xref: /openbmc/qemu/hw/net/vhost_net.c (revision f2426947decc96bde4f9ea50097dac66a0a48acc) 
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 void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
192 {
193     net->dev.vq_index = vq_index;
194 }
195 
196 static int vhost_net_start_one(struct vhost_net *net,
197                                VirtIODevice *dev)
198 {
199     struct vhost_vring_file file = { };
200     int r;
201 
202     net->dev.nvqs = 2;
203     net->dev.vqs = net->vqs;
204 
205     r = vhost_dev_enable_notifiers(&net->dev, dev);
206     if (r < 0) {
207         goto fail_notifiers;
208     }
209 
210     r = vhost_dev_start(&net->dev, dev);
211     if (r < 0) {
212         goto fail_start;
213     }
214 
215     if (net->nc->info->poll) {
216         net->nc->info->poll(net->nc, false);
217     }
218 
219     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
220         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
221         file.fd = net->backend;
222         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
223             const VhostOps *vhost_ops = net->dev.vhost_ops;
224             r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
225                                       &file);
226             if (r < 0) {
227                 r = -errno;
228                 goto fail;
229             }
230         }
231     }
232     return 0;
233 fail:
234     file.fd = -1;
235     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
236         while (file.index-- > 0) {
237             const VhostOps *vhost_ops = net->dev.vhost_ops;
238             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
239                                           &file);
240             assert(r >= 0);
241         }
242     }
243     if (net->nc->info->poll) {
244         net->nc->info->poll(net->nc, true);
245     }
246     vhost_dev_stop(&net->dev, dev);
247 fail_start:
248     vhost_dev_disable_notifiers(&net->dev, dev);
249 fail_notifiers:
250     return r;
251 }
252 
253 static void vhost_net_stop_one(struct vhost_net *net,
254                                VirtIODevice *dev)
255 {
256     struct vhost_vring_file file = { .fd = -1 };
257 
258     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
259         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
260             const VhostOps *vhost_ops = net->dev.vhost_ops;
261             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
262                                           &file);
263             assert(r >= 0);
264         }
265     }
266     if (net->nc->info->poll) {
267         net->nc->info->poll(net->nc, true);
268     }
269     vhost_dev_stop(&net->dev, dev);
270     vhost_dev_disable_notifiers(&net->dev, dev);
271 }
272 
273 static bool vhost_net_device_endian_ok(VirtIODevice *vdev)
274 {
275 #ifdef TARGET_IS_BIENDIAN
276 #ifdef HOST_WORDS_BIGENDIAN
277     return virtio_is_big_endian(vdev);
278 #else
279     return !virtio_is_big_endian(vdev);
280 #endif
281 #else
282     return true;
283 #endif
284 }
285 
286 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
287                     int total_queues)
288 {
289     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
290     VirtioBusState *vbus = VIRTIO_BUS(qbus);
291     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
292     int r, i = 0;
293 
294     if (!vhost_net_device_endian_ok(dev)) {
295         error_report("vhost-net does not support cross-endian");
296         r = -ENOSYS;
297         goto err;
298     }
299 
300     if (!k->set_guest_notifiers) {
301         error_report("binding does not support guest notifiers");
302         r = -ENOSYS;
303         goto err;
304     }
305 
306     for (i = 0; i < total_queues; i++) {
307         vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
308     }
309 
310     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
311     if (r < 0) {
312         error_report("Error binding guest notifier: %d", -r);
313         goto err;
314     }
315 
316     for (i = 0; i < total_queues; i++) {
317         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
318 
319         if (r < 0) {
320             goto err;
321         }
322     }
323 
324     return 0;
325 
326 err:
327     while (--i >= 0) {
328         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
329     }
330     return r;
331 }
332 
333 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
334                     int total_queues)
335 {
336     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
337     VirtioBusState *vbus = VIRTIO_BUS(qbus);
338     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
339     int i, r;
340 
341     for (i = 0; i < total_queues; i++) {
342         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
343     }
344 
345     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
346     if (r < 0) {
347         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
348         fflush(stderr);
349     }
350     assert(r >= 0);
351 }
352 
353 void vhost_net_cleanup(struct vhost_net *net)
354 {
355     vhost_dev_cleanup(&net->dev);
356     g_free(net);
357 }
358 
359 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
360 {
361     return vhost_virtqueue_pending(&net->dev, idx);
362 }
363 
364 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
365                               int idx, bool mask)
366 {
367     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
368 }
369 
370 VHostNetState *get_vhost_net(NetClientState *nc)
371 {
372     VHostNetState *vhost_net = 0;
373 
374     if (!nc) {
375         return 0;
376     }
377 
378     switch (nc->info->type) {
379     case NET_CLIENT_OPTIONS_KIND_TAP:
380         vhost_net = tap_get_vhost_net(nc);
381         break;
382     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
383         vhost_net = vhost_user_get_vhost_net(nc);
384         break;
385     default:
386         break;
387     }
388 
389     return vhost_net;
390 }
391 #else
392 struct vhost_net *vhost_net_init(VhostNetOptions *options)
393 {
394     error_report("vhost-net support is not compiled in");
395     return NULL;
396 }
397 
398 bool vhost_net_query(VHostNetState *net, VirtIODevice *dev)
399 {
400     return false;
401 }
402 
403 int vhost_net_start(VirtIODevice *dev,
404                     NetClientState *ncs,
405                     int total_queues)
406 {
407     return -ENOSYS;
408 }
409 void vhost_net_stop(VirtIODevice *dev,
410                     NetClientState *ncs,
411                     int total_queues)
412 {
413 }
414 
415 void vhost_net_cleanup(struct vhost_net *net)
416 {
417 }
418 
419 unsigned vhost_net_get_features(struct vhost_net *net, unsigned features)
420 {
421     return features;
422 }
423 void vhost_net_ack_features(struct vhost_net *net, unsigned features)
424 {
425 }
426 
427 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
428 {
429     return false;
430 }
431 
432 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
433                               int idx, bool mask)
434 {
435 }
436 
437 VHostNetState *get_vhost_net(NetClientState *nc)
438 {
439     return 0;
440 }
441 #endif
442