xref: /openbmc/qemu/hw/net/vhost_net.c (revision c5e5269d8a955a0f924218911c2f4a0b34e87a21)
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 
38 
39 /* Features supported by host kernel. */
40 static const int kernel_feature_bits[] = {
41     VIRTIO_F_NOTIFY_ON_EMPTY,
42     VIRTIO_RING_F_INDIRECT_DESC,
43     VIRTIO_RING_F_EVENT_IDX,
44     VIRTIO_NET_F_MRG_RXBUF,
45     VIRTIO_F_VERSION_1,
46     VIRTIO_NET_F_MTU,
47     VIRTIO_F_IOMMU_PLATFORM,
48     VIRTIO_F_RING_PACKED,
49     VIRTIO_NET_F_HASH_REPORT,
50     VHOST_INVALID_FEATURE_BIT
51 };
52 
53 /* Features supported by others. */
54 static const int user_feature_bits[] = {
55     VIRTIO_F_NOTIFY_ON_EMPTY,
56     VIRTIO_RING_F_INDIRECT_DESC,
57     VIRTIO_RING_F_EVENT_IDX,
58 
59     VIRTIO_F_ANY_LAYOUT,
60     VIRTIO_F_VERSION_1,
61     VIRTIO_NET_F_CSUM,
62     VIRTIO_NET_F_GUEST_CSUM,
63     VIRTIO_NET_F_GSO,
64     VIRTIO_NET_F_GUEST_TSO4,
65     VIRTIO_NET_F_GUEST_TSO6,
66     VIRTIO_NET_F_GUEST_ECN,
67     VIRTIO_NET_F_GUEST_UFO,
68     VIRTIO_NET_F_HOST_TSO4,
69     VIRTIO_NET_F_HOST_TSO6,
70     VIRTIO_NET_F_HOST_ECN,
71     VIRTIO_NET_F_HOST_UFO,
72     VIRTIO_NET_F_MRG_RXBUF,
73     VIRTIO_NET_F_MTU,
74     VIRTIO_F_IOMMU_PLATFORM,
75     VIRTIO_F_RING_PACKED,
76     VIRTIO_NET_F_RSS,
77     VIRTIO_NET_F_HASH_REPORT,
78 
79     /* This bit implies RARP isn't sent by QEMU out of band */
80     VIRTIO_NET_F_GUEST_ANNOUNCE,
81 
82     VIRTIO_NET_F_MQ,
83 
84     VHOST_INVALID_FEATURE_BIT
85 };
86 
87 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
88 {
89     const int *feature_bits = 0;
90 
91     switch (net->nc->info->type) {
92     case NET_CLIENT_DRIVER_TAP:
93         feature_bits = kernel_feature_bits;
94         break;
95     case NET_CLIENT_DRIVER_VHOST_USER:
96         feature_bits = user_feature_bits;
97         break;
98 #ifdef CONFIG_VHOST_NET_VDPA
99     case NET_CLIENT_DRIVER_VHOST_VDPA:
100         feature_bits = vdpa_feature_bits;
101         break;
102 #endif
103     default:
104         error_report("Feature bits not defined for this type: %d",
105                 net->nc->info->type);
106         break;
107     }
108 
109     return feature_bits;
110 }
111 
112 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
113 {
114     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
115             features);
116 }
117 int vhost_net_get_config(struct vhost_net *net,  uint8_t *config,
118                          uint32_t config_len)
119 {
120     return vhost_dev_get_config(&net->dev, config, config_len, NULL);
121 }
122 int vhost_net_set_config(struct vhost_net *net, const uint8_t *data,
123                          uint32_t offset, uint32_t size, uint32_t flags)
124 {
125     return vhost_dev_set_config(&net->dev, data, offset, size, flags);
126 }
127 
128 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
129 {
130     net->dev.acked_features = net->dev.backend_features;
131     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
132 }
133 
134 uint64_t vhost_net_get_max_queues(VHostNetState *net)
135 {
136     return net->dev.max_queues;
137 }
138 
139 uint64_t vhost_net_get_acked_features(VHostNetState *net)
140 {
141     return net->dev.acked_features;
142 }
143 
144 static int vhost_net_get_fd(NetClientState *backend)
145 {
146     switch (backend->info->type) {
147     case NET_CLIENT_DRIVER_TAP:
148         return tap_get_fd(backend);
149     default:
150         fprintf(stderr, "vhost-net requires tap backend\n");
151         return -ENOSYS;
152     }
153 }
154 
155 struct vhost_net *vhost_net_init(VhostNetOptions *options)
156 {
157     int r;
158     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
159     struct vhost_net *net = g_new0(struct vhost_net, 1);
160     uint64_t features = 0;
161     Error *local_err = NULL;
162 
163     if (!options->net_backend) {
164         fprintf(stderr, "vhost-net requires net backend to be setup\n");
165         goto fail;
