xref: /qemu/hw/net/vhost_net.c (revision fc48ffc3) 
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 <netpacket/packet.h>
32 #include <net/ethernet.h>
33 #include <net/if.h>
34 #include <netinet/in.h>
35 
36 #include <stdio.h>
37 
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"
42 
43 struct vhost_net {
44     struct vhost_dev dev;
45     struct vhost_virtqueue vqs[2];
46     int backend;
47     NetClientState *nc;
48 };
49 
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,
56     VIRTIO_F_VERSION_1,
57     VHOST_INVALID_FEATURE_BIT
58 };
59 
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,
65 
66     VIRTIO_F_ANY_LAYOUT,
67     VIRTIO_F_VERSION_1,
68     VIRTIO_NET_F_CSUM,
69     VIRTIO_NET_F_GUEST_CSUM,
70     VIRTIO_NET_F_GSO,
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,
80     VIRTIO_NET_F_STATUS,
81     VIRTIO_NET_F_CTRL_VQ,
82     VIRTIO_NET_F_CTRL_RX,
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,
87 
88     VIRTIO_NET_F_MQ,
89 
90     VHOST_INVALID_FEATURE_BIT
91 };
92 
93 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
94 {
95     const int *feature_bits = 0;
96 
97     switch (net->nc->info->type) {
98     case NET_CLIENT_OPTIONS_KIND_TAP:
99         feature_bits = kernel_feature_bits;
100         break;
101     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
102         feature_bits = user_feature_bits;
103         break;
104     default:
105         error_report("Feature bits not defined for this type: %d",
106                 net->nc->info->type);
107         break;
108     }
109 
110     return feature_bits;
111 }
112 
113 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
114 {
115     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
116             features);
117 }
118 
119 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
120 {
121     net->dev.acked_features = net->dev.backend_features;
122     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
123 }
124 
125 static int vhost_net_get_fd(NetClientState *backend)
126 {
127     switch (backend->info->type) {
128     case NET_CLIENT_OPTIONS_KIND_TAP:
129         return tap_get_fd(backend);
130     default:
131         fprintf(stderr, "vhost-net requires tap backend\n");
132         return -EBADFD;
133     }
134 }
135 
136 struct vhost_net *vhost_net_init(VhostNetOptions *options)
137 {
138     int r;
139     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
140     struct vhost_net *net = g_malloc(sizeof *net);
141 
142     if (!options->net_backend) {
143         fprintf(stderr, "vhost-net requires net backend to be setup\n");
144         goto fail;
145     }
146 
147     if (backend_kernel) {
148         r = vhost_net_get_fd(options->net_backend);
149         if (r < 0) {
150             goto fail;
151         }
152         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
153             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
154         net->backend = r;
155     } else {
156         net->dev.backend_features = 0;
157         net->backend = -1;
158     }
159     net->nc = options->net_backend;
160 
161     net->dev.nvqs = 2;
162     net->dev.vqs = net->vqs;
163     net->dev.vq_index = net->nc->queue_index;
164 
165     r = vhost_dev_init(&net->dev, options->opaque,
166                        options->backend_type);
167     if (r < 0) {
168         goto fail;
169     }
170     if (backend_kernel) {
171         if (!qemu_has_vnet_hdr_len(options->net_backend,
172                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
173             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
174         }
175         if (~net->dev.features & net->dev.backend_features) {
176             fprintf(stderr, "vhost lacks feature mask %" PRIu64
177                    " for backend\n",
178                    (uint64_t)(~net->dev.features & net->dev.backend_features));
179             vhost_dev_cleanup(&net->dev);
180             goto fail;
181         }
182     }
183     /* Set sane init value. Override when guest acks. */
184     vhost_net_ack_features(net, 0);
185     return net;
186 fail:
187     g_free(net);
188     return NULL;
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_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
197                                      bool set)
198 {
199     int r = 0;
200 
201     if (virtio_has_feature(dev, VIRTIO_F_VERSION_1) ||
202         (virtio_legacy_is_cross_endian(dev) && !virtio_is_big_endian(dev))) {
203         r = qemu_set_vnet_le(peer, set);
204         if (r) {
205             error_report("backend does not support LE vnet headers");
206         }
207     } else if (virtio_legacy_is_cross_endian(dev)) {
208         r = qemu_set_vnet_be(peer, set);
209         if (r) {
210             error_report("backend does not support BE vnet headers");
211         }
212     }
213 
214     return r;
215 }
216 
217 static int vhost_net_start_one(struct vhost_net *net,
218                                VirtIODevice *dev)
219 {
220     struct vhost_vring_file file = { };
221     int r;
222 
223     net->dev.nvqs = 2;
224     net->dev.vqs = net->vqs;
225 
226     r = vhost_dev_enable_notifiers(&net->dev, dev);
227     if (r < 0) {
228         goto fail_notifiers;
229     }
230 
231     r = vhost_dev_start(&net->dev, dev);
232     if (r < 0) {
233         goto fail_start;
234     }
235 
236     if (net->nc->info->poll) {
237         net->nc->info->poll(net->nc, false);
238     }
239 
240     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
241         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
242         file.fd = net->backend;
243         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
244             const VhostOps *vhost_ops = net->dev.vhost_ops;
245             r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
246                                       &file);
247             if (r < 0) {
248                 r = -errno;
249                 goto fail;
250             }
251         }
252     }
253     return 0;
