vhostfds[i], VHOST_SET_FEATURES, &features);
+ if (ret < 0)
+ return ret;
+ }
+
+ return 0;
+}
static int
add_memseg_list(const struct rte_memseg_list *msl, void *arg)
* have much more memory regions. Below function will treat each
* contiguous memory space reserved by DPDK as one region.
*/
-static struct vhost_memory_kernel *
-prepare_vhost_memory_kernel(void)
+static int
+vhost_kernel_set_memory_table(struct virtio_user_dev *dev)
{
+ uint32_t i;
+ struct vhost_kernel_data *data = dev->backend_data;
struct vhost_memory_kernel *vm;
+ int ret;
vm = malloc(sizeof(struct vhost_memory_kernel) +
max_regions *
sizeof(struct vhost_memory_region));
if (!vm)
- return NULL;
+ goto err;
vm->nregions = 0;
vm->padding = 0;
* The memory lock has already been taken by memory subsystem
* or virtio_user_start_device().
*/
- if (rte_memseg_list_walk_thread_unsafe(add_memseg_list, vm) < 0) {
- free(vm);
- return NULL;
- }
+ ret = rte_memseg_list_walk_thread_unsafe(add_memseg_list, vm);
+ if (ret < 0)
+ goto err_free;
- return vm;
-}
+ for (i = 0; i < dev->max_queue_pairs; ++i) {
+ if (data->vhostfds[i] < 0)
+ continue;
-/* with below features, vhost kernel does not need to do the checksum and TSO,
- * these info will be passed to virtio_user through virtio net header.
- */
-#define VHOST_KERNEL_GUEST_OFFLOADS_MASK \
- ((1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
- (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
- (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
- (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
- (1ULL << VIRTIO_NET_F_GUEST_UFO))
+ ret = vhost_kernel_ioctl(data->vhostfds[i], VHOST_SET_MEM_TABLE, vm);
+ if (ret < 0)
+ goto err_free;
+ }
-/* with below features, when flows from virtio_user to vhost kernel
- * (1) if flows goes up through the kernel networking stack, it does not need
- * to verify checksum, which can save CPU cycles;
- * (2) if flows goes through a Linux bridge and outside from an interface
- * (kernel driver), checksum and TSO will be done by GSO in kernel or even
- * offloaded into real physical device.
- */
-#define VHOST_KERNEL_HOST_OFFLOADS_MASK \
- ((1ULL << VIRTIO_NET_F_HOST_TSO4) | \
- (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
- (1ULL << VIRTIO_NET_F_CSUM))
+ free(vm);
+
+ return 0;
+err_free:
+ free(vm);
+err:
+ PMD_DRV_LOG(ERR, "Failed to set memory table");
+ return -1;
+}
-static unsigned int
-tap_support_features(void)
+static int
+vhost_kernel_set_vring(struct virtio_user_dev *dev, uint64_t req, struct vhost_vring_state *state)
{
- int tapfd;
- unsigned int tap_features;
+ int ret, fd;
+ unsigned int index = state->index;
+ struct vhost_kernel_data *data = dev->backend_data;
- 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;
- }
+ /* Convert from queue index to queue-pair & offset */
+ fd = data->vhostfds[state->index / 2];
+ state->index %= 2;
- if (ioctl(tapfd, TUNGETFEATURES, &tap_features) == -1) {
- PMD_DRV_LOG(ERR, "TUNGETFEATURES failed: %s", strerror(errno));
- close(tapfd);
+ ret = vhost_kernel_ioctl(fd, req, state);
+ if (ret < 0) {
+ PMD_DRV_LOG(ERR, "Failed to set vring (request %" PRIu64 ")", req);
return -1;
}
- close(tapfd);
- return tap_features;
+ /* restore index back to queue index */
+ state->index = index;
+
+ return 0;
}
static int
-vhost_kernel_ioctl(struct virtio_user_dev *dev,
- enum vhost_user_request req,
- void *arg)
+vhost_kernel_set_vring_num(struct virtio_user_dev *dev, struct vhost_vring_state *state)
