xref: /openbmc/qemu/hw/net/virtio-net.c (revision 01c22f2c)
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 RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
226 {
227     VirtIONet *n = qemu_get_nic_opaque(nc);
228     RxFilterInfo *info;
229     strList *str_list, *entry;
230     intList *int_list, *int_entry;
231     int i, j;
232 
233     info = g_malloc0(sizeof(*info));
234     info->name = g_strdup(nc->name);
235     info->promiscuous = n->promisc;
236 
237     if (n->nouni) {
238         info->unicast = RX_STATE_NONE;
239     } else if (n->alluni) {
240         info->unicast = RX_STATE_ALL;
241     } else {
242         info->unicast = RX_STATE_NORMAL;
243     }
244 
245     if (n->nomulti) {
246         info->multicast = RX_STATE_NONE;
247     } else if (n->allmulti) {
248         info->multicast = RX_STATE_ALL;
249     } else {
250         info->multicast = RX_STATE_NORMAL;
251     }
252 
253     info->broadcast_allowed = n->nobcast;
254     info->multicast_overflow = n->mac_table.multi_overflow;
255     info->unicast_overflow = n->mac_table.uni_overflow;
256 
257     info->main_mac = mac_strdup_printf(n->mac);
258 
259     str_list = NULL;
260     for (i = 0; i < n->mac_table.first_multi; i++) {
261         entry = g_malloc0(sizeof(*entry));
262         entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
263         entry->next = str_list;
264         str_list = entry;
265     }
266     info->unicast_table = str_list;
267 
268     str_list = NULL;
269     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
270         entry = g_malloc0(sizeof(*entry));
271         entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
272         entry->next = str_list;
273         str_list = entry;
274     }
275     info->multicast_table = str_list;
276 
277     int_list = NULL;
278     for (i = 0; i < MAX_VLAN >> 5; i++) {
279         for (j = 0; n->vlans[i] && j < 0x1f; j++) {
280             if (n->vlans[i] & (1U << j)) {
281                 int_entry = g_malloc0(sizeof(*int_entry));
282                 int_entry->value = (i << 5) + j;
283                 int_entry->next = int_list;
284                 int_list = int_entry;
285             }
286         }
287     }
288     info->vlan_table = int_list;
289 
290     /* enable event notification after query */
291     nc->rxfilter_notify_enabled = 1;
292 
293     return info;
294 }
295 
296 static void virtio_net_reset(VirtIODevice *vdev)
297 {
298     VirtIONet *n = VIRTIO_NET(vdev);
299 
300     /* Reset back to compatibility mode */
301     n->promisc = 1;
302     n->allmulti = 0;
303     n->alluni = 0;
304     n->nomulti = 0;
305     n->nouni = 0;
306     n->nobcast = 0;
307     /* multiqueue is disabled by default */
308     n->curr_queues = 1;
309 
310     /* Flush any MAC and VLAN filter table state */
311     n->mac_table.in_use = 0;
312     n->mac_table.first_multi = 0;
313     n->mac_table.multi_overflow = 0;
314     n->mac_table.uni_overflow = 0;
315     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
316     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
317     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
318     memset(n->vlans, 0, MAX_VLAN >> 3);
319 }
320 
321 static void peer_test_vnet_hdr(VirtIONet *n)
322 {
323     NetClientState *nc = qemu_get_queue(n->nic);
324     if (!nc->peer) {
325         return;
326     }
327 
328     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
329 }
330 
331 static int peer_has_vnet_hdr(VirtIONet *n)
332 {
333     return n->has_vnet_hdr;
334 }
335 
336 static int peer_has_ufo(VirtIONet *n)
337 {
338     if (!peer_has_vnet_hdr(n))
339         return 0;
340 
341     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
342 
343     return n->has_ufo;
344 }
345 
346 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs)
347 {
348     int i;
349     NetClientState *nc;
350 
351     n->mergeable_rx_bufs = mergeable_rx_bufs;
352 
353     n->guest_hdr_len = n->mergeable_rx_bufs ?