166     }
167     net->nc = options->net_backend;
168     net->dev.nvqs = options->nvqs;
169 
170     net->dev.max_queues = 1;
171     net->dev.vqs = net->vqs;
172 
173     if (backend_kernel) {
174         r = vhost_net_get_fd(options->net_backend);
175         if (r < 0) {
176             goto fail;
177         }
178         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
179             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
180         net->backend = r;
181         net->dev.protocol_features = 0;
182     } else {
183         net->dev.backend_features = 0;
184         net->dev.protocol_features = 0;
185         net->backend = -1;
186 
187         /* vhost-user needs vq_index to initiate a specific queue pair */
188         net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
189     }
190 
191     r = vhost_dev_init(&net->dev, options->opaque,
192                        options->backend_type, options->busyloop_timeout,
193                        &local_err);
194     if (r < 0) {
195         error_report_err(local_err);
196         goto fail;
197     }
198     if (backend_kernel) {
199         if (!qemu_has_vnet_hdr_len(options->net_backend,
200                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
201             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
202         }
203         if (~net->dev.features & net->dev.backend_features) {
204             fprintf(stderr, "vhost lacks feature mask 0x%" PRIx64
205                    " for backend\n",
206                    (uint64_t)(~net->dev.features & net->dev.backend_features));
207             goto fail;
208         }
209     }
210 
211     /* Set sane init value. Override when guest acks. */
212 #ifdef CONFIG_VHOST_NET_USER
213     if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
214         features = vhost_user_get_acked_features(net->nc);
215         if (~net->dev.features & features) {
216             fprintf(stderr, "vhost lacks feature mask 0x%" PRIx64
217                     " for backend\n",
218                     (uint64_t)(~net->dev.features & features));
219             goto fail;
220         }
221     }
222 #endif
223 
224     vhost_net_ack_features(net, features);
225 
226     return net;
227 
228 fail:
229     vhost_dev_cleanup(&net->dev);
230     g_free(net);
231     return NULL;
232 }
233 
234 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index,
235                                    int vq_index_end)
236 {
237     net->dev.vq_index = vq_index;
238     net->dev.vq_index_end = vq_index_end;
239 }
240 
241 static int vhost_net_start_one(struct vhost_net *net,
242                                VirtIODevice *dev)
243 {
244     struct vhost_vring_file file = { };
245     int r;
246 
247     if (net->nc->info->start) {
248         r = net->nc->info->start(net->nc);
249         if (r < 0) {
250             return r;
251         }
252     }
253 
254     r = vhost_dev_enable_notifiers(&net->dev, dev);
255     if (r < 0) {
256         goto fail_notifiers;
257     }
258 
259     r = vhost_dev_start(&net->dev, dev);
260     if (r < 0) {
261         goto fail_start;
262     }
263 
264     if (net->nc->info->poll) {
265         net->nc->info->poll(net->nc, false);
266     }
267 
268     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
269         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
270         file.fd = net->backend;
271         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
272             if (!virtio_queue_enabled(dev, net->dev.vq_index +
273                                       file.index)) {
274                 /* Queue might not be ready for start */
275                 continue;
276             }
277             r = vhost_net_set_backend(&net->dev, &file);
278             if (r < 0) {
279                 r = -errno;
280                 goto fail;
281             }
282         }
283     }
284     return 0;
