xref: /openbmc/qemu/hw/net/vhost_net.c (revision 07a32d6b) 
1 /*
2  * vhost-net support
3  *
4  * Copyright Red Hat, Inc. 2010
5  *
6  * Authors:
7  *  Michael S. Tsirkin <mst@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
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.
14  */
15 
16 #include "net/net.h"
17 #include "net/tap.h"
18 #include "net/vhost-user.h"
19 
20 #include "hw/virtio/virtio-net.h"
21 #include "net/vhost_net.h"
22 #include "qemu/error-report.h"
23 
24 #include "config.h"
25 
26 #ifdef CONFIG_VHOST_NET
27 #include <linux/vhost.h>
28 #include <sys/socket.h>
29 #include <linux/kvm.h>
30 #include <fcntl.h>
31 #include <linux/virtio_ring.h>
32 #include <netpacket/packet.h>
33 #include <net/ethernet.h>
34 #include <net/if.h>
35 #include <netinet/in.h>
36 
37 #include <stdio.h>
38 
39 #include "hw/virtio/vhost.h"
40 #include "hw/virtio/virtio-bus.h"
41 
42 struct vhost_net {
43     struct vhost_dev dev;
44     struct vhost_virtqueue vqs[2];
45     int backend;
46     NetClientState *nc;
47 };
48 
49 /* Features supported by host kernel. */
50 static const int kernel_feature_bits[] = {
51     VIRTIO_F_NOTIFY_ON_EMPTY,
52     VIRTIO_RING_F_INDIRECT_DESC,
53     VIRTIO_RING_F_EVENT_IDX,
54     VIRTIO_NET_F_MRG_RXBUF,
55     VHOST_INVALID_FEATURE_BIT
56 };
57 
58 /* Features supported by others. */
59 const int user_feature_bits[] = {
60     VIRTIO_F_NOTIFY_ON_EMPTY,
61     VIRTIO_RING_F_INDIRECT_DESC,
62     VIRTIO_RING_F_EVENT_IDX,
63 
64     VIRTIO_F_ANY_LAYOUT,
65     VIRTIO_NET_F_CSUM,
66     VIRTIO_NET_F_GUEST_CSUM,
67     VIRTIO_NET_F_GSO,
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,
77     VIRTIO_NET_F_STATUS,
78     VIRTIO_NET_F_CTRL_VQ,
79     VIRTIO_NET_F_CTRL_RX,
80     VIRTIO_NET_F_CTRL_VLAN,
81     VIRTIO_NET_F_CTRL_RX_EXTRA,
82     VIRTIO_NET_F_CTRL_MAC_ADDR,
83     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,
84 
85     VIRTIO_NET_F_MQ,
86 
87     VHOST_INVALID_FEATURE_BIT
88 };
89 
90 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
91 {
92     const int *feature_bits = 0;
93 
94     switch (net->nc->info->type) {
95     case NET_CLIENT_OPTIONS_KIND_TAP:
96         feature_bits = kernel_feature_bits;
97         break;
98     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
99         feature_bits = user_feature_bits;
100         break;
101     default:
102         error_report("Feature bits not defined for this type: %d",
103                 net->nc->info->type);
104         break;
105     }
106 
107     return feature_bits;
108 }
109 
110 unsigned vhost_net_get_features(struct vhost_net *net, unsigned features)
111 {
112     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
113             features);
114 }
115 
116 void vhost_net_ack_features(struct vhost_net *net, unsigned features)
117 {
118     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
119 }
120 
121 static int vhost_net_get_fd(NetClientState *backend)
122 {
123     switch (backend->info->type) {
124     case NET_CLIENT_OPTIONS_KIND_TAP:
125         return tap_get_fd(backend);
126     default:
127         fprintf(stderr, "vhost-net requires tap backend\n");
128         return -EBADFD;
129     }
130 }
131 
132 struct vhost_net *vhost_net_init(VhostNetOptions *options)
133 {
134     int r;
135     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
136     struct vhost_net *net = g_malloc(sizeof *net);
137 
138     if (!options->net_backend) {
139         fprintf(stderr, "vhost-net requires net backend to be setup\n");
140         goto fail;
141     }
142 
143     if (backend_kernel) {
144         r = vhost_net_get_fd(options->net_backend);
145         if (r < 0) {
146             goto fail;
147         }
148         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
149             ? 0 : (1 << VHOST_NET_F_VIRTIO_NET_HDR);
150         net->backend = r;
151     } else {
152         net->dev.backend_features = 0;
153         net->backend = -1;
154     }
155     net->nc = options->net_backend;
156 
157     net->dev.nvqs = 2;
158     net->dev.vqs = net->vqs;
159 
160     r = vhost_dev_init(&net->dev, options->opaque,
161                        options->backend_type, options->force);
162     if (r < 0) {
163         goto fail;
164     }
165     if (!qemu_has_vnet_hdr_len(options->net_backend,
166                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
167         net->dev.features &= ~(1 << VIRTIO_NET_F_MRG_RXBUF);
168     }
169     if (backend_kernel) {
170         if (~net->dev.features & net->dev.backend_features) {
171             fprintf(stderr, "vhost lacks feature mask %" PRIu64
172                    " for backend\n",
173                    (uint64_t)(~net->dev.features & net->dev.backend_features));
174             vhost_dev_cleanup(&net->dev);
175             goto fail;
176         }
177     }
178     /* Set sane init value. Override when guest acks. */
179     vhost_net_ack_features(net, 0);
180     return net;
181 fail:
182     g_free(net);
183     return NULL;
184 }
185 
186 bool vhost_net_query(VHostNetState *net, VirtIODevice *dev)
187 {
188     return vhost_dev_query(&net->dev, dev);
189 }
190 
191 static int vhost_net_start_one(struct vhost_net *net,
192                                VirtIODevice *dev,
193                                int vq_index)
194 {
195     struct vhost_vring_file file = { };
196     int r;
197 
198     if (net->dev.started) {
199         return 0;
200     }
201 
202     net->dev.nvqs = 2;
203     net->dev.vqs = net->vqs;
204     net->dev.vq_index = vq_index;
205 
206     r = vhost_dev_enable_notifiers(&net->dev, dev);
207     if (r < 0) {
208         goto fail_notifiers;
209     }
210 
211     r = vhost_dev_start(&net->dev, dev);
212     if (r < 0) {
213         goto fail_start;
214     }
215 
216     if (net->nc->info->poll) {
217         net->nc->info->poll(net->nc, false);
