xref: /openbmc/qemu/hw/net/virtio-net.c (revision acb0ef58)
1 /*
2  * Virtio Network Device
3  *
4  * Copyright IBM, Corp. 2007
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.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  */
13 
14 #include "qemu/iov.h"
15 #include "hw/virtio/virtio.h"
16 #include "net/net.h"
17 #include "net/checksum.h"
18 #include "net/tap.h"
19 #include "qemu/error-report.h"
20 #include "qemu/timer.h"
21 #include "hw/virtio/virtio-net.h"
22 #include "net/vhost_net.h"
23 #include "hw/virtio/virtio-bus.h"
24 #include "qapi/qmp/qjson.h"
25 #include "monitor/monitor.h"
26 
27 #define VIRTIO_NET_VM_VERSION    11
28 
29 #define MAC_TABLE_ENTRIES    64
30 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
31 
32 /*
33  * Calculate the number of bytes up to and including the given 'field' of
34  * 'container'.
35  */
36 #define endof(container, field) \
37     (offsetof(container, field) + sizeof(((container *)0)->field))
38 
39 typedef struct VirtIOFeature {
40     uint32_t flags;
41     size_t end;
42 } VirtIOFeature;
43 
44 static VirtIOFeature feature_sizes[] = {
45     {.flags = 1 << VIRTIO_NET_F_MAC,
46      .end = endof(struct virtio_net_config, mac)},
47     {.flags = 1 << VIRTIO_NET_F_STATUS,
48      .end = endof(struct virtio_net_config, status)},
49     {.flags = 1 << VIRTIO_NET_F_MQ,
50      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
51     {}
52 };
53 
54 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
55 {
56     VirtIONet *n = qemu_get_nic_opaque(nc);
57 
58     return &n->vqs[nc->queue_index];
59 }
60 
61 static int vq2q(int queue_index)
62 {
63     return queue_index / 2;
64 }
65 
66 /* TODO
67  * - we could suppress RX interrupt if we were so inclined.
68  */
69 
70 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
71 {
72     VirtIONet *n = VIRTIO_NET(vdev);
73     struct virtio_net_config netcfg;
74 
75     stw_p(&netcfg.status, n->status);
76     stw_p(&netcfg.max_virtqueue_pairs, n->max_queues);
77     memcpy(netcfg.mac, n->mac, ETH_ALEN);
78     memcpy(config, &netcfg, n->config_size);
79 }
80 
81 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
82 {
83     VirtIONet *n = VIRTIO_NET(vdev);
84     struct virtio_net_config netcfg = {};
85 
86     memcpy(&netcfg, config, n->config_size);
87 
88     if (!(vdev->guest_features >> VIRTIO_NET_F_CTRL_MAC_ADDR & 1) &&
89         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
90         memcpy(n->mac, netcfg.mac, ETH_ALEN);
91         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
92     }
93 }
94 
95 static bool virtio_net_started(VirtIONet *n, uint8_t status)
96 {
97     VirtIODevice *vdev = VIRTIO_DEVICE(n);
98     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
99         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
100 }
101 
102 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
103 {
104     VirtIODevice *vdev = VIRTIO_DEVICE(n);
105     NetClientState *nc = qemu_get_queue(n->nic);
106     int queues = n->multiqueue ? n->max_queues : 1;
107 
108     if (!nc->peer) {
109         return;
110     }
111     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
112         return;
113     }
114 
115     if (!tap_get_vhost_net(nc->peer)) {
116         return;
117     }
118 
119     if (!!n->vhost_started ==
120         (virtio_net_started(n, status) && !nc->peer->link_down)) {
121         return;
122     }
123     if (!n->vhost_started) {
124         int r;
125         if (!vhost_net_query(tap_get_vhost_net(nc->peer), vdev)) {
126             return;
127         }
128         n->vhost_started = 1;
129         r = vhost_net_start(vdev, n->nic->ncs, queues);
130         if (r < 0) {
131             error_report("unable to start vhost net: %d: "
132                          "falling back on userspace virtio", -r);
133             n->vhost_started = 0;
134         }
135     } else {
136         vhost_net_stop(vdev, n->nic->ncs, queues);
137         n->vhost_started = 0;
138     }
139 }
140 
141 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
142 {
143     VirtIONet *n = VIRTIO_NET(vdev);
144     VirtIONetQueue *q;
145     int i;
146     uint8_t queue_status;
147 
148     virtio_net_vhost_status(n, status);
149 
150     for (i = 0; i < n->max_queues; i++) {
151         q = &n->vqs[i];
152 
153         if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
154             queue_status = 0;
155         } else {
156             queue_status = status;
157         }
158 
159         if (!q->tx_waiting) {
160             continue;
161         }
162 
163         if (virtio_net_started(n, queue_status) && !n->vhost_started) {
164             if (q->tx_timer) {
165                 timer_mod(q->tx_timer,
166                                qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
167             } else {
168                 qemu_bh_schedule(q->tx_bh);
169             }
170         } else {
171             if (q->tx_timer) {
172                 timer_del(q->tx_timer);
173             } else {
174                 qemu_bh_cancel(q->tx_bh);
175             }
176         }
177     }
178 }
179 
180 static void virtio_net_set_link_status(NetClientState *nc)
181 {
182     VirtIONet *n = qemu_get_nic_opaque(nc);
183     VirtIODevice *vdev = VIRTIO_DEVICE(n);
184     uint16_t old_status = n->status;
185 
186     if (nc->link_down)
187         n->status &= ~VIRTIO_NET_S_LINK_UP;
188     else
189         n->status |= VIRTIO_NET_S_LINK_UP;
