4 * Copyright Red Hat, Inc. 2010
7 * Michael S. Tsirkin <mst@redhat.com>
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.
16 #include "qemu/osdep.h"
19 #include "net/vhost-user.h"
21 #include "hw/virtio/virtio-net.h"
22 #include "net/vhost_net.h"
23 #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>
30 #include <netpacket/packet.h>
31 #include <net/ethernet.h>
33 #include <netinet/in.h>
36 #include "standard-headers/linux/virtio_ring.h"
37 #include "hw/virtio/vhost.h"
38 #include "hw/virtio/virtio-bus.h"
42 struct vhost_virtqueue vqs[2];
47 /* Features supported by host kernel. */
48 static const int kernel_feature_bits[] = {
49 VIRTIO_F_NOTIFY_ON_EMPTY,
50 VIRTIO_RING_F_INDIRECT_DESC,
51 VIRTIO_RING_F_EVENT_IDX,
52 VIRTIO_NET_F_MRG_RXBUF,
54 VHOST_INVALID_FEATURE_BIT
57 /* Features supported by others. */
58 static const int user_feature_bits[] = {
59 VIRTIO_F_NOTIFY_ON_EMPTY,
60 VIRTIO_RING_F_INDIRECT_DESC,
61 VIRTIO_RING_F_EVENT_IDX,
66 VIRTIO_NET_F_GUEST_CSUM,
68 VIRTIO_NET_F_GUEST_TSO4,
69 VIRTIO_NET_F_GUEST_TSO6,
70 VIRTIO_NET_F_GUEST_ECN,
71 VIRTIO_NET_F_GUEST_UFO,
72 VIRTIO_NET_F_HOST_TSO4,
73 VIRTIO_NET_F_HOST_TSO6,
74 VIRTIO_NET_F_HOST_ECN,
75 VIRTIO_NET_F_HOST_UFO,
76 VIRTIO_NET_F_MRG_RXBUF,
78 /* This bit implies RARP isn't sent by QEMU out of band */
79 VIRTIO_NET_F_GUEST_ANNOUNCE,
83 VHOST_INVALID_FEATURE_BIT
86 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
88 const int *feature_bits = 0;
90 switch (net->nc->info->type) {
91 case NET_CLIENT_OPTIONS_KIND_TAP:
92 feature_bits = kernel_feature_bits;
94 case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
95 feature_bits = user_feature_bits;
98 error_report("Feature bits not defined for this type: %d",
106 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
108 return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
112 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
114 net->dev.acked_features = net->dev.backend_features;
115 vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
118 uint64_t vhost_net_get_max_queues(VHostNetState *net)
120 return net->dev.max_queues;
123 static int vhost_net_get_fd(NetClientState *backend)
125 switch (backend->info->type) {
126 case NET_CLIENT_OPTIONS_KIND_TAP:
127 return tap_get_fd(backend);
129 fprintf(stderr, "vhost-net requires tap backend\n");
134 struct vhost_net *vhost_net_init(VhostNetOptions *options)
137 bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
138 struct vhost_net *net = g_malloc(sizeof *net);
140 if (!options->net_backend) {
141 fprintf(stderr, "vhost-net requires net backend to be setup\n");
144 net->nc = options->net_backend;
146 net->dev.max_queues = 1;
148 net->dev.vqs = net->vqs;
150 if (backend_kernel) {
151 r = vhost_net_get_fd(options->net_backend);
155 net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
156 ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
158 net->dev.protocol_features = 0;
160 net->dev.backend_features = 0;
161 net->dev.protocol_features = 0;
164 /* vhost-user needs vq_index to initiate a specific queue pair */
165 net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
168 r = vhost_dev_init(&net->dev, options->opaque,
169 options->backend_type);
173 if (backend_kernel) {
174 if (!qemu_has_vnet_hdr_len(options->net_backend,
175 sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
176 net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
178 if (~net->dev.features & net->dev.backend_features) {
179 fprintf(stderr, "vhost lacks feature mask %" PRIu64
181 (uint64_t)(~net->dev.features & net->dev.backend_features));
182 vhost_dev_cleanup(&net->dev);
186 /* Set sane init value. Override when guest acks. */
187 vhost_net_ack_features(net, 0);
194 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
196 net->dev.vq_index = vq_index;
199 static int vhost_net_start_one(struct vhost_net *net,
202 struct vhost_vring_file file = { };
206 net->dev.vqs = net->vqs;
208 r = vhost_dev_enable_notifiers(&net->dev, dev);
213 r = vhost_dev_start(&net->dev, dev);
218 if (net->nc->info->poll) {
219 net->nc->info->poll(net->nc, false);
222 if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
223 qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
224 file.fd = net->backend;
225 for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