218     }
219 
220     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
221         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
222         file.fd = net->backend;
223         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
224             const VhostOps *vhost_ops = net->dev.vhost_ops;
225             r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
226                                       &file);
227             if (r < 0) {
228                 r = -errno;
229                 goto fail;
230             }
231         }
232     }
233     return 0;
234 fail:
235     file.fd = -1;
236     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
237         while (file.index-- > 0) {
238             const VhostOps *vhost_ops = net->dev.vhost_ops;
239             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
240                                           &file);
241             assert(r >= 0);
242         }
243     }
244     if (net->nc->info->poll) {
245         net->nc->info->poll(net->nc, true);
246     }
247     vhost_dev_stop(&net->dev, dev);
248 fail_start:
249     vhost_dev_disable_notifiers(&net->dev, dev);
250 fail_notifiers:
251     return r;
252 }
253 
254 static void vhost_net_stop_one(struct vhost_net *net,
255                                VirtIODevice *dev)
256 {
257     struct vhost_vring_file file = { .fd = -1 };
258 
259     if (!net->dev.started) {
260         return;
261     }
262 
263     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
264         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
265             const VhostOps *vhost_ops = net->dev.vhost_ops;
266             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
267                                           &file);
268             assert(r >= 0);
269         }
270     }
271     if (net->nc->info->poll) {
272         net->nc->info->poll(net->nc, true);
273     }
274     vhost_dev_stop(&net->dev, dev);
275     vhost_dev_disable_notifiers(&net->dev, dev);
276 }
277 
278 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
279                     int total_queues)
280 {
281     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
282     VirtioBusState *vbus = VIRTIO_BUS(qbus);
283     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
284     int r, i = 0;
285 
286     if (!k->set_guest_notifiers) {
287         error_report("binding does not support guest notifiers");
288         r = -ENOSYS;
289         goto err;
290     }
291 
292     for (i = 0; i < total_queues; i++) {
293         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev, i * 2);
294 
295         if (r < 0) {
296             goto err;
297         }
298     }
299 
300     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
301     if (r < 0) {
302         error_report("Error binding guest notifier: %d", -r);
303         goto err;
304     }
305 
306     return 0;
307 
308 err:
309     while (--i >= 0) {
310         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
311     }
312     return r;
313 }
314 
315 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
316                     int total_queues)
317 {
318     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
319     VirtioBusState *vbus = VIRTIO_BUS(qbus);
320     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
321     int i, r;
322 
323     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
324     if (r < 0) {
325         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
326         fflush(stderr);
327     }
328     assert(r >= 0);
329 
330     for (i = 0; i < total_queues; i++) {
331         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
332     }
333 }
334 
335 void vhost_net_cleanup(struct vhost_net *net)
336 {
337     vhost_dev_cleanup(&net->dev);
338     g_free(net);
339 }
340 
341 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
342 {
343     return vhost_virtqueue_pending(&net->dev, idx);
344 }
345 
346 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
347                               int idx, bool mask)
348 {
349     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
350 }
351 
352 VHostNetState *get_vhost_net(NetClientState *nc)
353 {
354     VHostNetState *vhost_net = 0;
355 
356     if (!nc) {
357         return 0;
358     }
359 
360     switch (nc->info->type) {
361     case NET_CLIENT_OPTIONS_KIND_TAP:
362         vhost_net = tap_get_vhost_net(nc);
363         break;
364     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
365         vhost_net = vhost_user_get_vhost_net(nc);
366         break;
367     default:
368         break;
369     }
370 
371     return vhost_net;
372 }
373 #else
374 struct vhost_net *vhost_net_init(VhostNetOptions *options)
375 {
376     error_report("vhost-net support is not compiled in");
377     return NULL;
378 }
379 
380 bool vhost_net_query(VHostNetState *net, VirtIODevice *dev)
381 {
382     return false;
383 }
384 
385 int vhost_net_start(VirtIODevice *dev,
386                     NetClientState *ncs,
387                     int total_queues)
388 {
389     return -ENOSYS;
390 }
391 void vhost_net_stop(VirtIODevice *dev,
392                     NetClientState *ncs,
393                     int total_queues)
394 {
395 }
396 
397 void vhost_net_cleanup(struct vhost_net *net)
398 {
399 }
400 
401 unsigned vhost_net_get_features(struct vhost_net *net, unsigned features)
402 {
403     return features;
404 }
405 void vhost_net_ack_features(struct vhost_net *net, unsigned features)
406 {
407 }
408 
409 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
410 {
411     return false;
412 }
413 
414 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
415                               int idx, bool mask)
416 {
417 }
418 
419 VHostNetState *get_vhost_net(NetClientState *nc)
420 {
421     return 0;
422 }
423 #endif
424