190 
191     if (n->status != old_status)
192         virtio_notify_config(vdev);
193 
194     virtio_net_set_status(vdev, vdev->status);
195 }
196 
197 static void rxfilter_notify(NetClientState *nc)
198 {
199     QObject *event_data;
200     VirtIONet *n = qemu_get_nic_opaque(nc);
201 
202     if (nc->rxfilter_notify_enabled) {
203         gchar *path = object_get_canonical_path(OBJECT(n->qdev));
204         if (n->netclient_name) {
205             event_data = qobject_from_jsonf("{ 'name': %s, 'path': %s }",
206                                     n->netclient_name, path);
207         } else {
208             event_data = qobject_from_jsonf("{ 'path': %s }", path);
209         }
210         monitor_protocol_event(QEVENT_NIC_RX_FILTER_CHANGED, event_data);
211         qobject_decref(event_data);
212         g_free(path);
213 
214         /* disable event notification to avoid events flooding */
215         nc->rxfilter_notify_enabled = 0;
216     }
217 }
218 
219 static char *mac_strdup_printf(const uint8_t *mac)
220 {
221     return g_strdup_printf("%.2x:%.2x:%.2x:%.2x:%.2x:%.2x", mac[0],
222                             mac[1], mac[2], mac[3], mac[4], mac[5]);
223 }
224 
225 static intList *get_vlan_table(VirtIONet *n)
226 {
227     intList *list, *entry;
228     int i, j;
229 
230     list = NULL;
231     for (i = 0; i < MAX_VLAN >> 5; i++) {
232         for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
233             if (n->vlans[i] & (1U << j)) {
234                 entry = g_malloc0(sizeof(*entry));
235                 entry->value = (i << 5) + j;
236                 entry->next = list;
237                 list = entry;
238             }
239         }
240     }
241 
242     return list;
243 }
244 
245 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
246 {
247     VirtIONet *n = qemu_get_nic_opaque(nc);
248     VirtIODevice *vdev = VIRTIO_DEVICE(n);
249     RxFilterInfo *info;
250     strList *str_list, *entry;
251     int i;
252 
253     info = g_malloc0(sizeof(*info));
254     info->name = g_strdup(nc->name);
255     info->promiscuous = n->promisc;
256 
257     if (n->nouni) {
258         info->unicast = RX_STATE_NONE;
259     } else if (n->alluni) {
260         info->unicast = RX_STATE_ALL;
261     } else {
262         info->unicast = RX_STATE_NORMAL;
263     }
264 
265     if (n->nomulti) {
266         info->multicast = RX_STATE_NONE;
267     } else if (n->allmulti) {
268         info->multicast = RX_STATE_ALL;
269     } else {
270         info->multicast = RX_STATE_NORMAL;
271     }
272 
273     info->broadcast_allowed = n->nobcast;
274     info->multicast_overflow = n->mac_table.multi_overflow;
275     info->unicast_overflow = n->mac_table.uni_overflow;
276 
277     info->main_mac = mac_strdup_printf(n->mac);
278 
279     str_list = NULL;
280     for (i = 0; i < n->mac_table.first_multi; i++) {
281         entry = g_malloc0(sizeof(*entry));
282         entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
283         entry->next = str_list;
284         str_list = entry;
285     }
286     info->unicast_table = str_list;
287 
288     str_list = NULL;
289     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
290         entry = g_malloc0(sizeof(*entry));
291         entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
292         entry->next = str_list;
293         str_list = entry;
294     }
295     info->multicast_table = str_list;
296     info->vlan_table = get_vlan_table(n);
297 
298     if (!((1 << VIRTIO_NET_F_CTRL_VLAN) & vdev->guest_features)) {
299         info->vlan = RX_STATE_ALL;
300     } else if (!info->vlan_table) {
301         info->vlan = RX_STATE_NONE;
302     } else {
303         info->vlan = RX_STATE_NORMAL;
304     }
305 
306     /* enable event notification after query */
307     nc->rxfilter_notify_enabled = 1;
308 
309     return info;
310 }
311 
312 static void virtio_net_reset(VirtIODevice *vdev)
313 {
314     VirtIONet *n = VIRTIO_NET(vdev);
315 
316     /* Reset back to compatibility mode */
317     n->promisc = 1;
318     n->allmulti = 0;
319     n->alluni = 0;
320     n->nomulti = 0;
321     n->nouni = 0;
322     n->nobcast = 0;
323     /* multiqueue is disabled by default */
324     n->curr_queues = 1;
325 
326     /* Flush any MAC and VLAN filter table state */
327     n->mac_table.in_use = 0;
328     n->mac_table.first_multi = 0;
329     n->mac_table.multi_overflow = 0;
330     n->mac_table.uni_overflow = 0;
331     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
332     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
333     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
334     memset(n->vlans, 0, MAX_VLAN >> 3);
335 }
336 
337 static void peer_test_vnet_hdr(VirtIONet *n)
338 {
339     NetClientState *nc = qemu_get_queue(n->nic);
340     if (!nc->peer) {
341         return;
342     }
343 
344     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
345 }
346 
347 static int peer_has_vnet_hdr(VirtIONet *n)
348 {
349     return n->has_vnet_hdr;
350 }
351 
352 static int peer_has_ufo(VirtIONet *n)
353 {
354     if (!peer_has_vnet_hdr(n))
355         return 0;
356 
357     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
358 
359     return n->has_ufo;
360 }
361 
362 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs)
363 {
364     int i;
365     NetClientState *nc;
366 
367     n->mergeable_rx_bufs = mergeable_rx_bufs;
368 
369     n->guest_hdr_len = n->mergeable_rx_bufs ?