{
- int ret = -1;
- unsigned int i;
- uint64_t req_kernel;
- struct vhost_memory_kernel *vm = NULL;
- int vhostfd;
- unsigned int queue_sel;
- unsigned int features;
+ return vhost_kernel_set_vring(dev, VHOST_SET_VRING_NUM, state);
+}
+
+static int
+vhost_kernel_set_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
+{
+ return vhost_kernel_set_vring(dev, VHOST_SET_VRING_BASE, state);
+}
+
+static int
+vhost_kernel_get_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
+{
+ return vhost_kernel_set_vring(dev, VHOST_GET_VRING_BASE, state);
+}
- PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
+static int
+vhost_kernel_set_vring_file(struct virtio_user_dev *dev, uint64_t req,
+ struct vhost_vring_file *file)
+{
+ int ret, fd;
+ unsigned int index = file->index;
+ struct vhost_kernel_data *data = dev->backend_data;
- req_kernel = vhost_req_user_to_kernel[req];
+ /* Convert from queue index to queue-pair & offset */
+ fd = data->vhostfds[file->index / 2];
+ file->index %= 2;
- if (req_kernel == VHOST_SET_MEM_TABLE) {
- vm = prepare_vhost_memory_kernel();
- if (!vm)
- return -1;
- arg = (void *)vm;
+ ret = vhost_kernel_ioctl(fd, req, file);
+ if (ret < 0) {
+ PMD_DRV_LOG(ERR, "Failed to set vring file (request %" PRIu64 ")", req);
+ return -1;
}
- if (req_kernel == VHOST_SET_FEATURES) {
- /* We don't need memory protection here */
- *(uint64_t *)arg &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
+ /* restore index back to queue index */
+ file->index = index;
- /* VHOST kernel does not know about below flags */
- *(uint64_t *)arg &= ~VHOST_KERNEL_GUEST_OFFLOADS_MASK;
- *(uint64_t *)arg &= ~VHOST_KERNEL_HOST_OFFLOADS_MASK;
+ return 0;
+}
- *(uint64_t *)arg &= ~(1ULL << VIRTIO_NET_F_MQ);
- }
+static int
+vhost_kernel_set_vring_kick(struct virtio_user_dev *dev, struct vhost_vring_file *file)
+{
+ return vhost_kernel_set_vring_file(dev, VHOST_SET_VRING_KICK, file);
+}
- switch (req_kernel) {
- case VHOST_SET_VRING_NUM:
- case VHOST_SET_VRING_ADDR:
- case VHOST_SET_VRING_BASE:
- case VHOST_GET_VRING_BASE:
- case VHOST_SET_VRING_KICK:
- case VHOST_SET_VRING_CALL:
- queue_sel = *(unsigned int *)arg;
- vhostfd = dev->vhostfds[queue_sel / 2];
- *(unsigned int *)arg = queue_sel % 2;
- PMD_DRV_LOG(DEBUG, "vhostfd=%d, index=%u",
- vhostfd, *(unsigned int *)arg);
- break;
- default:
- vhostfd = -1;
- }
- if (vhostfd == -1) {
- 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;
- }
- } else {
- ret = ioctl(vhostfd, req_kernel, arg);
- }
+static int
+vhost_kernel_set_vring_call(struct virtio_user_dev *dev, struct vhost_vring_file *file)
+{
+ return vhost_kernel_set_vring_file(dev, VHOST_SET_VRING_CALL, file);
+}
- if (!ret && req_kernel == VHOST_GET_FEATURES) {
- features = tap_support_features();
- /* with tap as the backend, all these features are supported
- * but not claimed by vhost-net, so we add them back when
- * reporting to upper layer.
- */
- if (features & IFF_VNET_HDR) {
- *((uint64_t *)arg) |= VHOST_KERNEL_GUEST_OFFLOADS_MASK;
- *((uint64_t *)arg) |= VHOST_KERNEL_HOST_OFFLOADS_MASK;
- }
+static int
+vhost_kernel_set_vring_addr(struct virtio_user_dev *dev, struct vhost_vring_addr *addr)
+{
+ int ret, fd;
+ unsigned int index = addr->index;
+ struct vhost_kernel_data *data = dev->backend_data;
+
+ /* Convert from queue index to queue-pair & offset */
+ fd = data->vhostfds[addr->index / 2];
+ addr->index %= 2;
- /* vhost_kernel will not declare this feature, but it does
- * support multi-queue.
