xref: /openbmc/qemu/hw/net/vhost_net.c (revision dcb10c000cdd4d14f5ac4f07b04fb666494ef4a8) 
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         net->dev.protocol_features = 0;
156     } else {
157         net->dev.backend_features = 0;
158         net->dev.protocol_features = 0;
159         net->backend = -1;
160     }
161     net->nc = options->net_backend;
162 
163     net->dev.nvqs = 2;
164     net->dev.vqs = net->vqs;
165 
166     r = vhost_dev_init(&net->dev, options->opaque,
167                        options->backend_type);
168     if (r < 0) {
169         goto fail;
170     }
171     if (backend_kernel) {
172         if (!qemu_has_vnet_hdr_len(options->net_backend,
173                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
174             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
175         }
176         if (~net->dev.features & net->dev.backend_features) {
177             fprintf(stderr, "vhost lacks feature mask %" PRIu64
178                    " for backend\n",
179                    (uint64_t)(~net->dev.features & net->dev.backend_features));
180             vhost_dev_cleanup(&net->dev);
181             goto fail;
182         }
183     }
184     /* Set sane init value. Override when guest acks. */
185     vhost_net_ack_features(net, 0);
186     return net;
187 fail:
188     g_free(net);
189     return NULL;
190 }
191 
192 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
193 {
194     net->dev.vq_index = vq_index;
195 }
196 
197 static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
198                                      bool set)
199 {
200     int r = 0;
201 
202     if (virtio_vdev_has_feature(dev, VIRTIO_F_VERSION_1) ||
203         (virtio_legacy_is_cross_endian(dev) && !virtio_is_big_endian(dev))) {
204         r = qemu_set_vnet_le(peer, set);
205         if (r) {
206             error_report("backend does not support LE vnet headers");
207         }
208     } else if (virtio_legacy_is_cross_endian(dev)) {
209         r = qemu_set_vnet_be(peer, set);
210         if (r) {
211             error_report("backend does not support BE vnet headers");
212         }
213     }
214 
215     return r;
216 }
217 
218 static int vhost_net_start_one(struct vhost_net *net,
219                                VirtIODevice *dev)
220 {
221     struct vhost_vring_file file = { };
222     int r;
223 
224     net->dev.nvqs = 2;
225     net->dev.vqs = net->vqs;
226 
227     r = vhost_dev_enable_notifiers(&net->dev, dev);
228     if (r < 0) {
229         goto fail_notifiers;
230     }
231 
232     r = vhost_dev_start(&net->dev, dev);
233     if (r < 0) {
234         goto fail_start;
235     }
236 
237     if (net->nc->info->poll) {
238         net->nc->info->poll(net->nc, false);
239     }
240 
241     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_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             const VhostOps *vhost_ops = net->dev.vhost_ops;
246             r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
247                                       &file);
248             if (r < 0) {
249                 r = -errno;
250                 goto fail;
251             }
252         }
253     }
254     return 0;
255 fail:
256     file.fd = -1;
257     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
258         while (file.index-- > 0) {
259             const VhostOps *vhost_ops = net->dev.vhost_ops;
260             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
261                                           &file);
262             assert(r >= 0);
263         }
264     }
265     if (net->nc->info->poll) {
266         net->nc->info->poll(net->nc, true);
267     }
268     vhost_dev_stop(&net->dev, dev);
269 fail_start:
270     vhost_dev_disable_notifiers(&net->dev, dev);
271 fail_notifiers:
272     return r;
273 }
274 
275 static void vhost_net_stop_one(struct vhost_net *net,
276                                VirtIODevice *dev)
277 {
278     struct vhost_vring_file file = { .fd = -1 };
279 
280     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
281         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
282             const VhostOps *vhost_ops = net->dev.vhost_ops;
283             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
284                                           &file);
285             assert(r >= 0);
286         }
287     } else if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
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_RESET_OWNER,
291                                           NULL);
292             assert(r >= 0);
293         }
294     }
295     if (net->nc->info->poll) {
296         net->nc->info->poll(net->nc, true);
297     }
298     vhost_dev_stop(&net->dev, dev);
299     vhost_dev_disable_notifiers(&net->dev, dev);
300 }
301 
302 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
303                     int total_queues)
304 {
305     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
306     VirtioBusState *vbus = VIRTIO_BUS(qbus);
307     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
308     int r, e, i;
309 
310     if (!k->set_guest_notifiers) {
311         error_report("binding does not support guest notifiers");
312         r = -ENOSYS;
313         goto err;
314     }
315 
316     r = vhost_net_set_vnet_endian(dev, ncs[0].peer, true);
317     if (r < 0) {
318         goto err;
319     }
320 
321     for (i = 0; i < total_queues; i++) {
322         vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
323     }
324 
325     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
326     if (r < 0) {
327         error_report("Error binding guest notifier: %d", -r);
328         goto err_endian;
329     }
330 
331     for (i = 0; i < total_queues; i++) {
332         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
333 
334         if (r < 0) {
335             goto err_start;
336         }
337     }
338 
339     return 0;
340 
341 err_start:
342     while (--i >= 0) {
343         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
344     }
345     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
346     if (e < 0) {
347         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
348         fflush(stderr);
349     }
350 err_endian:
351     vhost_net_set_vnet_endian(dev, ncs[0].peer, false);
352 err:
353     return r;
354 }
355 
356 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
357                     int total_queues)
358 {
359     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
360     VirtioBusState *vbus = VIRTIO_BUS(qbus);
361     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
362     int i, r;
363 
364     for (i = 0; i < total_queues; i++) {
365         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
366     }
367 
368     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
369     if (r < 0) {
370         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
371         fflush(stderr);
372     }
373     assert(r >= 0);
374 
375     assert(vhost_net_set_vnet_endian(dev, ncs[0].peer, false) >= 0);
376 }
377 
378 void vhost_net_cleanup(struct vhost_net *net)
379 {
380     vhost_dev_cleanup(&net->dev);
381     g_free(net);
382 }
383 
384 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
385 {
386     return vhost_virtqueue_pending(&net->dev, idx);
387 }
388 
389 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
390                               int idx, bool mask)
391 {
392     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
393 }
394 
395 VHostNetState *get_vhost_net(NetClientState *nc)
396 {
397     VHostNetState *vhost_net = 0;
398 
399     if (!nc) {
400         return 0;
401     }
402 
403     switch (nc->info->type) {
404     case NET_CLIENT_OPTIONS_KIND_TAP:
405         vhost_net = tap_get_vhost_net(nc);
406         break;
407     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
408         vhost_net = vhost_user_get_vhost_net(nc);
409         break;
410     default:
411         break;
412     }
413 
414     return vhost_net;
415 }
416 #else
417 struct vhost_net *vhost_net_init(VhostNetOptions *options)
418 {
419     error_report("vhost-net support is not compiled in");
420     return NULL;
421 }
422 
423 int vhost_net_start(VirtIODevice *dev,
424                     NetClientState *ncs,
425                     int total_queues)
426 {
427     return -ENOSYS;
428 }
429 void vhost_net_stop(VirtIODevice *dev,
430                     NetClientState *ncs,
431                     int total_queues)
432 {
433 }
434 
435 void vhost_net_cleanup(struct vhost_net *net)
436 {
437 }
438 
439 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
440 {
441     return features;
442 }
443 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
444 {
445 }
446 
447 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
448 {
449     return false;
450 }
451 
452 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
453                               int idx, bool mask)
454 {
455 }
456 
457 VHostNetState *get_vhost_net(NetClientState *nc)
458 {
459     return 0;
460 }
461 #endif
462