370         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
371 
372     for (i = 0; i < n->max_queues; i++) {
373         nc = qemu_get_subqueue(n->nic, i);
374 
375         if (peer_has_vnet_hdr(n) &&
376             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
377             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
378             n->host_hdr_len = n->guest_hdr_len;
379         }
380     }
381 }
382 
383 static int peer_attach(VirtIONet *n, int index)
384 {
385     NetClientState *nc = qemu_get_subqueue(n->nic, index);
386 
387     if (!nc->peer) {
388         return 0;
389     }
390 
391     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
392         return 0;
393     }
394 
395     return tap_enable(nc->peer);
396 }
397 
398 static int peer_detach(VirtIONet *n, int index)
399 {
400     NetClientState *nc = qemu_get_subqueue(n->nic, index);
401 
402     if (!nc->peer) {
403         return 0;
404     }
405 
406     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
407         return 0;
408     }
409 
410     return tap_disable(nc->peer);
411 }
412 
413 static void virtio_net_set_queues(VirtIONet *n)
414 {
415     int i;
416     int r;
417 
418     for (i = 0; i < n->max_queues; i++) {
419         if (i < n->curr_queues) {
420             r = peer_attach(n, i);
421             assert(!r);
422         } else {
423             r = peer_detach(n, i);
424             assert(!r);
425         }
426     }
427 }
428 
429 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
430 
431 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
432 {
433     VirtIONet *n = VIRTIO_NET(vdev);
434     NetClientState *nc = qemu_get_queue(n->nic);
435 
436     features |= (1 << VIRTIO_NET_F_MAC);
437 
438     if (!peer_has_vnet_hdr(n)) {
439         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
440         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
441         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
442         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
443 
444         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
445         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
446         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
447         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
448     }
449 
450     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
451         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
452         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
453     }
454 
455     if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
456         return features;
457     }
458     if (!tap_get_vhost_net(nc->peer)) {
459         return features;
460     }
461     return vhost_net_get_features(tap_get_vhost_net(nc->peer), features);
462 }
463 
464 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
465 {
466     uint32_t features = 0;
467 
468     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
469      * but also these: */
470     features |= (1 << VIRTIO_NET_F_MAC);
471     features |= (1 << VIRTIO_NET_F_CSUM);
472     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
473     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
474     features |= (1 << VIRTIO_NET_F_HOST_ECN);
475 
476     return features;
477 }
478 
479 static void virtio_net_apply_guest_offloads(VirtIONet *n)
480 {
481     qemu_set_offload(qemu_get_queue(n->nic)->peer,
482             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
483             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
484             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
485             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
486             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
487 }
488 
489 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
490 {
491     static const uint64_t guest_offloads_mask =
492         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
493         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
494         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
495         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
496         (1ULL << VIRTIO_NET_F_GUEST_UFO);
497 
498     return guest_offloads_mask & features;
499 }
500 
501 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
502 {
503     VirtIODevice *vdev = VIRTIO_DEVICE(n);
504     return virtio_net_guest_offloads_by_features(vdev->guest_features);
505 }
506 
507 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
508 {
509     VirtIONet *n = VIRTIO_NET(vdev);
510     int i;
511 
512     virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)));
513 
514     virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
515 
516     if (n->has_vnet_hdr) {
517         n->curr_guest_offloads =
518             virtio_net_guest_offloads_by_features(features);
519         virtio_net_apply_guest_offloads(n);
520     }
521 
522     for (i = 0;  i < n->max_queues; i++) {
523         NetClientState *nc = qemu_get_subqueue(n->nic, i);
524 
525         if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
526             continue;
527         }
528         if (!tap_get_vhost_net(nc->peer)) {
529             continue;
530         }
531         vhost_net_ack_features(tap_get_vhost_net(nc->peer), features);
532     }
533 
534     if ((1 << VIRTIO_NET_F_CTRL_VLAN) & features) {
535         memset(n->vlans, 0, MAX_VLAN >> 3);
536     } else {
537         memset(n->vlans, 0xff, MAX_VLAN >> 3);
538     }
539 }
540 
541 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
542                                      struct iovec *iov, unsigned int iov_cnt)
543 {
544     uint8_t on;
545     size_t s;
546     NetClientState *nc = qemu_get_queue(n->nic);
547 
548     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
549     if (s != sizeof(on)) {
550         return VIRTIO_NET_ERR;
551     }
552 
553     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
554         n->promisc = on;
555     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
556         n->allmulti = on;
557     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
558         n->alluni = on;
559     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
560         n->nomulti = on;
561     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
562         n->nouni = on;
563     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
564         n->nobcast = on;
565     } else {
566         return VIRTIO_NET_ERR;
567     }
568 
569     rxfilter_notify(nc);
570 
571     return VIRTIO_NET_OK;
572 }
573 
574 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
575                                      struct iovec *iov, unsigned int iov_cnt)
576 {
577     VirtIODevice *vdev = VIRTIO_DEVICE(n);
578     uint64_t offloads;
579     size_t s;
580 
581     if (!((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features)) {
582         return VIRTIO_NET_ERR;
583     }
584 
585     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
586     if (s != sizeof(offloads)) {
587         return VIRTIO_NET_ERR;
588     }
589 
590     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
591         uint64_t supported_offloads;
592 
593         if (!n->has_vnet_hdr) {
594             return VIRTIO_NET_ERR;
595         }
596 
597         supported_offloads = virtio_net_supported_guest_offloads(n);
598         if (offloads & ~supported_offloads) {
599             return VIRTIO_NET_ERR;
600         }
601 
602         n->curr_guest_offloads = offloads;
603         virtio_net_apply_guest_offloads(n);
604 
605         return VIRTIO_NET_OK;
606     } else {
607         return VIRTIO_NET_ERR;
608     }
609 }
610 
611 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
612                                  struct iovec *iov, unsigned int iov_cnt)
613 {
614     struct virtio_net_ctrl_mac mac_data;
615     size_t s;
616     NetClientState *nc = qemu_get_queue(n->nic);
617 
618     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
619         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
620             return VIRTIO_NET_ERR;
621         }
622         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
623         assert(s == sizeof(n->mac));
624         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
625         rxfilter_notify(nc);
626 
627         return VIRTIO_NET_OK;
628     }
629 