- */
- if (features & IFF_MULTI_QUEUE)
- *(uint64_t *)arg |= (1ull << VIRTIO_NET_F_MQ);
+ ret = vhost_kernel_ioctl(fd, VHOST_SET_VRING_ADDR, addr);
+ if (ret < 0) {
+ PMD_DRV_LOG(ERR, "Failed to set vring address");
+ return -1;
}
- if (vm)
- free(vm);
+ /* restore index back to queue index */
+ addr->index = index;
- if (ret < 0)
- PMD_DRV_LOG(ERR, "%s failed: %s",
- vhost_msg_strings[req], strerror(errno));
+ return 0;
+}
- return ret;
+static int
+vhost_kernel_get_status(struct virtio_user_dev *dev __rte_unused, uint8_t *status __rte_unused)
+{
+ return -ENOTSUP;
+}
+
+static int
+vhost_kernel_set_status(struct virtio_user_dev *dev __rte_unused, uint8_t status __rte_unused)
+{
+ return -ENOTSUP;
}
/**
static int
vhost_kernel_setup(struct virtio_user_dev *dev)
{
+ struct vhost_kernel_data *data;
+ unsigned int tap_features;
+ unsigned int tap_flags;
+ const char *ifname;
+ uint32_t q, i;
int vhostfd;
- uint32_t i;
+
+ if (tap_support_features(&tap_features) < 0)
+ return -1;
+
+ if ((tap_features & IFF_VNET_HDR) == 0) {
+ PMD_INIT_LOG(ERR, "TAP does not support IFF_VNET_HDR");
+ return -1;
+ }
+
+ data = malloc(sizeof(*data));
+ if (!data) {
+ PMD_INIT_LOG(ERR, "(%s) Failed to allocate Vhost-kernel data", dev->path);
+ return -1;
+ }
+
+ data->vhostfds = malloc(dev->max_queue_pairs * sizeof(int));
+ if (!data->vhostfds) {
+ PMD_INIT_LOG(ERR, "(%s) Failed to allocate Vhost FDs", dev->path);
+ goto err_data;
+ }
+ data->tapfds = malloc(dev->max_queue_pairs * sizeof(int));
+ if (!data->tapfds) {
+ PMD_INIT_LOG(ERR, "(%s) Failed to allocate TAP FDs", dev->path);
+ goto err_vhostfds;
+ }
+
+ for (q = 0; q < dev->max_queue_pairs; ++q) {
+ data->vhostfds[q] = -1;
+ data->tapfds[q] = -1;
+ }
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;
+ PMD_DRV_LOG(ERR, "fail to open %s, %s", dev->path, strerror(errno));
+ goto err_tapfds;
}
+ data->vhostfds[i] = vhostfd;
+ }
+
+ ifname = dev->ifname != NULL ? dev->ifname : "tap%d";
+ data->tapfds[0] = tap_open(ifname, (tap_features & IFF_MULTI_QUEUE) != 0);
+ if (data->tapfds[0] < 0)
+ goto err_tapfds;
+ if (dev->ifname == NULL && tap_get_name(data->tapfds[0], &dev->ifname) < 0) {
+ PMD_DRV_LOG(ERR, "fail to get tap name (%d)", data->tapfds[0]);
+ goto err_tapfds;
+ }
+ if (tap_get_flags(data->tapfds[0], &tap_flags) < 0) {
+ PMD_DRV_LOG(ERR, "fail to get tap flags for tap %s", dev->ifname);
+ goto err_tapfds;
+ }
+ if ((tap_flags & IFF_MULTI_QUEUE) == 0 && dev->max_queue_pairs > 1) {
+ PMD_DRV_LOG(ERR, "tap %s does not support multi queue", dev->ifname);
+ goto err_tapfds;
+ }
+
+ for (i = 1; i < dev->max_queue_pairs; i++) {
+ data->tapfds[i] = tap_open(dev->ifname, true);
+ if (data->tapfds[i] < 0)
+ goto err_tapfds;
+ }
+
+ dev->backend_data = data;
+
+ return 0;
+
+err_tapfds:
+ for (i = 0; i < dev->max_queue_pairs; i++) {
+ if (data->vhostfds[i] >= 0)
+ close(data->vhostfds[i]);
+ if (data->tapfds[i] >= 0)
+ close(data->tapfds[i]);
+ }
+
+ free(data->tapfds);
+err_vhostfds:
+ free(data->vhostfds);
+err_data:
+ free(data);
+
+ return -1;
+}
+
+static int
+vhost_kernel_destroy(struct virtio_user_dev *dev)
