]>
Commit | Line | Data |
---|---|---|
d5970055 MT |
1 | /* |
2 | * vhost-net support | |
3 | * | |
4 | * Copyright Red Hat, Inc. 2010 | |
5 | * | |
6 | * Authors: | |
7 | * Michael S. Tsirkin <[email protected]> | |
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. | |
6b620ca3 PB |
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. | |
d5970055 MT |
14 | */ |
15 | ||
1422e32d | 16 | #include "net/net.h" |
d5970055 | 17 | #include "net/tap.h" |
03ce5744 | 18 | #include "net/vhost-user.h" |
d5970055 | 19 | |
0d09e41a PB |
20 | #include "hw/virtio/virtio-net.h" |
21 | #include "net/vhost_net.h" | |
1de7afc9 | 22 | #include "qemu/error-report.h" |
d5970055 MT |
23 | |
24 | #include "config.h" | |
25 | ||
26 | #ifdef CONFIG_VHOST_NET | |
27 | #include <linux/vhost.h> | |
d5970055 MT |
28 | #include <sys/socket.h> |
29 | #include <linux/kvm.h> | |
30 | #include <fcntl.h> | |
d5970055 MT |
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 | ||
4fbe0f32 | 38 | #include "standard-headers/linux/virtio_ring.h" |
0d09e41a | 39 | #include "hw/virtio/vhost.h" |
1c819449 | 40 | #include "hw/virtio/virtio-bus.h" |
5be7d9f1 | 41 | #include "hw/virtio/virtio-access.h" |
d5970055 MT |
42 | |
43 | struct vhost_net { | |
44 | struct vhost_dev dev; | |
45 | struct vhost_virtqueue vqs[2]; | |
46 | int backend; | |
35277d14 | 47 | NetClientState *nc; |
d5970055 MT |
48 | }; |
49 | ||
2e6d46d7 NN |
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, | |
b1506132 | 56 | VIRTIO_F_VERSION_1, |
2e6d46d7 NN |
57 | VHOST_INVALID_FEATURE_BIT |
58 | }; | |
59 | ||
5f4c01ca | 60 | /* Features supported by others. */ |
d122f1a2 | 61 | static const int user_feature_bits[] = { |
5f4c01ca NN |
62 | VIRTIO_F_NOTIFY_ON_EMPTY, |
63 | VIRTIO_RING_F_INDIRECT_DESC, | |
64 | VIRTIO_RING_F_EVENT_IDX, | |
65 | ||
66 | VIRTIO_F_ANY_LAYOUT, | |
b1506132 | 67 | VIRTIO_F_VERSION_1, |
5f4c01ca NN |
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 | ||
2e6d46d7 | 93 | static const int *vhost_net_get_feature_bits(struct vhost_net *net) |
d5970055 | 94 | { |
2e6d46d7 NN |
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; | |
5f4c01ca NN |
101 | case NET_CLIENT_OPTIONS_KIND_VHOST_USER: |
102 | feature_bits = user_feature_bits; | |
103 | break; | |
2e6d46d7 NN |
104 | default: |
105 | error_report("Feature bits not defined for this type: %d", | |
106 | net->nc->info->type); | |
107 | break; | |
ca736c8e | 108 | } |
2e6d46d7 NN |
109 | |
110 | return feature_bits; | |
111 | } | |
112 | ||
9a2ba823 | 113 | uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features) |
2e6d46d7 NN |
114 | { |
115 | return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net), | |
116 | features); | |
d5970055 MT |
117 | } |
118 | ||
9a2ba823 | 119 | void vhost_net_ack_features(struct vhost_net *net, uint64_t features) |
d5970055 | 120 | { |
b49ae913 | 121 | net->dev.acked_features = net->dev.backend_features; |
2e6d46d7 | 122 | vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features); |
d5970055 MT |
123 | } |
124 | ||
4e68f7a0 | 125 | static int vhost_net_get_fd(NetClientState *backend) |
d5970055 MT |
126 | { |
127 | switch (backend->info->type) { | |
2be64a68 | 128 | case NET_CLIENT_OPTIONS_KIND_TAP: |
d5970055 MT |
129 | return tap_get_fd(backend); |
130 | default: | |
131 | fprintf(stderr, "vhost-net requires tap backend\n"); | |
132 | return -EBADFD; | |
133 | } | |
134 | } | |
135 | ||
81647a65 | 136 | struct vhost_net *vhost_net_init(VhostNetOptions *options) |
d5970055 MT |
137 | { |
138 | int r; | |
1a1bfac9 | 139 | bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL; |
7267c094 | 140 | struct vhost_net *net = g_malloc(sizeof *net); |
81647a65 NN |
141 | |
142 | if (!options->net_backend) { | |