630     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
631         return VIRTIO_NET_ERR;
632     }
633 
634     int in_use = 0;
635     int first_multi = 0;
636     uint8_t uni_overflow = 0;
637     uint8_t multi_overflow = 0;
638     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
639 
640     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
641                    sizeof(mac_data.entries));
642     mac_data.entries = ldl_p(&mac_data.entries);
643     if (s != sizeof(mac_data.entries)) {
644         goto error;
645     }
646     iov_discard_front(&iov, &iov_cnt, s);
647 
648     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
649         goto error;
650     }
651 
652     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
653         s = iov_to_buf(iov, iov_cnt, 0, macs,
654                        mac_data.entries * ETH_ALEN);
655         if (s != mac_data.entries * ETH_ALEN) {
656             goto error;
657         }
658         in_use += mac_data.entries;
659     } else {
660         uni_overflow = 1;
661     }
662 
663     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
664 
665     first_multi = in_use;
666 
667     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
668                    sizeof(mac_data.entries));
669     mac_data.entries = ldl_p(&mac_data.entries);
670     if (s != sizeof(mac_data.entries)) {
671         goto error;
672     }
673 
674     iov_discard_front(&iov, &iov_cnt, s);
675 
676     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
677         goto error;
678     }
679 
680     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
681         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
682                        mac_data.entries * ETH_ALEN);
683         if (s != mac_data.entries * ETH_ALEN) {
684             goto error;
685         }
686         in_use += mac_data.entries;
687     } else {
688         multi_overflow = 1;
689     }
690 
691     n->mac_table.in_use = in_use;
692     n->mac_table.first_multi = first_multi;
693     n->mac_table.uni_overflow = uni_overflow;
694     n->mac_table.multi_overflow = multi_overflow;
695     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
696     g_free(macs);
697     rxfilter_notify(nc);
698 
699     return VIRTIO_NET_OK;
700 
701 error:
702     g_free(macs);
703     return VIRTIO_NET_ERR;
704 }
705 
706 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
707                                         struct iovec *iov, unsigned int iov_cnt)
708 {
709     uint16_t vid;
710     size_t s;
711     NetClientState *nc = qemu_get_queue(n->nic);
712 
713     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
714     vid = lduw_p(&vid);
715     if (s != sizeof(vid)) {
716         return VIRTIO_NET_ERR;
717     }
718 
719     if (vid >= MAX_VLAN)
720         return VIRTIO_NET_ERR;
721 
722     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
723         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
724     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
725         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
726     else
727         return VIRTIO_NET_ERR;
728 
729     rxfilter_notify(nc);
730 
731     return VIRTIO_NET_OK;
732 }
733 
734 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
735                                 struct iovec *iov, unsigned int iov_cnt)
736 {
737     VirtIODevice *vdev = VIRTIO_DEVICE(n);
738     struct virtio_net_ctrl_mq mq;
739     size_t s;
740     uint16_t queues;
741 
742     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
743     if (s != sizeof(mq)) {
744         return VIRTIO_NET_ERR;
745     }
746 
747     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
748         return VIRTIO_NET_ERR;
749     }
750 
751     queues = lduw_p(&mq.virtqueue_pairs);
752 
753     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
754         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
755         queues > n->max_queues ||
756         !n->multiqueue) {
757         return VIRTIO_NET_ERR;
758     }
759 
760     n->curr_queues = queues;
761     /* stop the backend before changing the number of queues to avoid handling a
762      * disabled queue */
763     virtio_net_set_status(vdev, vdev->status);
764     virtio_net_set_queues(n);
765 
766     return VIRTIO_NET_OK;
767 }
768 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
769 {
770     VirtIONet *n = VIRTIO_NET(vdev);
771     struct virtio_net_ctrl_hdr ctrl;
772     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
773     VirtQueueElement elem;
774     size_t s;
775     struct iovec *iov;
776     unsigned int iov_cnt;
777 
778     while (virtqueue_pop(vq, &elem)) {
779         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
780             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
781             error_report("virtio-net ctrl missing headers");
782             exit(1);
783         }
784 
785         iov = elem.out_sg;
786         iov_cnt = elem.out_num;
787         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
788         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
789         if (s != sizeof(ctrl)) {
790             status = VIRTIO_NET_ERR;
791         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
792             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
793         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
794             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
795         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
796             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
797         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
798             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
799         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
800             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
801         }
802 
803         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
804         assert(s == sizeof(status));
805 
806         virtqueue_push(vq, &elem, sizeof(status));
807         virtio_notify(vdev, vq);
808     }
809 }
810 
811 /* RX */
812 
813 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
814 {
815     VirtIONet *n = VIRTIO_NET(vdev);
816     int queue_index = vq2q(virtio_get_queue_index(vq));
817 
818     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
819 }
820 
821 static int virtio_net_can_receive(NetClientState *nc)
822 {
823     VirtIONet *n = qemu_get_nic_opaque(nc);
824     VirtIODevice *vdev = VIRTIO_DEVICE(n);
825     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
826 
827     if (!vdev->vm_running) {
828         return 0;
829     }
830 
831     if (nc->queue_index >= n->curr_queues) {
832         return 0;
833     }
834 
835     if (!virtio_queue_ready(q->rx_vq) ||
836         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
837         return 0;
838     }
839 
840     return 1;
841 }
842 
843 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
844 {
845     VirtIONet *n = q->n;
846     if (virtio_queue_empty(q->rx_vq) ||
847         (n->mergeable_rx_bufs &&
848          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
849         virtio_queue_set_notification(q->rx_vq, 1);
850 
851         /* To avoid a race condition where the guest has made some buffers
852          * available after the above check but before notification was
853          * enabled, check for available buffers again.
854          */
855         if (virtio_queue_empty(q->rx_vq) ||
856             (n->mergeable_rx_bufs &&
857              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
858             return 0;
859         }
860     }
861 
862     virtio_queue_set_notification(q->rx_vq, 0);
863     return 1;
864 }
865 
866 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
867  * it never finds out that the packets don't have valid checksums.  This
868  * causes dhclient to get upset.  Fedora's carried a patch for ages to
869  * fix this with Xen but it hasn't appeared in an upstream release of
870  * dhclient yet.
871  *
872  * To avoid breaking existing guests, we catch udp packets and add
873  * checksums.  This is terrible but it's better than hacking the guest
874  * kernels.
875  *
876  * N.B. if we introduce a zero-copy API, this operation is no longer free so
877  * we should provide a mechanism to disable it to avoid polluting the host
878  * cache.