+{
+ struct vhost_kernel_data *data = dev->backend_data;
+ uint32_t i;
+
+ if (!data)
+ return 0;
- dev->vhostfds[i] = vhostfd;
+ for (i = 0; i < dev->max_queue_pairs; ++i) {
+ if (data->vhostfds[i] >= 0)
+ close(data->vhostfds[i]);
+ if (data->tapfds[i] >= 0)
+ close(data->tapfds[i]);
}
+ free(data->vhostfds);
+ free(data->tapfds);
+ free(data);
+ dev->backend_data = NULL;
+
return 0;
}
uint16_t pair_idx,
int enable)
{
+ struct vhost_kernel_data *data = dev->backend_data;
int hdr_size;
int vhostfd;
int tapfd;
- int req_mq = (dev->max_queue_pairs > 1);
- vhostfd = dev->vhostfds[pair_idx];
+ if (dev->qp_enabled[pair_idx] == enable)
+ return 0;
+
+ vhostfd = data->vhostfds[pair_idx];
+ tapfd = data->tapfds[pair_idx];
if (!enable) {
- if (dev->tapfds[pair_idx] >= 0) {
- close(dev->tapfds[pair_idx]);
- dev->tapfds[pair_idx] = -1;
+ if (vhost_kernel_set_backend(vhostfd, -1) < 0) {
+ PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
+ return -1;
}
- return vhost_kernel_set_backend(vhostfd, -1);
- } else if (dev->tapfds[pair_idx] >= 0) {
+ dev->qp_enabled[pair_idx] = false;
return 0;
}
else
hdr_size = sizeof(struct virtio_net_hdr);
- tapfd = vhost_kernel_open_tap(&dev->ifname, hdr_size, req_mq,
- (char *)dev->mac_addr, dev->features);
- if (tapfd < 0) {
- PMD_DRV_LOG(ERR, "fail to open tap for vhost kernel");
+ /* Set mac on tap only once when starting */
+ if (!dev->started && pair_idx == 0 &&
+ tap_set_mac(data->tapfds[pair_idx], dev->mac_addr) < 0)
+ return -1;
+
+ if (vhost_kernel_tap_setup(tapfd, hdr_size, dev->features) < 0) {
+ PMD_DRV_LOG(ERR, "fail to setup 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;
+ dev->qp_enabled[pair_idx] = true;
+ return 0;
+}
+
+static int
+vhost_kernel_get_backend_features(uint64_t *features)
+{
+ *features = 0;
+
return 0;
}
+static int
+vhost_kernel_update_link_state(struct virtio_user_dev *dev __rte_unused)
+{
+ /* Nothing to update (Maybe get TAP interface link state?) */
+ return 0;
+}
+
+static int
+vhost_kernel_get_intr_fd(struct virtio_user_dev *dev __rte_unused)
+{
+ /* No link state interrupt with Vhost-kernel */
+ return -1;
+}
+
struct virtio_user_backend_ops virtio_ops_kernel = {
.setup = vhost_kernel_setup,
- .send_request = vhost_kernel_ioctl,
- .enable_qp = vhost_kernel_enable_queue_pair
+ .destroy = vhost_kernel_destroy,
+ .get_backend_features = vhost_kernel_get_backend_features,
+ .set_owner = vhost_kernel_set_owner,
+ .get_features = vhost_kernel_get_features,
+ .set_features = vhost_kernel_set_features,
+ .set_memory_table = vhost_kernel_set_memory_table,
+ .set_vring_num = vhost_kernel_set_vring_num,
+ .set_vring_base = vhost_kernel_set_vring_base,
+ .get_vring_base = vhost_kernel_get_vring_base,
+ .set_vring_call = vhost_kernel_set_vring_call,
+ .set_vring_kick = vhost_kernel_set_vring_kick,
+ .set_vring_addr = vhost_kernel_set_vring_addr,
+ .get_status = vhost_kernel_get_status,
+ .set_status = vhost_kernel_set_status,
+ .enable_qp = vhost_kernel_enable_queue_pair,
+ .update_link_state = vhost_kernel_update_link_state,
+ .get_intr_fd = vhost_kernel_get_intr_fd,
};