285 fail:
286     file.fd = -1;
287     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
288         while (file.index-- > 0) {
289             if (!virtio_queue_enabled(dev, net->dev.vq_index +
290                                       file.index)) {
291                 /* Queue might not be ready for start */
292                 continue;
293             }
294             int r = vhost_net_set_backend(&net->dev, &file);
295             assert(r >= 0);
296         }
297     }
298     if (net->nc->info->poll) {
299         net->nc->info->poll(net->nc, true);
300     }
301     vhost_dev_stop(&net->dev, dev);
302 fail_start:
303     vhost_dev_disable_notifiers(&net->dev, dev);
304 fail_notifiers:
305     return r;
306 }
307 
308 static void vhost_net_stop_one(struct vhost_net *net,
309                                VirtIODevice *dev)
310 {
311     struct vhost_vring_file file = { .fd = -1 };
312 
313     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
314         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
315             int r = vhost_net_set_backend(&net->dev, &file);
316             assert(r >= 0);
317         }
318     }
319     if (net->nc->info->poll) {
320         net->nc->info->poll(net->nc, true);
321     }
322     vhost_dev_stop(&net->dev, dev);
323     if (net->nc->info->stop) {
324         net->nc->info->stop(net->nc);
325     }
326     vhost_dev_disable_notifiers(&net->dev, dev);
327 }
328 
329 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
330                     int data_queue_pairs, int cvq)
331 {
332     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
333     VirtioBusState *vbus = VIRTIO_BUS(qbus);
334     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
335     int total_notifiers = data_queue_pairs * 2 + cvq;
336     VirtIONet *n = VIRTIO_NET(dev);
337     int nvhosts = data_queue_pairs + cvq;
338     struct vhost_net *net;
339     int r, e, i, index_end = data_queue_pairs * 2;
340     NetClientState *peer;
341 
342     if (cvq) {
343         index_end += 1;
344     }
345 
346     if (!k->set_guest_notifiers) {
347         error_report("binding does not support guest notifiers");
348         return -ENOSYS;
349     }
350 
351     for (i = 0; i < nvhosts; i++) {
352 
353         if (i < data_queue_pairs) {
354             peer = qemu_get_peer(ncs, i);
355         } else { /* Control Virtqueue */
356             peer = qemu_get_peer(ncs, n->max_queue_pairs);
357         }
358 
359         net = get_vhost_net(peer);
360         vhost_net_set_vq_index(net, i * 2, index_end);
361 
362         /* Suppress the masking guest notifiers on vhost user
363          * because vhost user doesn't interrupt masking/unmasking
364          * properly.
365          */
366         if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
367             dev->use_guest_notifier_mask = false;
368         }
369      }
370 
371     r = k->set_guest_notifiers(qbus->parent, total_notifiers, true);
372     if (r < 0) {
373         error_report("Error binding guest notifier: %d", -r);
374         goto err;
375     }
376 
377     for (i = 0; i < nvhosts; i++) {
378         if (i < data_queue_pairs) {
379             peer = qemu_get_peer(ncs, i);
380         } else {
381             peer = qemu_get_peer(ncs, n->max_queue_pairs);
382         }
383         r = vhost_net_start_one(get_vhost_net(peer), dev);
384 
385         if (r < 0) {
386             goto err_start;
387         }
388 
389         if (peer->vring_enable) {
390             /* restore vring enable state */
391             r = vhost_set_vring_enable(peer, peer->vring_enable);
392 
393             if (r < 0) {
394                 vhost_net_stop_one(get_vhost_net(peer), dev);
395                 goto err_start;
396             }
397         }
398     }
399 
400     return 0;
401 
402 err_start:
403     while (--i >= 0) {
404         peer = qemu_get_peer(ncs, i < data_queue_pairs ?