143 | fprintf(stderr, "vhost-net requires net backend to be setup\n"); | |
d5970055 MT |
144 | goto fail; |
145 | } | |
81647a65 | 146 | |
1a1bfac9 NN |
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) | |
9a2ba823 | 153 | ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR); |
1a1bfac9 NN |
154 | net->backend = r; |
155 | } else { | |
156 | net->dev.backend_features = 0; | |
157 | net->backend = -1; | |
d5970055 | 158 | } |
81647a65 | 159 | net->nc = options->net_backend; |
d5970055 | 160 | |
f56a1247 MT |
161 | net->dev.nvqs = 2; |
162 | net->dev.vqs = net->vqs; | |
830d70db | 163 | net->dev.vq_index = net->nc->queue_index; |
f56a1247 | 164 | |
81647a65 | 165 | r = vhost_dev_init(&net->dev, options->opaque, |
1a1bfac9 | 166 | options->backend_type, options->force); |
d5970055 MT |
167 | if (r < 0) { |
168 | goto fail; | |
169 | } | |
1a1bfac9 | 170 | if (backend_kernel) { |
d8e80ae3 DM |
171 | if (!qemu_has_vnet_hdr_len(options->net_backend, |
172 | sizeof(struct virtio_net_hdr_mrg_rxbuf))) { | |
9a2ba823 | 173 | net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF); |
d8e80ae3 | 174 | } |
1a1bfac9 NN |
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 | } | |
d5970055 | 182 | } |
d5970055 MT |
183 | /* Set sane init value. Override when guest acks. */ |
184 | vhost_net_ack_features(net, 0); | |
185 | return net; | |
186 | fail: | |
7267c094 | 187 | g_free(net); |
d5970055 MT |
188 | return NULL; |
189 | } | |
190 | ||
5430a28f MT |
191 | bool vhost_net_query(VHostNetState *net, VirtIODevice *dev) |
192 | { | |
193 | return vhost_dev_query(&net->dev, dev); | |
194 | } | |
195 | ||
cd7d1d26 JW |
196 | static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index) |
197 | { | |
198 | net->dev.vq_index = vq_index; | |
199 | } | |
200 | ||
5be7d9f1 GK |
201 | static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer, |
202 | bool set) | |
203 | { | |
204 | int r = 0; | |
205 | ||
206 | if (virtio_has_feature(dev, VIRTIO_F_VERSION_1) || | |
207 | (virtio_legacy_is_cross_endian(dev) && !virtio_is_big_endian(dev))) { | |
208 | r = qemu_set_vnet_le(peer, set); | |
209 | if (r) { | |
210 | error_report("backend does not support LE vnet headers"); | |
211 | } | |
212 | } else if (virtio_legacy_is_cross_endian(dev)) { | |
213 | r = qemu_set_vnet_be(peer, set); | |
214 | if (r) { | |
215 | error_report("backend does not support BE vnet headers"); | |
216 | } | |
217 | } | |
218 | ||
219 | return r; | |
220 | } | |
221 | ||
a9f98bb5 | 222 | static int vhost_net_start_one(struct vhost_net *net, |
cd7d1d26 | 223 | VirtIODevice *dev) |
d5970055 MT |
224 | { |
225 | struct vhost_vring_file file = { }; | |
226 | int r; | |
b0b3db79 | 227 | |
a9f98bb5 JW |
228 | net->dev.nvqs = 2; |
229 | net->dev.vqs = net->vqs; | |
a9f98bb5 | 230 | |
b0b3db79 MT |
231 | r = vhost_dev_enable_notifiers(&net->dev, dev); |
232 | if (r < 0) { | |
233 | goto fail_notifiers; | |
234 | } | |
d5970055 | 235 | |
d5970055 MT |
236 | r = vhost_dev_start(&net->dev, dev); |
237 | if (r < 0) { | |
b0b3db79 | 238 | goto fail_start; |
d5970055 MT |
239 | } |
240 | ||
212d69f2 NN |
241 | if (net->nc->info->poll) { |
242 | net->nc->info->poll(net->nc, false); | |
243 | } | |
244 | ||
1a1bfac9 NN |
245 | if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) { |
246 | qemu_set_fd_handler(net->backend, NULL, NULL, NULL); | |
247 | file.fd = net->backend; | |
248 | for (file.index = 0; file.index < net->dev.nvqs; ++file.index) { | |
249 | const VhostOps *vhost_ops = net->dev.vhost_ops; | |
250 | r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND, | |
251 | &file); | |
252 | if (r < 0) { | |
253 | r = -errno; | |
254 | goto fail; | |
255 | } | |
d5970055 MT |
256 | } |