879  */
880 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
881                                         uint8_t *buf, size_t size)
882 {
883     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
884         (size > 27 && size < 1500) && /* normal sized MTU */
885         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
886         (buf[23] == 17) && /* ip.protocol == UDP */
887         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
888         net_checksum_calculate(buf, size);
889         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
890     }
891 }
892 
893 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
894                            const void *buf, size_t size)
895 {
896     if (n->has_vnet_hdr) {
897         /* FIXME this cast is evil */
898         void *wbuf = (void *)buf;
899         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
900                                     size - n->host_hdr_len);
901         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
902     } else {
903         struct virtio_net_hdr hdr = {
904             .flags = 0,
905             .gso_type = VIRTIO_NET_HDR_GSO_NONE
906         };
907         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
908     }
909 }
910 
911 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
912 {
913     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
914     static const uint8_t vlan[] = {0x81, 0x00};
915     uint8_t *ptr = (uint8_t *)buf;
916     int i;
917 
918     if (n->promisc)
919         return 1;
920 
921     ptr += n->host_hdr_len;
922 
923     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
924         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
925         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
926             return 0;
927     }
928 
929     if (ptr[0] & 1) { // multicast
930         if (!memcmp(ptr, bcast, sizeof(bcast))) {
931             return !n->nobcast;
932         } else if (n->nomulti) {
933             return 0;
934         } else if (n->allmulti || n->mac_table.multi_overflow) {
935             return 1;
936         }
937 
938         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
939             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
940                 return 1;
941             }
942         }
943     } else { // unicast
944         if (n->nouni) {
945             return 0;
946         } else if (n->alluni || n->mac_table.uni_overflow) {
947             return 1;
948         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
949             return 1;
950         }
951 
952         for (i = 0; i < n->mac_table.first_multi; i++) {
953             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
954                 return 1;
955             }
956         }
957     }
958 
959     return 0;
960 }
961 
962 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
963 {
964     VirtIONet *n = qemu_get_nic_opaque(nc);
965     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
966     VirtIODevice *vdev = VIRTIO_DEVICE(n);
967     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
968     struct virtio_net_hdr_mrg_rxbuf mhdr;
969     unsigned mhdr_cnt = 0;
970     size_t offset, i, guest_offset;
971 
972     if (!virtio_net_can_receive(nc)) {
973         return -1;
974     }
975 
976     /* hdr_len refers to the header we supply to the guest */
977     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
978         return 0;
979     }
980 
981     if (!receive_filter(n, buf, size))
982         return size;
983 
984     offset = i = 0;
985 
986     while (offset < size) {
987         VirtQueueElement elem;
988         int len, total;
989         const struct iovec *sg = elem.in_sg;
990 
991         total = 0;
992 
993         if (virtqueue_pop(q->rx_vq, &elem) == 0) {
994             if (i == 0)
995                 return -1;
996             error_report("virtio-net unexpected empty queue: "
997                     "i %zd mergeable %d offset %zd, size %zd, "
998                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
999                     i, n->mergeable_rx_bufs, offset, size,
1000                     n->guest_hdr_len, n->host_hdr_len, vdev->guest_features);
1001             exit(1);
1002         }
1003 
1004         if (elem.in_num < 1) {
1005             error_report("virtio-net receive queue contains no in buffers");
1006             exit(1);
1007         }
1008 
1009         if (i == 0) {
1010             assert(offset == 0);
1011             if (n->mergeable_rx_bufs) {
1012                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1013                                     sg, elem.in_num,
1014                                     offsetof(typeof(mhdr), num_buffers),
1015                                     sizeof(mhdr.num_buffers));
1016             }
1017 
1018             receive_header(n, sg, elem.in_num, buf, size);
1019             offset = n->host_hdr_len;
1020             total += n->guest_hdr_len;
1021             guest_offset = n->guest_hdr_len;
1022         } else {
1023             guest_offset = 0;
1024         }
1025 
1026         /* copy in packet.  ugh */
1027         len = iov_from_buf(sg, elem.in_num, guest_offset,
1028                            buf + offset, size - offset);
1029         total += len;
1030         offset += len;
1031         /* If buffers can't be merged, at this point we
1032          * must have consumed the complete packet.
1033          * Otherwise, drop it. */
1034         if (!n->mergeable_rx_bufs && offset < size) {
1035 #if 0
1036             error_report("virtio-net truncated non-mergeable packet: "
1037                          "i %zd mergeable %d offset %zd, size %zd, "
1038                          "guest hdr len %zd, host hdr len %zd",
1039                          i, n->mergeable_rx_bufs,
1040                          offset, size, n->guest_hdr_len, n->host_hdr_len);
1041 #endif
1042             return size;
1043         }
1044 
1045         /* signal other side */
1046         virtqueue_fill(q->rx_vq, &elem, total, i++);
1047     }
1048 
1049     if (mhdr_cnt) {
1050         stw_p(&mhdr.num_buffers, i);
1051         iov_from_buf(mhdr_sg, mhdr_cnt,
1052                      0,
1053                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1054     }
1055 
1056     virtqueue_flush(q->rx_vq, i);
1057     virtio_notify(vdev, q->rx_vq);
1058 
1059     return size;
1060 }
1061 
1062 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1063 
1064 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1065 {
1066     VirtIONet *n = qemu_get_nic_opaque(nc);
1067     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1068     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1069 
1070     virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
1071     virtio_notify(vdev, q->tx_vq);
1072 
1073     q->async_tx.elem.out_num = q->async_tx.len = 0;
1074 
1075     virtio_queue_set_notification(q->tx_vq, 1);
1076     virtio_net_flush_tx(q);
1077 }
1078 
1079 /* TX */
1080 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
1081 {
1082     VirtIONet *n = q->n;
1083     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1084     VirtQueueElement elem;
1085     int32_t num_packets = 0;
1086     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
1087     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1088         return num_packets;
1089     }
1090 
1091     assert(vdev->vm_running);
1092 
1093     if (q->async_tx.elem.out_num) {
1094         virtio_queue_set_notification(q->tx_vq, 0);
1095         return num_packets;
1096     }
1097 
1098     while (virtqueue_pop(q->tx_vq, &elem)) {
1099         ssize_t ret, len;
1100         unsigned int out_num = elem.out_num;
1101         struct iovec *out_sg = &elem.out_sg[0];
1102         struct iovec sg[VIRTQUEUE_MAX_SIZE];
1103 
1104         if (out_num < 1) {
1105             error_report("virtio-net header not in first element");
1106             exit(1);
1107         }
1108 
1109         /*
1110          * If host wants to see the guest header as is, we can
1111          * pass it on unchanged. Otherwise, copy just the parts
1112          * that host is interested in.
