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.
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_DRIVER_TAP:
92 feature_bits = kernel_feature_bits;
94 case NET_CLIENT_DRIVER_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 uint64_t vhost_net_get_acked_features(VHostNetState *net)
125 return net->dev.acked_features;
128 static int vhost_net_get_fd(NetClientState *backend)
130 switch (backend->info->type) {
131 case NET_CLIENT_DRIVER_TAP:
132 return tap_get_fd(backend);
134 fprintf(stderr, "vhost-net requires tap backend\n");
139 struct vhost_net *vhost_net_init(VhostNetOptions *options)
142 bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
143 struct vhost_net *net = g_new0(struct vhost_net, 1);
144 uint64_t features = 0;
146 if (!options->net_backend) {
147 fprintf(stderr, "vhost-net requires net backend to be setup\n");
150 net->nc = options->net_backend;
152 net->dev.max_queues = 1;
154 net->dev.vqs = net->vqs;
156 if (backend_kernel) {
157 r = vhost_net_get_fd(options->net_backend);
161 net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
162 ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
164 net->dev.protocol_features = 0;
166 net->dev.backend_features = 0;
167 net->dev.protocol_features = 0;
170 /* vhost-user needs vq_index to initiate a specific queue pair */
171 net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
174 r = vhost_dev_init(&net->dev, options->opaque,
175 options->backend_type, options->busyloop_timeout);
179 if (backend_kernel) {
180 if (!qemu_has_vnet_hdr_len(options->net_backend,
181 sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
182 net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
184 if (~net->dev.features & net->dev.backend_features) {
185 fprintf(stderr, "vhost lacks feature mask %" PRIu64
187 (uint64_t)(~net->dev.features & net->dev.backend_features));
192 /* Set sane init value. Override when guest acks. */
193 if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
194 features = vhost_user_get_acked_features(net->nc);
195 if (~net->dev.features & features) {
196 fprintf(stderr, "vhost lacks feature mask %" PRIu64
198 (uint64_t)(~net->dev.features & features));
203 vhost_net_ack_features(net, features);
208 vhost_dev_cleanup(&net->dev);
213 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
215 net->dev.vq_index = vq_index;
218 static int vhost_net_start_one(struct vhost_net *net,
221 struct vhost_vring_file file = { };
225 net->dev.vqs = net->vqs;
227 r = vhost_dev_enable_notifiers(&net->dev, dev);
232 r = vhost_dev_start(&net->dev, dev);
237 if (net->nc->info->poll) {
238 net->nc->info->poll(net->nc, false);
241 if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
242 qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
243 file.fd = net->backend;
244 for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
245 r = vhost_net_set_backend(&net->dev, &file);
255 if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
256 while (file.index-- > 0) {
257 int r = vhost_net_set_backend(&net->dev, &file);
261 if (net->nc->info->poll) {
262 net->nc->info->poll(net->nc, true);
264 vhost_dev_stop(&net->dev, dev);
266 vhost_dev_disable_notifiers(&net->dev, dev);
271 static void vhost_net_stop_one(struct vhost_net *net,
274 struct vhost_vring_file file = { .fd = -1 };
276 if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
277 for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
278 int r = vhost_net_set_backend(&net->dev, &file);
282 if (net->nc->info->poll) {
283 net->nc->info->poll(net->nc, true);
285 vhost_dev_stop(&net->dev, dev);
286 vhost_dev_disable_notifiers(&net->dev, dev);
289 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
292 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
293 VirtioBusState *vbus = VIRTIO_BUS(qbus);
294 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
297 if (!k->set_guest_notifiers) {
298 error_report("binding does not support guest notifiers");
302 for (i = 0; i < total_queues; i++) {
303 struct vhost_net *net;
305 net = get_vhost_net(ncs[i].peer);
306 vhost_net_set_vq_index(net, i * 2);
308 /* Suppress the masking guest notifiers on vhost user
309 * because vhost user doesn't interrupt masking/unmasking
312 if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
313 dev->use_guest_notifier_mask = false;
317 r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
319 error_report("Error binding guest notifier: %d", -r);
323 for (i = 0; i < total_queues; i++) {
324 r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
330 if (ncs[i].peer->vring_enable) {
331 /* restore vring enable state */
332 r = vhost_set_vring_enable(ncs[i].peer, ncs[i].peer->vring_enable);
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);
355 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
358 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
359 VirtioBusState *vbus = VIRTIO_BUS(qbus);
360 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
363 for (i = 0; i < total_queues; i++) {
364 vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
367 r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
369 fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
375 void vhost_net_cleanup(struct vhost_net *net)
377 vhost_dev_cleanup(&net->dev);
380 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
382 const VhostOps *vhost_ops = net->dev.vhost_ops;
384 assert(vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
385 assert(vhost_ops->vhost_migration_done);
387 return vhost_ops->vhost_migration_done(&net->dev, mac_addr);
390 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
392 return vhost_virtqueue_pending(&net->dev, idx);
395 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
398 vhost_virtqueue_mask(&net->dev, dev, idx, mask);
401 VHostNetState *get_vhost_net(NetClientState *nc)
403 VHostNetState *vhost_net = 0;
409 switch (nc->info->type) {
410 case NET_CLIENT_DRIVER_TAP:
411 vhost_net = tap_get_vhost_net(nc);
413 case NET_CLIENT_DRIVER_VHOST_USER:
414 vhost_net = vhost_user_get_vhost_net(nc);
424 int vhost_set_vring_enable(NetClientState *nc, int enable)
426 VHostNetState *net = get_vhost_net(nc);
427 const VhostOps *vhost_ops = net->dev.vhost_ops;
429 nc->vring_enable = enable;
431 if (vhost_ops && vhost_ops->vhost_set_vring_enable) {
432 return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
439 uint64_t vhost_net_get_max_queues(VHostNetState *net)
444 struct vhost_net *vhost_net_init(VhostNetOptions *options)
446 error_report("vhost-net support is not compiled in");
450 int vhost_net_start(VirtIODevice *dev,
456 void vhost_net_stop(VirtIODevice *dev,
462 void vhost_net_cleanup(struct vhost_net *net)
466 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
471 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
475 uint64_t vhost_net_get_acked_features(VHostNetState *net)
480 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
485 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
490 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
495 VHostNetState *get_vhost_net(NetClientState *nc)
500 int vhost_set_vring_enable(NetClientState *nc, int enable)