226 const VhostOps *vhost_ops = net->dev.vhost_ops;
227 r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
237 if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
238 while (file.index-- > 0) {
239 const VhostOps *vhost_ops = net->dev.vhost_ops;
240 int r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
244 if (net->nc->info->poll) {
245 net->nc->info->poll(net->nc, true);
247 vhost_dev_stop(&net->dev, dev);
249 vhost_dev_disable_notifiers(&net->dev, dev);
254 static void vhost_net_stop_one(struct vhost_net *net,
257 struct vhost_vring_file file = { .fd = -1 };
259 if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
260 for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
261 const VhostOps *vhost_ops = net->dev.vhost_ops;
262 int r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
266 if (net->nc->info->poll) {
267 net->nc->info->poll(net->nc, true);
269 vhost_dev_stop(&net->dev, dev);
270 vhost_dev_disable_notifiers(&net->dev, dev);
273 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
276 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
277 VirtioBusState *vbus = VIRTIO_BUS(qbus);
278 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
281 if (!k->set_guest_notifiers) {
282 error_report("binding does not support guest notifiers");
286 for (i = 0; i < total_queues; i++) {
287 struct vhost_net *net;
289 net = get_vhost_net(ncs[i].peer);
290 vhost_net_set_vq_index(net, i * 2);
292 /* Suppress the masking guest notifiers on vhost user
293 * because vhost user doesn't interrupt masking/unmasking
296 if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
297 dev->use_guest_notifier_mask = false;
301 r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
303 error_report("Error binding guest notifier: %d", -r);
307 for (i = 0; i < total_queues; i++) {
308 r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
319 vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
321 e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
323 fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
330 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
333 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
334 VirtioBusState *vbus = VIRTIO_BUS(qbus);
335 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
338 for (i = 0; i < total_queues; i++) {
339 vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
342 r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
344 fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
350 void vhost_net_cleanup(struct vhost_net *net)
352 vhost_dev_cleanup(&net->dev);
356 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
358 const VhostOps *vhost_ops = net->dev.vhost_ops;
361 if (vhost_ops->vhost_migration_done) {
362 r = vhost_ops->vhost_migration_done(&net->dev, mac_addr);
368 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
370 return vhost_virtqueue_pending(&net->dev, idx);
373 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
376 vhost_virtqueue_mask(&net->dev, dev, idx, mask);
379 VHostNetState *get_vhost_net(NetClientState *nc)
381 VHostNetState *vhost_net = 0;
387 switch (nc->info->type) {
388 case NET_CLIENT_OPTIONS_KIND_TAP:
389 vhost_net = tap_get_vhost_net(nc);
391 case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
392 vhost_net = vhost_user_get_vhost_net(nc);
401 int vhost_set_vring_enable(NetClientState *nc, int enable)
403 VHostNetState *net = get_vhost_net(nc);
404 const VhostOps *vhost_ops = net->dev.vhost_ops;
406 if (vhost_ops->vhost_set_vring_enable) {
407 return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
414 uint64_t vhost_net_get_max_queues(VHostNetState *net)
419 struct vhost_net *vhost_net_init(VhostNetOptions *options)
421 error_report("vhost-net support is not compiled in");
425 int vhost_net_start(VirtIODevice *dev,
431 void vhost_net_stop(VirtIODevice *dev,
437 void vhost_net_cleanup(struct vhost_net *net)
441 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
445 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
449 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
454 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
459 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
464 VHostNetState *get_vhost_net(NetClientState *nc)
469 int vhost_set_vring_enable(NetClientState *nc, int enable)