xref: /qemu/hw/net/vhost_net.c (revision b2a3cbb8) 
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 "qemu/osdep.h"
17 #include "net/net.h"
18 #include "net/tap.h"
19 #include "net/vhost-user.h"
20 #include "net/vhost-vdpa.h"
21 
22 #include "standard-headers/linux/vhost_types.h"
23 #include "hw/virtio/virtio-net.h"
24 #include "net/vhost_net.h"
25 #include "qapi/error.h"
26 #include "qemu/error-report.h"
27 #include "qemu/main-loop.h"
28 
29 #include <sys/socket.h>
30 #include <net/if.h>
31 #include <netinet/in.h>
32 
33 
34 #include "standard-headers/linux/virtio_ring.h"
35 #include "hw/virtio/vhost.h"
36 #include "hw/virtio/virtio-bus.h"
37 #include "linux-headers/linux/vhost.h"
38 
39 
40 /* Features supported by host kernel. */
41 static const int kernel_feature_bits[] = {
42     VIRTIO_F_NOTIFY_ON_EMPTY,
43     VIRTIO_RING_F_INDIRECT_DESC,
44     VIRTIO_RING_F_EVENT_IDX,
45     VIRTIO_NET_F_MRG_RXBUF,
46     VIRTIO_F_VERSION_1,
47     VIRTIO_NET_F_MTU,
48     VIRTIO_F_IOMMU_PLATFORM,
49     VIRTIO_F_RING_PACKED,
50     VIRTIO_F_RING_RESET,
51     VIRTIO_NET_F_HASH_REPORT,
52     VHOST_INVALID_FEATURE_BIT
53 };
54 
55 /* Features supported by others. */
56 static const int user_feature_bits[] = {
57     VIRTIO_F_NOTIFY_ON_EMPTY,
58     VIRTIO_RING_F_INDIRECT_DESC,
59     VIRTIO_RING_F_EVENT_IDX,
60 
61     VIRTIO_F_ANY_LAYOUT,
62     VIRTIO_F_VERSION_1,
63     VIRTIO_NET_F_CSUM,
64     VIRTIO_NET_F_GUEST_CSUM,
65     VIRTIO_NET_F_GSO,
66     VIRTIO_NET_F_GUEST_TSO4,
67     VIRTIO_NET_F_GUEST_TSO6,
68     VIRTIO_NET_F_GUEST_ECN,
69     VIRTIO_NET_F_GUEST_UFO,
70     VIRTIO_NET_F_HOST_TSO4,
71     VIRTIO_NET_F_HOST_TSO6,
72     VIRTIO_NET_F_HOST_ECN,
73     VIRTIO_NET_F_HOST_UFO,
74     VIRTIO_NET_F_MRG_RXBUF,
75     VIRTIO_NET_F_MTU,
76     VIRTIO_F_IOMMU_PLATFORM,
77     VIRTIO_F_RING_PACKED,
78     VIRTIO_NET_F_RSS,
79     VIRTIO_NET_F_HASH_REPORT,
80 
81     /* This bit implies RARP isn't sent by QEMU out of band */
82     VIRTIO_NET_F_GUEST_ANNOUNCE,
83 
84     VIRTIO_NET_F_MQ,
85 
86     VHOST_INVALID_FEATURE_BIT
87 };
88 
89 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
90 {
91     const int *feature_bits = 0;
92 
93     switch (net->nc->info->type) {
94     case NET_CLIENT_DRIVER_TAP:
95         feature_bits = kernel_feature_bits;
96         break;
97     case NET_CLIENT_DRIVER_VHOST_USER:
98         feature_bits = user_feature_bits;
99         break;
100 #ifdef CONFIG_VHOST_NET_VDPA
101     case NET_CLIENT_DRIVER_VHOST_VDPA:
102         feature_bits = vdpa_feature_bits;
103         break;
104 #endif
105     default:
106         error_report("Feature bits not defined for this type: %d",
107                 net->nc->info->type);
108         break;
109     }
110 
111     return feature_bits;
112 }
113 
114 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
115 {
116     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
117             features);
118 }
119 int vhost_net_get_config(struct vhost_net *net,  uint8_t *config,
120                          uint32_t config_len)
121 {
122     return vhost_dev_get_config(&net->dev, config, config_len, NULL);
123 }
124 int vhost_net_set_config(struct vhost_net *net, const uint8_t *data,
125                          uint32_t offset, uint32_t size, uint32_t flags)
126 {
127     return vhost_dev_set_config(&net->dev, data, offset, size, flags);
128 }
129 
130 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
131 {
132     net->dev.acked_features = net->dev.backend_features;
133     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
134 }
135 
136 uint64_t vhost_net_get_max_queues(VHostNetState *net)
137 {
138     return net->dev.max_queues;
139 }
140 
141 uint64_t vhost_net_get_acked_features(VHostNetState *net)
142 {
143     return net->dev.acked_features;
144 }
145 
146 static int vhost_net_get_fd(NetClientState *backend)
147 {
148     switch (backend->info->type) {
149     case NET_CLIENT_DRIVER_TAP:
150         return tap_get_fd(backend);
151     default:
152         fprintf(stderr, "vhost-net requires tap backend\n");
153         return -ENOSYS;
154     }
155 }
156 
157 struct vhost_net *vhost_net_init(VhostNetOptions *options)
