xref: /openbmc/qemu/hw/net/vhost_net.c (revision e2051e9e004649b53af4db34f78c689fb44e075b) 
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 uint64_t vhost_net_get_max_queues(VHostNetState *net)
126 {
127     return net->dev.max_queues;
128 }
129 
130 static int vhost_net_get_fd(NetClientState *backend)
131 {
132     switch (backend->info->type) {
133     case NET_CLIENT_OPTIONS_KIND_TAP:
134         return tap_get_fd(backend);
135     default:
136         fprintf(stderr, "vhost-net requires tap backend\n");
137         return -EBADFD;
138     }
139 }
140 
141 struct vhost_net *vhost_net_init(VhostNetOptions *options)
142 {
143     int r;
144     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
145     struct vhost_net *net = g_malloc(sizeof *net);
146 
147     if (!options->net_backend) {
148         fprintf(stderr, "vhost-net requires net backend to be setup\n");
149         goto fail;
150     }
151 
152     net->dev.max_queues = 1;
153 
154     if (backend_kernel) {
155         r = vhost_net_get_fd(options->net_backend);
156         if (r < 0) {
157             goto fail;
158         }
159         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
160             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
161         net->backend = r;
162         net->dev.protocol_features = 0;
163     } else {
164         net->dev.backend_features = 0;
165         net->dev.protocol_features = 0;
166         net->backend = -1;
167     }
168     net->nc = options->net_backend;
169 
170     net->dev.nvqs = 2;
171     net->dev.vqs = net->vqs;
172 
173     r = vhost_dev_init(&net->dev, options->opaque,
174                        options->backend_type);
175     if (r < 0) {
176         goto fail;
177     }
178     if (backend_kernel) {
179         if (!qemu_has_vnet_hdr_len(options->net_backend,
180                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
181             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
182         }
183         if (~net->dev.features & net->dev.backend_features) {
184             fprintf(stderr, "vhost lacks feature mask %" PRIu64
185                    " for backend\n",
186                    (uint64_t)(~net->dev.features & net->dev.backend_features));
187             vhost_dev_cleanup(&net->dev);
188             goto fail;
189         }
190     }
191     /* Set sane init value. Override when guest acks. */
192     vhost_net_ack_features(net, 0);
193     return net;
194 fail:
195     g_free(net);
196     return NULL;
197 }
198 
199 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
200 {
201     net->dev.vq_index = vq_index;
202 }
203 
204 static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
205                                      bool set)
206 {
207     int r = 0;
208 
209     if (virtio_vdev_has_feature(dev, VIRTIO_F_VERSION_1) ||
210         (virtio_legacy_is_cross_endian(dev) && !virtio_is_big_endian(dev))) {
211         r = qemu_set_vnet_le(peer, set);
212         if (r) {
213             error_report("backend does not support LE vnet headers");
214         }
215     } else if (virtio_legacy_is_cross_endian(dev)) {
216         r = qemu_set_vnet_be(peer, set);
217         if (r) {
218             error_report("backend does not support BE vnet headers");
219         }
220     }
221 
222     return r;
223 }
224 
225 static int vhost_net_start_one(struct vhost_net *net,
226                                VirtIODevice *dev)
227 {
228     struct vhost_vring_file file = { };
229     int r;
230 
231     net->dev.nvqs = 2;
232     net->dev.vqs = net->vqs;
233 
234     r = vhost_dev_enable_notifiers(&net->dev, dev);
235     if (r < 0) {
236         goto fail_notifiers;
237     }
238 
239     r = vhost_dev_start(&net->dev, dev);
240     if (r < 0) {
241         goto fail_start;
242     }
243 
244     if (net->nc->info->poll) {
245         net->nc->info->poll(net->nc, false);
246     }
247 
248     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
249         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
250         file.fd = net->backend;
251         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
252             const VhostOps *vhost_ops = net->dev.vhost_ops;
253             r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
254                                       &file);
255             if (r < 0) {
256                 r = -errno;
257                 goto fail;
258             }
259         }
260     }
261     return 0;
262 fail:
263     file.fd = -1;
264     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
265         while (file.index-- > 0) {
266             const VhostOps *vhost_ops = net->dev.vhost_ops;
267             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
268                                           &file);
269             assert(r >= 0);
270         }
271     }
272     if (net->nc->info->poll) {
273         net->nc->info->poll(net->nc, true);
274     }
275     vhost_dev_stop(&net->dev, dev);
276 fail_start:
277     vhost_dev_disable_notifiers(&net->dev, dev);
278 fail_notifiers:
279     return r;
280 }
281 
282 static void vhost_net_stop_one(struct vhost_net *net,
283                                VirtIODevice *dev)
284 {
285     struct vhost_vring_file file = { .fd = -1 };
286 