257 | } | |
258 | return 0; | |
259 | fail: | |
260 | file.fd = -1; | |
1a1bfac9 NN |
261 | if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) { |
262 | while (file.index-- > 0) { | |
263 | const VhostOps *vhost_ops = net->dev.vhost_ops; | |
264 | int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND, | |
265 | &file); | |
266 | assert(r >= 0); | |
267 | } | |
d5970055 | 268 | } |
212d69f2 NN |
269 | if (net->nc->info->poll) { |
270 | net->nc->info->poll(net->nc, true); | |
271 | } | |
d5970055 | 272 | vhost_dev_stop(&net->dev, dev); |
b0b3db79 MT |
273 | fail_start: |
274 | vhost_dev_disable_notifiers(&net->dev, dev); | |
275 | fail_notifiers: | |
d5970055 MT |
276 | return r; |
277 | } | |
278 | ||
a9f98bb5 JW |
279 | static void vhost_net_stop_one(struct vhost_net *net, |
280 | VirtIODevice *dev) | |
d5970055 MT |
281 | { |
282 | struct vhost_vring_file file = { .fd = -1 }; | |
283 | ||
1a1bfac9 NN |
284 | if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) { |
285 | for (file.index = 0; file.index < net->dev.nvqs; ++file.index) { | |
286 | const VhostOps *vhost_ops = net->dev.vhost_ops; | |
287 | int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND, | |
288 | &file); | |
289 | assert(r >= 0); | |
290 | } | |
294ce717 LG |
291 | } else if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) { |
292 | for (file.index = 0; file.index < net->dev.nvqs; ++file.index) { | |
293 | const VhostOps *vhost_ops = net->dev.vhost_ops; | |
294 | int r = vhost_ops->vhost_call(&net->dev, VHOST_RESET_OWNER, | |
830d70db | 295 | &file); |
294ce717 LG |
296 | assert(r >= 0); |
297 | } | |
d5970055 | 298 | } |
212d69f2 NN |
299 | if (net->nc->info->poll) { |
300 | net->nc->info->poll(net->nc, true); | |
301 | } | |
d5970055 | 302 | vhost_dev_stop(&net->dev, dev); |
b0b3db79 | 303 | vhost_dev_disable_notifiers(&net->dev, dev); |
d5970055 MT |
304 | } |
305 | ||
a9f98bb5 JW |
306 | int vhost_net_start(VirtIODevice *dev, NetClientState *ncs, |
307 | int total_queues) | |
308 | { | |
1c819449 FK |
309 | BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev))); |
310 | VirtioBusState *vbus = VIRTIO_BUS(qbus); | |
311 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus); | |
cd7d1d26 | 312 | int r, e, i; |
a9f98bb5 | 313 | |
1c819449 | 314 | if (!k->set_guest_notifiers) { |
312fd5f2 | 315 | error_report("binding does not support guest notifiers"); |
a9f98bb5 JW |
316 | r = -ENOSYS; |
317 | goto err; | |
318 | } | |
319 | ||
5be7d9f1 GK |
320 | r = vhost_net_set_vnet_endian(dev, ncs[0].peer, true); |
321 | if (r < 0) { | |
322 | goto err; | |
323 | } | |
324 | ||
a9f98bb5 | 325 | for (i = 0; i < total_queues; i++) { |
cd7d1d26 | 326 | vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2); |
a9f98bb5 JW |
327 | } |
328 | ||
1c819449 | 329 | r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true); |
a9f98bb5 | 330 | if (r < 0) { |
312fd5f2 | 331 | error_report("Error binding guest notifier: %d", -r); |
5be7d9f1 | 332 | goto err_endian; |
a9f98bb5 JW |
333 | } |
334 | ||
cd7d1d26 JW |
335 | for (i = 0; i < total_queues; i++) { |
336 | r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev); | |
337 | ||
338 | if (r < 0) { | |
339 | goto err_start; | |
340 | } | |
341 | } | |
342 | ||
a9f98bb5 JW |
343 | return 0; |
344 | ||
cd7d1d26 | 345 | err_start: |
a9f98bb5 | 346 | while (--i >= 0) { |
ed8b4afe | 347 | vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev); |
a9f98bb5 | 348 | } |
cd7d1d26 JW |
349 | e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false); |
350 | if (e < 0) { | |
351 | fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e); | |
352 | fflush(stderr); | |
353 | } | |
5be7d9f1 GK |
354 | err_endian: |