1113          */
1114         assert(n->host_hdr_len <= n->guest_hdr_len);
1115         if (n->host_hdr_len != n->guest_hdr_len) {
1116             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
1117                                        out_sg, out_num,
1118                                        0, n->host_hdr_len);
1119             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
1120                              out_sg, out_num,
1121                              n->guest_hdr_len, -1);
1122             out_num = sg_num;
1123             out_sg = sg;
1124         }
1125 
1126         len = n->guest_hdr_len;
1127 
1128         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
1129                                       out_sg, out_num, virtio_net_tx_complete);
1130         if (ret == 0) {
1131             virtio_queue_set_notification(q->tx_vq, 0);
1132             q->async_tx.elem = elem;
1133             q->async_tx.len  = len;
1134             return -EBUSY;
1135         }
1136 
1137         len += ret;
1138 
1139         virtqueue_push(q->tx_vq, &elem, 0);
1140         virtio_notify(vdev, q->tx_vq);
1141 
1142         if (++num_packets >= n->tx_burst) {
1143             break;
1144         }
1145     }
1146     return num_packets;
1147 }
1148 
1149 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
1150 {
1151     VirtIONet *n = VIRTIO_NET(vdev);
1152     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1153 
1154     /* This happens when device was stopped but VCPU wasn't. */
1155     if (!vdev->vm_running) {
1156         q->tx_waiting = 1;
1157         return;
1158     }
1159 
1160     if (q->tx_waiting) {
1161         virtio_queue_set_notification(vq, 1);
1162         timer_del(q->tx_timer);
1163         q->tx_waiting = 0;
1164         virtio_net_flush_tx(q);
1165     } else {
1166         timer_mod(q->tx_timer,
1167                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
1168         q->tx_waiting = 1;
1169         virtio_queue_set_notification(vq, 0);
1170     }
1171 }
1172 
1173 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1174 {
1175     VirtIONet *n = VIRTIO_NET(vdev);
1176     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1177 
1178     if (unlikely(q->tx_waiting)) {
1179         return;
1180     }
1181     q->tx_waiting = 1;
1182     /* This happens when device was stopped but VCPU wasn't. */
1183     if (!vdev->vm_running) {
1184         return;
1185     }
1186     virtio_queue_set_notification(vq, 0);
1187     qemu_bh_schedule(q->tx_bh);
1188 }
1189 
1190 static void virtio_net_tx_timer(void *opaque)
1191 {
1192     VirtIONetQueue *q = opaque;
1193     VirtIONet *n = q->n;
1194     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1195     assert(vdev->vm_running);
1196 
1197     q->tx_waiting = 0;
1198 
1199     /* Just in case the driver is not ready on more */
1200     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1201         return;
1202     }
1203 
1204     virtio_queue_set_notification(q->tx_vq, 1);
1205     virtio_net_flush_tx(q);
1206 }
1207 
1208 static void virtio_net_tx_bh(void *opaque)
1209 {
1210     VirtIONetQueue *q = opaque;
1211     VirtIONet *n = q->n;
1212     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1213     int32_t ret;
1214 
1215     assert(vdev->vm_running);
1216 
1217     q->tx_waiting = 0;
1218 
1219     /* Just in case the driver is not ready on more */
1220     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1221         return;
1222     }
1223 
1224     ret = virtio_net_flush_tx(q);
1225     if (ret == -EBUSY) {
1226         return; /* Notification re-enable handled by tx_complete */
1227     }
1228 
1229     /* If we flush a full burst of packets, assume there are
1230      * more coming and immediately reschedule */
1231     if (ret >= n->tx_burst) {
1232         qemu_bh_schedule(q->tx_bh);
1233         q->tx_waiting = 1;
1234         return;
1235     }
1236 
1237     /* If less than a full burst, re-enable notification and flush
1238      * anything that may have come in while we weren't looking.  If
1239      * we find something, assume the guest is still active and reschedule */
1240     virtio_queue_set_notification(q->tx_vq, 1);
1241     if (virtio_net_flush_tx(q) > 0) {
1242         virtio_queue_set_notification(q->tx_vq, 0);
1243         qemu_bh_schedule(q->tx_bh);
1244         q->tx_waiting = 1;
1245     }
1246 }
1247 
1248 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1249 {
1250     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1251     int i, max = multiqueue ? n->max_queues : 1;
1252 
1253     n->multiqueue = multiqueue;
1254 
1255     for (i = 2; i <= n->max_queues * 2 + 1; i++) {
1256         virtio_del_queue(vdev, i);
1257     }
1258 
1259     for (i = 1; i < max; i++) {
1260         n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1261         if (n->vqs[i].tx_timer) {
1262             n->vqs[i].tx_vq =
1263                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1264             n->vqs[i].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
1265                                                    virtio_net_tx_timer,
1266                                                    &n->vqs[i]);
1267         } else {
1268             n->vqs[i].tx_vq =
1269                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1270             n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1271         }
1272 
1273         n->vqs[i].tx_waiting = 0;
1274         n->vqs[i].n = n;
1275     }
1276 
1277     /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack
1278      * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid
1279      * breaking them.