287     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
288         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
289             const VhostOps *vhost_ops = net->dev.vhost_ops;
290             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
291                                           &file);
292             assert(r >= 0);
293         }
294     } else if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
295         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
296             const VhostOps *vhost_ops = net->dev.vhost_ops;
297             int r = vhost_ops->vhost_call(&net->dev, VHOST_RESET_DEVICE,
298                                           NULL);
299             assert(r >= 0);
300         }
301     }
302     if (net->nc->info->poll) {
303         net->nc->info->poll(net->nc, true);
304     }
305     vhost_dev_stop(&net->dev, dev);
306     vhost_dev_disable_notifiers(&net->dev, dev);
307 }
308 
309 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
310                     int total_queues)
311 {
312     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
313     VirtioBusState *vbus = VIRTIO_BUS(qbus);
314     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
315     int r, e, i;
316 
317     if (!k->set_guest_notifiers) {
318         error_report("binding does not support guest notifiers");
319         r = -ENOSYS;
320         goto err;
321     }
322 
323     r = vhost_net_set_vnet_endian(dev, ncs[0].peer, true);
324     if (r < 0) {
325         goto err;
326     }
327 
328     for (i = 0; i < total_queues; i++) {
329         vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
330     }
331 
332     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
333     if (r < 0) {
334         error_report("Error binding guest notifier: %d", -r);
335         goto err_endian;
336     }
337 
338     for (i = 0; i < total_queues; i++) {
339         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
340 
341         if (r < 0) {
342             goto err_start;
343         }
344     }
345 
346     return 0;
347 
348 err_start:
349     while (--i >= 0) {
350         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
351     }
352     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
353     if (e < 0) {
354         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
355         fflush(stderr);
356     }
357 err_endian:
358     vhost_net_set_vnet_endian(dev, ncs[0].peer, false);
359 err:
360     return r;
361 }
362 
363 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
364                     int total_queues)
365 {
366     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
367     VirtioBusState *vbus = VIRTIO_BUS(qbus);
368     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
369     int i, r;
370 
371     for (i = 0; i < total_queues; i++) {
372         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
373     }
374 
375     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
376     if (r < 0) {
377         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
378         fflush(stderr);
379     }
380     assert(r >= 0);
381 
382     assert(vhost_net_set_vnet_endian(dev, ncs[0].peer, false) >= 0);
383 }
384 
385 void vhost_net_cleanup(struct vhost_net *net)
386 {
387     vhost_dev_cleanup(&net->dev);
388     g_free(net);
389 }
390 
391 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
392 {
393     return vhost_virtqueue_pending(&net->dev, idx);
394 }
395 
396 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
397                               int idx, bool mask)
398 {
399     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
400 }
401 
402 VHostNetState *get_vhost_net(NetClientState *nc)
403 {
404     VHostNetState *vhost_net = 0;
405 
406     if (!nc) {
407         return 0;
408     }
409 
410     switch (nc->info->type) {
411     case NET_CLIENT_OPTIONS_KIND_TAP:
412         vhost_net = tap_get_vhost_net(nc);
413         break;
414     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
415         vhost_net = vhost_user_get_vhost_net(nc);
416         break;
417     default:
418         break;
419     }
420 
421     return vhost_net;
422 }
423 #else
424 uint64_t vhost_net_get_max_queues(VHostNetState *net)
425 {
426     return 1;
427 }
428 
429 struct vhost_net *vhost_net_init(VhostNetOptions *options)
430 {
431     error_report("vhost-net support is not compiled in");
432     return NULL;
433 }
434 
435 int vhost_net_start(VirtIODevice *dev,
436                     NetClientState *ncs,
437                     int total_queues)
438 {
439     return -ENOSYS;
440 }
441 void vhost_net_stop(VirtIODevice *dev,
442                     NetClientState *ncs,
443                     int total_queues)
444 {
445 }
446 
447 void vhost_net_cleanup(struct vhost_net *net)
448 {
449 }
450 
451 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
452 {
453     return features;
454 }
455 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
456 {
457 }
458 
459 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
460 {
461     return false;
462 }
463 
464 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
465                               int idx, bool mask)
466 {
467 }
468 
469 VHostNetState *get_vhost_net(NetClientState *nc)
470 {
471     return 0;
472 }
473 #endif
474