158 {
159     int r;
160     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
161     struct vhost_net *net = g_new0(struct vhost_net, 1);
162     uint64_t features = 0;
163     Error *local_err = NULL;
164 
165     if (!options->net_backend) {
166         fprintf(stderr, "vhost-net requires net backend to be setup\n");
167         goto fail;
168     }
169     net->nc = options->net_backend;
170     net->dev.nvqs = options->nvqs;
171 
172     net->dev.max_queues = 1;
173     net->dev.vqs = net->vqs;
174 
175     if (backend_kernel) {
176         r = vhost_net_get_fd(options->net_backend);
177         if (r < 0) {
178             goto fail;
179         }
180         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
181             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
182         net->backend = r;
183         net->dev.protocol_features = 0;
184     } else {
185         net->dev.backend_features = 0;
186         net->dev.protocol_features = 0;
187         net->backend = -1;
188 
189         /* vhost-user needs vq_index to initiate a specific queue pair */
190         net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
191     }
192 
193     r = vhost_dev_init(&net->dev, options->opaque,
194                        options->backend_type, options->busyloop_timeout,
195                        &local_err);
196     if (r < 0) {
197         error_report_err(local_err);
198         goto fail;
199     }
200     if (backend_kernel) {
201         if (!qemu_has_vnet_hdr_len(options->net_backend,
202                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
203             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
204         }
205         if (~net->dev.features & net->dev.backend_features) {
206             fprintf(stderr, "vhost lacks feature mask 0x%" PRIx64
207                    " for backend\n",
208                    (uint64_t)(~net->dev.features & net->dev.backend_features));
209             goto fail;
210         }
211     }
212 
213     /* Set sane init value. Override when guest acks. */
214 #ifdef CONFIG_VHOST_NET_USER
215     if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
216         features = vhost_user_get_acked_features(net->nc);
217         if (~net->dev.features & features) {
218             fprintf(stderr, "vhost lacks feature mask 0x%" PRIx64
219                     " for backend\n",
220                     (uint64_t)(~net->dev.features & features));
221             goto fail;
222         }
223     }
224 #endif
225 
226     vhost_net_ack_features(net, features);
227 
228     return net;
229 
230 fail:
231     vhost_dev_cleanup(&net->dev);
232     g_free(net);
233     return NULL;
234 }
235 
236 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index,
237                                    int vq_index_end)
238 {
239     net->dev.vq_index = vq_index;
240     net->dev.vq_index_end = vq_index_end;
241 }
242 
243 static int vhost_net_start_one(struct vhost_net *net,
244                                VirtIODevice *dev)
245 {
246     struct vhost_vring_file file = { };
247     int r;
248 
249     if (net->nc->info->start) {
250         r = net->nc->info->start(net->nc);
251         if (r < 0) {
252             return r;
253         }
254     }
255 
256     r = vhost_dev_enable_notifiers(&net->dev, dev);
257     if (r < 0) {
258         goto fail_notifiers;
259     }
260 
261     r = vhost_dev_start(&net->dev, dev);
262     if (r < 0) {
263         goto fail_start;
264     }
265 
266     if (net->nc->info->poll) {
267         net->nc->info->poll(net->nc, false);
268     }
269 
270     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
271         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
272         file.fd = net->backend;
273         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
274             if (!virtio_queue_enabled(dev, net->dev.vq_index +
275                                       file.index)) {
276                 /* Queue might not be ready for start */
277                 continue;
278             }
279             r = vhost_net_set_backend(&net->dev, &file);
280             if (r < 0) {
