4 * Copyright Red Hat, Inc. 2010
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
18 #include "net/vhost-user.h"
20 #include "hw/virtio/virtio-net.h"
21 #include "net/vhost_net.h"
22 #include "qemu/error-report.h"
26 #ifdef CONFIG_VHOST_NET
27 #include <linux/vhost.h>
28 #include <sys/socket.h>
29 #include <linux/kvm.h>
31 #include <netpacket/packet.h>
32 #include <net/ethernet.h>
34 #include <netinet/in.h>
38 #include "standard-headers/linux/virtio_ring.h"
39 #include "hw/virtio/vhost.h"
40 #include "hw/virtio/virtio-bus.h"
41 #include "hw/virtio/virtio-access.h"
45 struct vhost_virtqueue vqs[2];
50 /* Features supported by host kernel. */
51 static const int kernel_feature_bits[] = {
52 VIRTIO_F_NOTIFY_ON_EMPTY,
53 VIRTIO_RING_F_INDIRECT_DESC,
54 VIRTIO_RING_F_EVENT_IDX,
55 VIRTIO_NET_F_MRG_RXBUF,
57 VHOST_INVALID_FEATURE_BIT
60 /* Features supported by others. */
61 static const int user_feature_bits[] = {
62 VIRTIO_F_NOTIFY_ON_EMPTY,
63 VIRTIO_RING_F_INDIRECT_DESC,
64 VIRTIO_RING_F_EVENT_IDX,
69 VIRTIO_NET_F_GUEST_CSUM,
71 VIRTIO_NET_F_GUEST_TSO4,
72 VIRTIO_NET_F_GUEST_TSO6,
73 VIRTIO_NET_F_GUEST_ECN,
74 VIRTIO_NET_F_GUEST_UFO,
75 VIRTIO_NET_F_HOST_TSO4,
76 VIRTIO_NET_F_HOST_TSO6,
77 VIRTIO_NET_F_HOST_ECN,
78 VIRTIO_NET_F_HOST_UFO,
79 VIRTIO_NET_F_MRG_RXBUF,
83 VIRTIO_NET_F_CTRL_VLAN,
84 VIRTIO_NET_F_CTRL_RX_EXTRA,
85 VIRTIO_NET_F_CTRL_MAC_ADDR,
86 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,
88 VIRTIO_NET_F_GUEST_ANNOUNCE,
92 VHOST_INVALID_FEATURE_BIT
95 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
97 const int *feature_bits = 0;
99 switch (net->nc->info->type) {
100 case NET_CLIENT_OPTIONS_KIND_TAP:
101 feature_bits = kernel_feature_bits;
103 case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
104 feature_bits = user_feature_bits;
107 error_report("Feature bits not defined for this type: %d",
108 net->nc->info->type);
115 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
117 return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
121 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
123 net->dev.acked_features = net->dev.backend_features;
124 vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
127 uint64_t vhost_net_get_max_queues(VHostNetState *net)
129 return net->dev.max_queues;
132 static int vhost_net_get_fd(NetClientState *backend)
134 switch (backend->info->type) {
135 case NET_CLIENT_OPTIONS_KIND_TAP:
136 return tap_get_fd(backend);
138 fprintf(stderr, "vhost-net requires tap backend\n");
143 struct vhost_net *vhost_net_init(VhostNetOptions *options)
146 bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
147 struct vhost_net *net = g_malloc(sizeof *net);
149 if (!options->net_backend) {
150 fprintf(stderr, "vhost-net requires net backend to be setup\n");
153 net->nc = options->net_backend;
155 net->dev.max_queues = 1;
157 net->dev.vqs = net->vqs;
159 if (backend_kernel) {
160 r = vhost_net_get_fd(options->net_backend);
164 net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
165 ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
167 net->dev.protocol_features = 0;
169 net->dev.backend_features = 0;
170 net->dev.protocol_features = 0;
173 /* vhost-user needs vq_index to initiate a specific queue pair */
174 net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
177 r = vhost_dev_init(&net->dev, options->opaque,
178 options->backend_type);
182 if (backend_kernel) {
183 if (!qemu_has_vnet_hdr_len(options->net_backend,
184 sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
185 net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
187 if (~net->dev.features & net->dev.backend_features) {
188 fprintf(stderr, "vhost lacks feature mask %" PRIu64
190 (uint64_t)(~net->dev.features & net->dev.backend_features));
191 vhost_dev_cleanup(&net->dev);
195 /* Set sane init value. Override when guest acks. */
196 vhost_net_ack_features(net, 0);
203 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
205 net->dev.vq_index = vq_index;
208 static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
213 if (virtio_vdev_has_feature(dev, VIRTIO_F_VERSION_1) ||
214 (virtio_legacy_is_cross_endian(dev) && !virtio_is_big_endian(dev))) {
215 r = qemu_set_vnet_le(peer, set);
217 error_report("backend does not support LE vnet headers");
219 } else if (virtio_legacy_is_cross_endian(dev)) {
220 r = qemu_set_vnet_be(peer, set);
222 error_report("backend does not support BE vnet headers");
229 static int vhost_net_start_one(struct vhost_net *net,
232 struct vhost_vring_file file = { };
236 net->dev.vqs = net->vqs;