254 fail:
255     file.fd = -1;
256     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
257         while (file.index-- > 0) {
258             const VhostOps *vhost_ops = net->dev.vhost_ops;
259             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
260                                           &file);
261             assert(r >= 0);
262         }
263     }
264     if (net->nc->info->poll) {
265         net->nc->info->poll(net->nc, true);
266     }
267     vhost_dev_stop(&net->dev, dev);
268 fail_start:
269     vhost_dev_disable_notifiers(&net->dev, dev);
270 fail_notifiers:
271     return r;
272 }
273 
274 static void vhost_net_stop_one(struct vhost_net *net,
275                                VirtIODevice *dev)
276 {
277     struct vhost_vring_file file = { .fd = -1 };
278 
279     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
280         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
281             const VhostOps *vhost_ops = net->dev.vhost_ops;
282             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
283                                           &file);
284             assert(r >= 0);
285         }
286     } else if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
287         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
288             const VhostOps *vhost_ops = net->dev.vhost_ops;
289             int r = vhost_ops->vhost_call(&net->dev, VHOST_RESET_OWNER,
290                                           &file);
291             assert(r >= 0);
292         }
293     }
294     if (net->nc->info->poll) {
295         net->nc->info->poll(net->nc, true);
296     }
297     vhost_dev_stop(&net->dev, dev);
298     vhost_dev_disable_notifiers(&net->dev, dev);
299 }
300 
301 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
302                     int total_queues)
303 {
304     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
305     VirtioBusState *vbus = VIRTIO_BUS(qbus);
306     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
307     int r, e, i;
308 
309     if (!k->set_guest_notifiers) {
310         error_report("binding does not support guest notifiers");
311         r = -ENOSYS;
312         goto err;
313     }
314 
315     r = vhost_net_set_vnet_endian(dev, ncs[0].peer, true);
316     if (r < 0) {
317         goto err;
318     }
319 
320     for (i = 0; i < total_queues; i++) {
321         vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
322     }
323 
324     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
325     if (r < 0) {
326         error_report("Error binding guest notifier: %d", -r);
327         goto err_endian;
328     }
329 
330     for (i = 0; i < total_queues; i++) {
331         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
332 
333         if (r < 0) {
334             goto err_start;
335         }
336     }
337 
338     return 0;
339 
340 err_start:
341     while (--i >= 0) {
342         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
343     }
344     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
345     if (e < 0) {
346         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
347         fflush(stderr);
348     }
349 err_endian:
350     vhost_net_set_vnet_endian(dev, ncs[0].peer, false);
351 err:
352     return r;
353 }
354 
355 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
356                     int total_queues)
357 {
358     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
359     VirtioBusState *vbus = VIRTIO_BUS(qbus);
360     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
361     int i, r;
362 
363     for (i = 0; i < total_queues; i++) {
364         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
365     }
366 
367     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
368     if (r < 0) {
369         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
370         fflush(stderr);
371     }
372     assert(r >= 0);
373 
374     assert(vhost_net_set_vnet_endian(dev, ncs[0].peer, false) >= 0);
375 }
376 
377 void vhost_net_cleanup(struct vhost_net *net)
378 {
379     vhost_dev_cleanup(&net->dev);
380     g_free(net);
381 }
382 
383 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
384 {
385     return vhost_virtqueue_pending(&net->dev, idx);
386 }
387 
388 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
389                               int idx, bool mask)
390 {
391     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
392 }
393 
394 VHostNetState *get_vhost_net(NetClientState *nc)
395 {
396     VHostNetState *vhost_net = 0;
397 
398     if (!nc) {
399         return 0;
400     }
401 
402     switch (nc->info->type) {
403     case NET_CLIENT_OPTIONS_KIND_TAP:
404         vhost_net = tap_get_vhost_net(nc);
405         break;
406     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
407         vhost_net = vhost_user_get_vhost_net(nc);
408         break;
409     default:
410         break;
411     }
412 
413     return vhost_net;
414 }
415 #else
416 struct vhost_net *vhost_net_init(VhostNetOptions *options)
417 {
418     error_report("vhost-net support is not compiled in");
419     return NULL;
420 }
421 
422 int vhost_net_start(VirtIODevice *dev,
423                     NetClientState *ncs,
424                     int total_queues)
425 {
426     return -ENOSYS;
427 }
428 void vhost_net_stop(VirtIODevice *dev,
429                     NetClientState *ncs,
430                     int total_queues)
431 {
432 }
433 
434 void vhost_net_cleanup(struct vhost_net *net)
435 {
436 }
437 
438 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
439 {
440     return features;
441 }
442 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
443 {
444 }
445 
446 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
447 {
448     return false;
449 }
450 
451 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
452                               int idx, bool mask)
453 {
454 }
455 
456 VHostNetState *get_vhost_net(NetClientState *nc)
457 {
458     return 0;
459 }
460 #endif
461