1280      */
1281     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1282 
1283     virtio_net_set_queues(n);
1284 }
1285 
1286 static void virtio_net_save(QEMUFile *f, void *opaque)
1287 {
1288     int i;
1289     VirtIONet *n = opaque;
1290     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1291 
1292     /* At this point, backend must be stopped, otherwise
1293      * it might keep writing to memory. */
1294     assert(!n->vhost_started);
1295     virtio_save(vdev, f);
1296 
1297     qemu_put_buffer(f, n->mac, ETH_ALEN);
1298     qemu_put_be32(f, n->vqs[0].tx_waiting);
1299     qemu_put_be32(f, n->mergeable_rx_bufs);
1300     qemu_put_be16(f, n->status);
1301     qemu_put_byte(f, n->promisc);
1302     qemu_put_byte(f, n->allmulti);
1303     qemu_put_be32(f, n->mac_table.in_use);
1304     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1305     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1306     qemu_put_be32(f, n->has_vnet_hdr);
1307     qemu_put_byte(f, n->mac_table.multi_overflow);
1308     qemu_put_byte(f, n->mac_table.uni_overflow);
1309     qemu_put_byte(f, n->alluni);
1310     qemu_put_byte(f, n->nomulti);
1311     qemu_put_byte(f, n->nouni);
1312     qemu_put_byte(f, n->nobcast);
1313     qemu_put_byte(f, n->has_ufo);
1314     if (n->max_queues > 1) {
1315         qemu_put_be16(f, n->max_queues);
1316         qemu_put_be16(f, n->curr_queues);
1317         for (i = 1; i < n->curr_queues; i++) {
1318             qemu_put_be32(f, n->vqs[i].tx_waiting);
1319         }
1320     }
1321 
1322     if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) {
1323         qemu_put_be64(f, n->curr_guest_offloads);
1324     }
1325 }
1326 
1327 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1328 {
1329     VirtIONet *n = opaque;
1330     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1331     int ret, i, link_down;
1332 
1333     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1334         return -EINVAL;
1335 
1336     ret = virtio_load(vdev, f);
1337     if (ret) {
1338         return ret;
1339     }
1340 
1341     qemu_get_buffer(f, n->mac, ETH_ALEN);
1342     n->vqs[0].tx_waiting = qemu_get_be32(f);
1343 
1344     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1345 
1346     if (version_id >= 3)
1347         n->status = qemu_get_be16(f);
1348 
1349     if (version_id >= 4) {
1350         if (version_id < 8) {
1351             n->promisc = qemu_get_be32(f);
1352             n->allmulti = qemu_get_be32(f);
1353         } else {
1354             n->promisc = qemu_get_byte(f);
1355             n->allmulti = qemu_get_byte(f);
1356         }
1357     }
1358 
1359     if (version_id >= 5) {
1360         n->mac_table.in_use = qemu_get_be32(f);
1361         /* MAC_TABLE_ENTRIES may be different from the saved image */
1362         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1363             qemu_get_buffer(f, n->mac_table.macs,
1364                             n->mac_table.in_use * ETH_ALEN);
1365         } else {
1366             int64_t i;
1367 
1368             /* Overflow detected - can happen if source has a larger MAC table.
1369              * We simply set overflow flag so there's no need to maintain the
1370              * table of addresses, discard them all.
1371              * Note: 64 bit math to avoid integer overflow.
1372              */
1373             for (i = 0; i < (int64_t)n->mac_table.in_use * ETH_ALEN; ++i) {
1374                 qemu_get_byte(f);
1375             }
1376             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1377             n->mac_table.in_use = 0;
1378         }
1379     }
1380 
1381     if (version_id >= 6)
1382         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1383 
1384     if (version_id >= 7) {
1385         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1386             error_report("virtio-net: saved image requires vnet_hdr=on");
1387             return -1;
1388         }
1389     }
1390 
1391     if (version_id >= 9) {
1392         n->mac_table.multi_overflow = qemu_get_byte(f);
1393         n->mac_table.uni_overflow = qemu_get_byte(f);
1394     }
1395 
1396     if (version_id >= 10) {
1397         n->alluni = qemu_get_byte(f);
1398         n->nomulti = qemu_get_byte(f);
1399         n->nouni = qemu_get_byte(f);
1400         n->nobcast = qemu_get_byte(f);
1401     }
1402 
1403     if (version_id >= 11) {
1404         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1405             error_report("virtio-net: saved image requires TUN_F_UFO support");
1406             return -1;
1407         }
1408     }
1409 
1410     if (n->max_queues > 1) {
1411         if (n->max_queues != qemu_get_be16(f)) {
1412             error_report("virtio-net: different max_queues ");
1413             return -1;
1414         }
1415 
1416         n->curr_queues = qemu_get_be16(f);
1417         if (n->curr_queues > n->max_queues) {
1418             error_report("virtio-net: curr_queues %x > max_queues %x",
1419                          n->curr_queues, n->max_queues);
1420             return -1;
1421         }
1422         for (i = 1; i < n->curr_queues; i++) {
1423             n->vqs[i].tx_waiting = qemu_get_be32(f);
1424         }
1425     }
1426 
1427     if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) {
1428         n->curr_guest_offloads = qemu_get_be64(f);
1429     } else {
1430         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
1431     }
1432 
1433     if (peer_has_vnet_hdr(n)) {
1434         virtio_net_apply_guest_offloads(n);
1435     }
1436 
1437     virtio_net_set_queues(n);
1438 
1439     /* Find the first multicast entry in the saved MAC filter */
1440     for (i = 0; i < n->mac_table.in_use; i++) {
1441         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1442             break;
1443         }
1444     }
1445     n->mac_table.first_multi = i;
1446 
1447     /* nc.link_down can't be migrated, so infer link_down according
1448      * to link status bit in n->status */
1449     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1450     for (i = 0; i < n->max_queues; i++) {
1451         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1452     }
1453 
1454     return 0;
1455 }
1456 
1457 static void virtio_net_cleanup(NetClientState *nc)
1458 {
1459     VirtIONet *n = qemu_get_nic_opaque(nc);
1460 
1461     n->nic = NULL;
1462 }
1463 
1464 static NetClientInfo net_virtio_info = {
1465     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1466     .size = sizeof(NICState),
1467     .can_receive = virtio_net_can_receive,
1468     .receive = virtio_net_receive,
1469     .cleanup = virtio_net_cleanup,
1470     .link_status_changed = virtio_net_set_link_status,
1471     .query_rx_filter = virtio_net_query_rxfilter,
1472 };
1473 
1474 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1475 {
1476     VirtIONet *n = VIRTIO_NET(vdev);
1477     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1478     assert(n->vhost_started);
1479     return vhost_net_virtqueue_pending(tap_get_vhost_net(nc->peer), idx);
1480 }
1481 
1482 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1483                                            bool mask)
1484 {
1485     VirtIONet *n = VIRTIO_NET(vdev);
1486     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1487     assert(n->vhost_started);
1488     vhost_net_virtqueue_mask(tap_get_vhost_net(nc->peer),
1489                              vdev, idx, mask);
1490 }
1491 
1492 void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features)
1493 {
1494     int i, config_size = 0;
1495     host_features |= (1 << VIRTIO_NET_F_MAC);
1496     for (i = 0; feature_sizes[i].flags != 0; i++) {
1497         if (host_features & feature_sizes[i].flags) {
1498             config_size = MAX(feature_sizes[i].end, config_size);
1499         }
1500     }
1501     n->config_size = config_size;
1502 }
1503 
1504 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
1505                                    const char *type)
1506 {
1507     /*
1508      * The name can be NULL, the netclient name will be type.x.