281                 r = -errno;
282                 goto fail;
283             }
284         }
285     }
286 
287     if (net->nc->info->load) {
288         r = net->nc->info->load(net->nc);
289         if (r < 0) {
290             goto fail;
291         }
292     }
293     return 0;
294 fail:
295     file.fd = -1;
296     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
297         while (file.index-- > 0) {
298             if (!virtio_queue_enabled(dev, net->dev.vq_index +
299                                       file.index)) {
300                 /* Queue might not be ready for start */
301                 continue;
302             }
303             int r = vhost_net_set_backend(&net->dev, &file);
304             assert(r >= 0);
305         }
306     }
307     if (net->nc->info->poll) {
308         net->nc->info->poll(net->nc, true);
309     }
310     vhost_dev_stop(&net->dev, dev);
311 fail_start:
312     vhost_dev_disable_notifiers(&net->dev, dev);
313 fail_notifiers:
314     return r;
315 }
316 
317 static void vhost_net_stop_one(struct vhost_net *net,
318                                VirtIODevice *dev)
319 {
320     struct vhost_vring_file file = { .fd = -1 };
321 
322     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
323         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
324             int r = vhost_net_set_backend(&net->dev, &file);
325             assert(r >= 0);
326         }
327     }
328     if (net->nc->info->poll) {
329         net->nc->info->poll(net->nc, true);
330     }
331     vhost_dev_stop(&net->dev, dev);
332     if (net->nc->info->stop) {
333         net->nc->info->stop(net->nc);
334     }
335     vhost_dev_disable_notifiers(&net->dev, dev);
336 }
337 
338 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
339                     int data_queue_pairs, int cvq)
340 {
341     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
342     VirtioBusState *vbus = VIRTIO_BUS(qbus);
343     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
344     int total_notifiers = data_queue_pairs * 2 + cvq;
345     VirtIONet *n = VIRTIO_NET(dev);
346     int nvhosts = data_queue_pairs + cvq;
347     struct vhost_net *net;
348     int r, e, i, index_end = data_queue_pairs * 2;
349     NetClientState *peer;
350 
351     if (cvq) {
352         index_end += 1;
353     }
354 
355     if (!k->set_guest_notifiers) {
356         error_report("binding does not support guest notifiers");
357         return -ENOSYS;
358     }
359 
360     for (i = 0; i < nvhosts; i++) {
361 
362         if (i < data_queue_pairs) {
363             peer = qemu_get_peer(ncs, i);
364         } else { /* Control Virtqueue */
365             peer = qemu_get_peer(ncs, n->max_queue_pairs);
366         }
367 
368         net = get_vhost_net(peer);
369         vhost_net_set_vq_index(net, i * 2, index_end);
370 
371         /* Suppress the masking guest notifiers on vhost user
372          * because vhost user doesn't interrupt masking/unmasking
373          * properly.
374          */
375         if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
376             dev->use_guest_notifier_mask = false;
377         }
378      }
379 
380     r = k->set_guest_notifiers(qbus->parent, total_notifiers, true);
381     if (r < 0) {
382         error_report("Error binding guest notifier: %d", -r);
383         goto err;
384     }
385 
386     for (i = 0; i < nvhosts; i++) {
387         if (i < data_queue_pairs) {
388             peer = qemu_get_peer(ncs, i);
389         } else {
390             peer = qemu_get_peer(ncs, n->max_queue_pairs);
391         }
392 
393         if (peer->vring_enable) {
394             /* restore vring enable state */
395             r = vhost_set_vring_enable(peer, peer->vring_enable);
396 
397             if (r < 0) {
398                 goto err_start;
399             }
400         }
401 
402         r = vhost_net_start_one(get_vhost_net(peer), dev);
403         if (r < 0) {
404             goto err_start;
405         }
406     }
407 
408     return 0;
409 
410 err_start:
411     while (--i >= 0) {
412         peer = qemu_get_peer(ncs, i < data_queue_pairs ?