405                                   i : n->max_queue_pairs);
406         vhost_net_stop_one(get_vhost_net(peer), dev);
407     }
408     e = k->set_guest_notifiers(qbus->parent, total_notifiers, false);
409     if (e < 0) {
410         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
411         fflush(stderr);
412     }
413 err:
414     return r;
415 }
416 
417 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
418                     int data_queue_pairs, int cvq)
419 {
420     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
421     VirtioBusState *vbus = VIRTIO_BUS(qbus);
422     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
423     VirtIONet *n = VIRTIO_NET(dev);
424     NetClientState *peer;
425     int total_notifiers = data_queue_pairs * 2 + cvq;
426     int nvhosts = data_queue_pairs + cvq;
427     int i, r;
428 
429     for (i = 0; i < nvhosts; i++) {
430         if (i < data_queue_pairs) {
431             peer = qemu_get_peer(ncs, i);
432         } else {
433             peer = qemu_get_peer(ncs, n->max_queue_pairs);
434         }
435         vhost_net_stop_one(get_vhost_net(peer), dev);
436     }
437 
438     r = k->set_guest_notifiers(qbus->parent, total_notifiers, false);
439     if (r < 0) {
440         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
441         fflush(stderr);
442     }
443     assert(r >= 0);
444 }
445 
446 void vhost_net_cleanup(struct vhost_net *net)
447 {
448     vhost_dev_cleanup(&net->dev);
449 }
450 
451 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
452 {
453     const VhostOps *vhost_ops = net->dev.vhost_ops;
454 
455     assert(vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
456     assert(vhost_ops->vhost_migration_done);
457 
458     return vhost_ops->vhost_migration_done(&net->dev, mac_addr);
459 }
460 
461 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
462 {
463     return vhost_virtqueue_pending(&net->dev, idx);
464 }
465 
466 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
467                               int idx, bool mask)
468 {
469     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
470 }
471 
472 VHostNetState *get_vhost_net(NetClientState *nc)
473 {
474     VHostNetState *vhost_net = 0;
475 
476     if (!nc) {
477         return 0;
478     }
479 
480     switch (nc->info->type) {
481     case NET_CLIENT_DRIVER_TAP:
482         vhost_net = tap_get_vhost_net(nc);
483         break;
484 #ifdef CONFIG_VHOST_NET_USER
485     case NET_CLIENT_DRIVER_VHOST_USER:
486         vhost_net = vhost_user_get_vhost_net(nc);
487         assert(vhost_net);
488         break;
489 #endif
490 #ifdef CONFIG_VHOST_NET_VDPA
491     case NET_CLIENT_DRIVER_VHOST_VDPA:
492         vhost_net = vhost_vdpa_get_vhost_net(nc);
493         assert(vhost_net);
494         break;
495 #endif
496     default:
497         break;
498     }
499 
500     return vhost_net;
501 }
502 
503 int vhost_set_vring_enable(NetClientState *nc, int enable)
504 {
505     VHostNetState *net = get_vhost_net(nc);
506     const VhostOps *vhost_ops = net->dev.vhost_ops;
507 
508     nc->vring_enable = enable;
509 
510     if (vhost_ops && vhost_ops->vhost_set_vring_enable) {
511         return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
512     }
513 
514     return 0;
515 }
516 
517 int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
518 {
519     const VhostOps *vhost_ops = net->dev.vhost_ops;
520 
521     if (!vhost_ops->vhost_net_set_mtu) {
522         return 0;
523     }
524 
525     return vhost_ops->vhost_net_set_mtu(&net->dev, mtu);
526 }
527