1509      */
1510     assert(type != NULL);
1511 
1512     if (n->netclient_name) {
1513         g_free(n->netclient_name);
1514         n->netclient_name = NULL;
1515     }
1516     if (n->netclient_type) {
1517         g_free(n->netclient_type);
1518         n->netclient_type = NULL;
1519     }
1520 
1521     if (name != NULL) {
1522         n->netclient_name = g_strdup(name);
1523     }
1524     n->netclient_type = g_strdup(type);
1525 }
1526 
1527 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
1528 {
1529     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1530     VirtIONet *n = VIRTIO_NET(dev);
1531     NetClientState *nc;
1532     int i;
1533 
1534     virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
1535 
1536     n->max_queues = MAX(n->nic_conf.queues, 1);
1537     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1538     n->vqs[0].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1539     n->curr_queues = 1;
1540     n->vqs[0].n = n;
1541     n->tx_timeout = n->net_conf.txtimer;
1542 
1543     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1544                        && strcmp(n->net_conf.tx, "bh")) {
1545         error_report("virtio-net: "
1546                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1547                      n->net_conf.tx);
1548         error_report("Defaulting to \"bh\"");
1549     }
1550 
1551     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1552         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1553                                            virtio_net_handle_tx_timer);
1554         n->vqs[0].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, virtio_net_tx_timer,
1555                                                &n->vqs[0]);
1556     } else {
1557         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1558                                            virtio_net_handle_tx_bh);
1559         n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]);
1560     }
1561     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1562     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1563     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1564     n->status = VIRTIO_NET_S_LINK_UP;
1565 
1566     if (n->netclient_type) {
1567         /*
1568          * Happen when virtio_net_set_netclient_name has been called.
1569          */
1570         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1571                               n->netclient_type, n->netclient_name, n);
1572     } else {
1573         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1574                               object_get_typename(OBJECT(dev)), dev->id, n);
1575     }
1576 
1577     peer_test_vnet_hdr(n);
1578     if (peer_has_vnet_hdr(n)) {
1579         for (i = 0; i < n->max_queues; i++) {
1580             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1581         }
1582         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1583     } else {
1584         n->host_hdr_len = 0;
1585     }
1586 
1587     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1588 
1589     n->vqs[0].tx_waiting = 0;
1590     n->tx_burst = n->net_conf.txburst;
1591     virtio_net_set_mrg_rx_bufs(n, 0);
1592     n->promisc = 1; /* for compatibility */
1593 
1594     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1595 
1596     n->vlans = g_malloc0(MAX_VLAN >> 3);
1597 
1598     nc = qemu_get_queue(n->nic);
1599     nc->rxfilter_notify_enabled = 1;
1600 
1601     n->qdev = dev;
1602     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1603                     virtio_net_save, virtio_net_load, n);
1604 
1605     add_boot_device_path(n->nic_conf.bootindex, dev, "/ethernet-phy@0");
1606 }
1607 
1608 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
1609 {
1610     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1611     VirtIONet *n = VIRTIO_NET(dev);
1612     int i;
1613 
1614     /* This will stop vhost backend if appropriate. */
1615     virtio_net_set_status(vdev, 0);
1616 
1617     unregister_savevm(dev, "virtio-net", n);
1618 
1619     if (n->netclient_name) {
1620         g_free(n->netclient_name);
1621         n->netclient_name = NULL;
1622     }
1623     if (n->netclient_type) {
1624         g_free(n->netclient_type);
1625         n->netclient_type = NULL;
1626     }
1627 
1628     g_free(n->mac_table.macs);
1629     g_free(n->vlans);
1630 
1631     for (i = 0; i < n->max_queues; i++) {
1632         VirtIONetQueue *q = &n->vqs[i];
1633         NetClientState *nc = qemu_get_subqueue(n->nic, i);
1634 
1635         qemu_purge_queued_packets(nc);
1636 
1637         if (q->tx_timer) {
1638             timer_del(q->tx_timer);
1639             timer_free(q->tx_timer);
1640         } else if (q->tx_bh) {
1641             qemu_bh_delete(q->tx_bh);
1642         }
1643     }
1644 
1645     g_free(n->vqs);
1646     qemu_del_nic(n->nic);
1647     virtio_cleanup(vdev);
1648 }
1649 
1650 static void virtio_net_instance_init(Object *obj)
1651 {
1652     VirtIONet *n = VIRTIO_NET(obj);
1653 
1654     /*
1655      * The default config_size is sizeof(struct virtio_net_config).
1656      * Can be overriden with virtio_net_set_config_size.
1657      */
1658     n->config_size = sizeof(struct virtio_net_config);
1659 }
1660 
1661 static Property virtio_net_properties[] = {
1662     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
1663     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
1664                                                TX_TIMER_INTERVAL),
1665     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
1666     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
1667     DEFINE_PROP_END_OF_LIST(),
1668 };
1669 
1670 static void virtio_net_class_init(ObjectClass *klass, void *data)
1671 {
1672     DeviceClass *dc = DEVICE_CLASS(klass);
1673     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1674 
1675     dc->props = virtio_net_properties;
1676     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
1677     vdc->realize = virtio_net_device_realize;
1678     vdc->unrealize = virtio_net_device_unrealize;
1679     vdc->get_config = virtio_net_get_config;
1680     vdc->set_config = virtio_net_set_config;
1681     vdc->get_features = virtio_net_get_features;
1682     vdc->set_features = virtio_net_set_features;
1683     vdc->bad_features = virtio_net_bad_features;
1684     vdc->reset = virtio_net_reset;
1685     vdc->set_status = virtio_net_set_status;
1686     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
1687     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
1688 }
1689 
1690 static const TypeInfo virtio_net_info = {
1691     .name = TYPE_VIRTIO_NET,
1692     .parent = TYPE_VIRTIO_DEVICE,
1693     .instance_size = sizeof(VirtIONet),
1694     .instance_init = virtio_net_instance_init,
1695     .class_init = virtio_net_class_init,
1696 };
1697 
1698 static void virtio_register_types(void)
1699 {
1700     type_register_static(&virtio_net_info);
1701 }
1702 
1703 type_init(virtio_register_types)
1704