355 | vhost_net_set_vnet_endian(dev, ncs[0].peer, false); | |
cd7d1d26 | 356 | err: |
a9f98bb5 JW |
357 | return r; |
358 | } | |
359 | ||
360 | void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs, | |
361 | int total_queues) | |
362 | { | |
1c819449 FK |
363 | BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev))); |
364 | VirtioBusState *vbus = VIRTIO_BUS(qbus); | |
365 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus); | |
a9f98bb5 JW |
366 | int i, r; |
367 | ||
cd7d1d26 JW |
368 | for (i = 0; i < total_queues; i++) { |
369 | vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev); | |
370 | } | |
371 | ||
1c819449 | 372 | r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false); |
a9f98bb5 JW |
373 | if (r < 0) { |
374 | fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r); | |
375 | fflush(stderr); | |
376 | } | |
377 | assert(r >= 0); | |
5be7d9f1 GK |
378 | |
379 | assert(vhost_net_set_vnet_endian(dev, ncs[0].peer, false) >= 0); | |
a9f98bb5 JW |
380 | } |
381 | ||
d5970055 MT |
382 | void vhost_net_cleanup(struct vhost_net *net) |
383 | { | |
384 | vhost_dev_cleanup(&net->dev); | |
7267c094 | 385 | g_free(net); |
d5970055 | 386 | } |
f56a1247 MT |
387 | |
388 | bool vhost_net_virtqueue_pending(VHostNetState *net, int idx) | |
389 | { | |
390 | return vhost_virtqueue_pending(&net->dev, idx); | |
391 | } | |
392 | ||
393 | void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev, | |
394 | int idx, bool mask) | |
395 | { | |
396 | vhost_virtqueue_mask(&net->dev, dev, idx, mask); | |
397 | } | |
ed8b4afe NN |
398 | |
399 | VHostNetState *get_vhost_net(NetClientState *nc) | |
400 | { | |
401 | VHostNetState *vhost_net = 0; | |
402 | ||
403 | if (!nc) { | |
404 | return 0; | |
405 | } | |
406 | ||
407 | switch (nc->info->type) { | |
408 | case NET_CLIENT_OPTIONS_KIND_TAP: | |
409 | vhost_net = tap_get_vhost_net(nc); | |
410 | break; | |
03ce5744 NN |
411 | case NET_CLIENT_OPTIONS_KIND_VHOST_USER: |
412 | vhost_net = vhost_user_get_vhost_net(nc); | |
413 | break; | |
ed8b4afe NN |
414 | default: |
415 | break; | |
416 | } | |
417 | ||
418 | return vhost_net; | |
419 | } | |
d5970055 | 420 | #else |
81647a65 | 421 | struct vhost_net *vhost_net_init(VhostNetOptions *options) |
5430a28f | 422 | { |
35f75462 | 423 | error_report("vhost-net support is not compiled in"); |
5430a28f MT |
424 | return NULL; |
425 | } | |
426 | ||
427 | bool vhost_net_query(VHostNetState *net, VirtIODevice *dev) | |
d5970055 | 428 | { |
5430a28f | 429 | return false; |
d5970055 MT |
430 | } |
431 | ||
a9f98bb5 JW |
432 | int vhost_net_start(VirtIODevice *dev, |
433 | NetClientState *ncs, | |
434 | int total_queues) | |
d5970055 | 435 | { |
5430a28f | 436 | return -ENOSYS; |
d5970055 | 437 | } |
a9f98bb5 JW |
438 | void vhost_net_stop(VirtIODevice *dev, |
439 | NetClientState *ncs, | |
440 | int total_queues) | |
d5970055 MT |
441 | { |
442 | } | |
443 | ||
444 | void vhost_net_cleanup(struct vhost_net *net) | |
445 | { | |
446 | } | |
447 | ||
9a2ba823 | 448 | uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features) |
d5970055 | 449 | { |
5430a28f | 450 | return features; |
d5970055 | 451 | } |
9a2ba823 | 452 | void vhost_net_ack_features(struct vhost_net *net, uint64_t features) |
d5970055 MT |
453 | { |
454 | } | |
f56a1247 MT |
455 | |
456 | bool vhost_net_virtqueue_pending(VHostNetState *net, int idx) | |
457 | { | |
4dd72e04 | 458 | return false; |
f56a1247 MT |
459 | } |
460 | ||
461 | void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev, | |
462 | int idx, bool mask) | |
463 | { | |
464 | } | |
ed8b4afe NN |
465 | |
466 | VHostNetState *get_vhost_net(NetClientState *nc) | |
467 | { | |
468 | return 0; | |
469 | } | |
d5970055 | 470 | #endif |