238 r = vhost_dev_enable_notifiers(&net->dev, dev);
243 r = vhost_dev_start(&net->dev, dev);
248 if (net->nc->info->poll) {
249 net->nc->info->poll(net->nc, false);
252 if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
253 qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
254 file.fd = net->backend;
255 for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
256 const VhostOps *vhost_ops = net->dev.vhost_ops;
257 r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
267 if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
268 while (file.index-- > 0) {
269 const VhostOps *vhost_ops = net->dev.vhost_ops;
270 int r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
274 if (net->nc->info->poll) {
275 net->nc->info->poll(net->nc, true);
277 vhost_dev_stop(&net->dev, dev);
279 vhost_dev_disable_notifiers(&net->dev, dev);
284 static void vhost_net_stop_one(struct vhost_net *net,
287 struct vhost_vring_file file = { .fd = -1 };
289 if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
290 for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
291 const VhostOps *vhost_ops = net->dev.vhost_ops;
292 int r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
296 if (net->nc->info->poll) {
297 net->nc->info->poll(net->nc, true);
299 vhost_dev_stop(&net->dev, dev);
300 vhost_dev_disable_notifiers(&net->dev, dev);
303 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
306 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
307 VirtioBusState *vbus = VIRTIO_BUS(qbus);
308 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
311 if (!k->set_guest_notifiers) {
312 error_report("binding does not support guest notifiers");
317 r = vhost_net_set_vnet_endian(dev, ncs[0].peer, true);
322 for (i = 0; i < total_queues; i++) {
323 vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
326 r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
328 error_report("Error binding guest notifier: %d", -r);
332 for (i = 0; i < total_queues; i++) {
333 r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
344 vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
346 e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
348 fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
352 vhost_net_set_vnet_endian(dev, ncs[0].peer, false);
357 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
360 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
361 VirtioBusState *vbus = VIRTIO_BUS(qbus);
362 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
365 for (i = 0; i < total_queues; i++) {
366 vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
369 r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
371 fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
376 assert(vhost_net_set_vnet_endian(dev, ncs[0].peer, false) >= 0);
379 void vhost_net_cleanup(struct vhost_net *net)
381 vhost_dev_cleanup(&net->dev);
385 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
387 const VhostOps *vhost_ops = net->dev.vhost_ops;
390 if (vhost_ops->vhost_migration_done) {
391 r = vhost_ops->vhost_migration_done(&net->dev, mac_addr);
397 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
399 return vhost_virtqueue_pending(&net->dev, idx);
402 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
405 vhost_virtqueue_mask(&net->dev, dev, idx, mask);
408 VHostNetState *get_vhost_net(NetClientState *nc)
410 VHostNetState *vhost_net = 0;
416 switch (nc->info->type) {
417 case NET_CLIENT_OPTIONS_KIND_TAP:
418 vhost_net = tap_get_vhost_net(nc);
420 case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
421 vhost_net = vhost_user_get_vhost_net(nc);
430 int vhost_set_vring_enable(NetClientState *nc, int enable)
432 VHostNetState *net = get_vhost_net(nc);
433 const VhostOps *vhost_ops = net->dev.vhost_ops;
435 if (vhost_ops->vhost_set_vring_enable) {
436 return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
443 uint64_t vhost_net_get_max_queues(VHostNetState *net)
448 struct vhost_net *vhost_net_init(VhostNetOptions *options)
450 error_report("vhost-net support is not compiled in");
454 int vhost_net_start(VirtIODevice *dev,
460 void vhost_net_stop(VirtIODevice *dev,
466 void vhost_net_cleanup(struct vhost_net *net)
470 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
474 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
478 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
483 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
488 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
493 VHostNetState *get_vhost_net(NetClientState *nc)
498 int vhost_set_vring_enable(NetClientState *nc, int enable)