413                                   i : n->max_queue_pairs);
414         vhost_net_stop_one(get_vhost_net(peer), dev);
415     }
416     e = k->set_guest_notifiers(qbus->parent, total_notifiers, false);
417     if (e < 0) {
418         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
419         fflush(stderr);
420     }
421 err:
422     return r;
423 }
424 
425 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
426                     int data_queue_pairs, int cvq)
427 {
428     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
429     VirtioBusState *vbus = VIRTIO_BUS(qbus);
430     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
431     VirtIONet *n = VIRTIO_NET(dev);
432     NetClientState *peer;
433     int total_notifiers = data_queue_pairs * 2 + cvq;
434     int nvhosts = data_queue_pairs + cvq;
435     int i, r;
436 
437     for (i = 0; i < nvhosts; i++) {
438         if (i < data_queue_pairs) {
439             peer = qemu_get_peer(ncs, i);
440         } else {
441             peer = qemu_get_peer(ncs, n->max_queue_pairs);
442         }
443         vhost_net_stop_one(get_vhost_net(peer), dev);
444     }
445 
446     r = k->set_guest_notifiers(qbus->parent, total_notifiers, false);
447     if (r < 0) {
448         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
449         fflush(stderr);
450     }
451     assert(r >= 0);
452 }
453 
454 void vhost_net_cleanup(struct vhost_net *net)
455 {
456     vhost_dev_cleanup(&net->dev);
457 }
458 
459 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
460 {
461     const VhostOps *vhost_ops = net->dev.vhost_ops;
462 
463     assert(vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
464     assert(vhost_ops->vhost_migration_done);
465 
466     return vhost_ops->vhost_migration_done(&net->dev, mac_addr);
467 }
468 
469 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
470 {
471     return vhost_virtqueue_pending(&net->dev, idx);
472 }
473 
474 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
475                               int idx, bool mask)
476 {
477     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
478 }
479 
480 VHostNetState *get_vhost_net(NetClientState *nc)
481 {
482     VHostNetState *vhost_net = 0;
483 
484     if (!nc) {
485         return 0;
486     }
487 
488     switch (nc->info->type) {
489     case NET_CLIENT_DRIVER_TAP:
490         vhost_net = tap_get_vhost_net(nc);
491         break;
492 #ifdef CONFIG_VHOST_NET_USER
493     case NET_CLIENT_DRIVER_VHOST_USER:
494         vhost_net = vhost_user_get_vhost_net(nc);
495         assert(vhost_net);
496         break;
497 #endif
498 #ifdef CONFIG_VHOST_NET_VDPA
499     case NET_CLIENT_DRIVER_VHOST_VDPA:
500         vhost_net = vhost_vdpa_get_vhost_net(nc);
501         assert(vhost_net);
502         break;
503 #endif
504     default:
505         break;
506     }
507 
508     return vhost_net;
509 }
510 
511 int vhost_set_vring_enable(NetClientState *nc, int enable)
512 {
513     VHostNetState *net = get_vhost_net(nc);
514     const VhostOps *vhost_ops = net->dev.vhost_ops;
515 
516     nc->vring_enable = enable;
517 
518     if (vhost_ops && vhost_ops->vhost_set_vring_enable) {
519         return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
520     }
521 
522     return 0;
523 }
524 
525 int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
526 {
527     const VhostOps *vhost_ops = net->dev.vhost_ops;
528 
529     if (!vhost_ops->vhost_net_set_mtu) {
530         return 0;
531     }
532 
533     return vhost_ops->vhost_net_set_mtu(&net->dev, mtu);
534 }
535 
536 void vhost_net_virtqueue_reset(VirtIODevice *vdev, NetClientState *nc,
537                                int vq_index)
538 {
539     VHostNetState *net = get_vhost_net(nc->peer);
540     const VhostOps *vhost_ops = net->dev.vhost_ops;
541     struct vhost_vring_file file = { .fd = -1 };
542     int idx;
543 
544     /* should only be called after backend is connected */
545     assert(vhost_ops);
546 
547     idx = vhost_ops->vhost_get_vq_index(&net->dev, vq_index);
548 
549     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
550         file.index = idx;
551         int r = vhost_net_set_backend(&net->dev, &file);
552         assert(r >= 0);
553     }
554 
555     vhost_virtqueue_stop(&net->dev,
556                          vdev,
557                          net->dev.vqs + idx,
558                          net->dev.vq_index + idx);
559 }
560 
561 int vhost_net_virtqueue_restart(VirtIODevice *vdev, NetClientState *nc,
562                                 int vq_index)
563 {
564     VHostNetState *net = get_vhost_net(nc->peer);
565     const VhostOps *vhost_ops = net->dev.vhost_ops;
566     struct vhost_vring_file file = { };
567     int idx, r;
568 
569     if (!net->dev.started) {
570         return -EBUSY;
571     }
572 
573     /* should only be called after backend is connected */
574     assert(vhost_ops);
575 
576     idx = vhost_ops->vhost_get_vq_index(&net->dev, vq_index);
577 
578     r = vhost_virtqueue_start(&net->dev,
579                               vdev,
580                               net->dev.vqs + idx,
581                               net->dev.vq_index + idx);
582     if (r < 0) {
583         goto err_start;
584     }
585 
586     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
587         file.index = idx;
588         file.fd = net->backend;
589         r = vhost_net_set_backend(&net->dev, &file);
590         if (r < 0) {
591             r = -errno;
592             goto err_start;
593         }
594     }
595 
596     return 0;
597 
598 err_start:
599     error_report("Error when restarting the queue.");
600 
601     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
602         file.fd = VHOST_FILE_UNBIND;
603         file.index = idx;
604         int r = vhost_net_set_backend(&net->dev, &file);
605         assert(r >= 0);
606     }
607 
608     vhost_dev_stop(&net->dev, vdev);
609 
610     return r;
611 }
612