354         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
355 
356     for (i = 0; i < n->max_queues; i++) {
357         nc = qemu_get_subqueue(n->nic, i);
358 
359         if (peer_has_vnet_hdr(n) &&
360             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
361             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
362             n->host_hdr_len = n->guest_hdr_len;
363         }
364     }
365 }
366 
367 static int peer_attach(VirtIONet *n, int index)
368 {
369     NetClientState *nc = qemu_get_subqueue(n->nic, index);
370 
371     if (!nc->peer) {
372         return 0;
373     }
374 
375     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
376         return 0;
377     }
378 
379     return tap_enable(nc->peer);
380 }
381 
382 static int peer_detach(VirtIONet *n, int index)
383 {
384     NetClientState *nc = qemu_get_subqueue(n->nic, index);
385 
386     if (!nc->peer) {
387         return 0;
388     }
389 
390     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
391         return 0;
392     }
393 
394     return tap_disable(nc->peer);
395 }
396 
397 static void virtio_net_set_queues(VirtIONet *n)
398 {
399     int i;
400     int r;
401 
402     for (i = 0; i < n->max_queues; i++) {
403         if (i < n->curr_queues) {
404             r = peer_attach(n, i);
405             assert(!r);
406         } else {
407             r = peer_detach(n, i);
408             assert(!r);
409         }
410     }
411 }
412 
413 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
414 
415 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
416 {
417     VirtIONet *n = VIRTIO_NET(vdev);
418     NetClientState *nc = qemu_get_queue(n->nic);
419 
420     features |= (1 << VIRTIO_NET_F_MAC);
421 
422     if (!peer_has_vnet_hdr(n)) {
423         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
424         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
425         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
426         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
427 
428         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
429         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
430         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
431         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
432     }
433 
434     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
435         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
436         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
437     }
438 
439     if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
440         return features;
441     }
442     if (!tap_get_vhost_net(nc->peer)) {
443         return features;
444     }
445     return vhost_net_get_features(tap_get_vhost_net(nc->peer), features);
446 }
447 
448 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
449 {
450     uint32_t features = 0;
451 
452     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
453      * but also these: */
454     features |= (1 << VIRTIO_NET_F_MAC);
455     features |= (1 << VIRTIO_NET_F_CSUM);
456     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
457     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
458     features |= (1 << VIRTIO_NET_F_HOST_ECN);
459 
460     return features;
461 }
462 
463 static void virtio_net_apply_guest_offloads(VirtIONet *n)
464 {
465     qemu_set_offload(qemu_get_queue(n->nic)->peer,
466             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
467             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
468             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
469             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
470             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
471 }
472 
473 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
474 {
475     static const uint64_t guest_offloads_mask =
476         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
477         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
478         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
479         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
480         (1ULL << VIRTIO_NET_F_GUEST_UFO);
481 
482     return guest_offloads_mask & features;
483 }
484 
485 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
486 {
487     VirtIODevice *vdev = VIRTIO_DEVICE(n);
488     return virtio_net_guest_offloads_by_features(vdev->guest_features);
489 }
490 
491 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
492 {
493     VirtIONet *n = VIRTIO_NET(vdev);
494     int i;
495 
496     virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)));
497 
498     virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
499 
500     if (n->has_vnet_hdr) {
501         n->curr_guest_offloads =
502             virtio_net_guest_offloads_by_features(features);
503         virtio_net_apply_guest_offloads(n);
504     }
505 
506     for (i = 0;  i < n->max_queues; i++) {
507         NetClientState *nc = qemu_get_subqueue(n->nic, i);
508 
509         if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
510             continue;
511         }
512         if (!tap_get_vhost_net(nc->peer)) {
513             continue;
514         }
515         vhost_net_ack_features(tap_get_vhost_net(nc->peer), features);
516     }
517 }
518 
519 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
520                                      struct iovec *iov, unsigned int iov_cnt)
521 {
522     uint8_t on;
523     size_t s;
524     NetClientState *nc = qemu_get_queue(n->nic);
525 
526     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
527     if (s != sizeof(on)) {
528         return VIRTIO_NET_ERR;
529     }
530 
531     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
532         n->promisc = on;
533     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
534         n->allmulti = on;
535     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
536         n->alluni = on;
537     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
538         n->nomulti = on;
539     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
540         n->nouni = on;
541     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
542         n->nobcast = on;
543     } else {
544         return VIRTIO_NET_ERR;
545     }
546 
547     rxfilter_notify(nc);
548 
549     return VIRTIO_NET_OK;
550 }
551 
552 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
553                                      struct iovec *iov, unsigned int iov_cnt)
554 {
555     VirtIODevice *vdev = VIRTIO_DEVICE(n);
556     uint64_t offloads;
557     size_t s;
558 
559     if (!((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features)) {
560         return VIRTIO_NET_ERR;
561     }
562 
563     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
564     if (s != sizeof(offloads)) {
565         return VIRTIO_NET_ERR;
566     }
567 
568     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
569         uint64_t supported_offloads;
570 
571         if (!n->has_vnet_hdr) {
572             return VIRTIO_NET_ERR;
573         }
574 
575         supported_offloads = virtio_net_supported_guest_offloads(n);
576         if (offloads & ~supported_offloads) {
577             return VIRTIO_NET_ERR;
578         }
579 
580         n->curr_guest_offloads = offloads;
581         virtio_net_apply_guest_offloads(n);
582 
583         return VIRTIO_NET_OK;
584     } else {
585         return VIRTIO_NET_ERR;
586     }
587 }
588 
589 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
590                                  struct iovec *iov, unsigned int iov_cnt)
591 {
592     struct virtio_net_ctrl_mac mac_data;
593     size_t s;
594     NetClientState *nc = qemu_get_queue(n->nic);
595 
596     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
597         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
598             return VIRTIO_NET_ERR;
599         }
600         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
601         assert(s == sizeof(n->mac));
602         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
603         rxfilter_notify(nc);
604 
605         return VIRTIO_NET_OK;
606     }
607 
608     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
609         return VIRTIO_NET_ERR;
610     }
611 
612     int in_use = 0;
613     int first_multi = 0;
614     uint8_t uni_overflow = 0;
615     uint8_t multi_overflow = 0;
616     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
617 
618     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
619                    sizeof(mac_data.entries));
620     mac_data.entries = ldl_p(&mac_data.entries);
621     if (s != sizeof(mac_data.entries)) {
622         goto error;
623     }
624     iov_discard_front(&iov, &iov_cnt, s);
625 
626     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
627         goto error;
628     }
629 
630     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
631         s = iov_to_buf(iov, iov_cnt, 0, macs,
632                        mac_data.entries * ETH_ALEN);
633         if (s != mac_data.entries * ETH_ALEN) {
634             goto error;
635         }
636         in_use += mac_data.entries;
637     } else {
638         uni_overflow = 1;
639     }
640 
641     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
642 
643     first_multi = in_use;
644 
645     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
646                    sizeof(mac_data.entries));
647     mac_data.entries = ldl_p(&mac_data.entries);
648     if (s != sizeof(mac_data.entries)) {
649         goto error;
650     }
651 
652     iov_discard_front(&iov, &iov_cnt, s);
653 
654     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
655         goto error;
656     }
657 
658     if (in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
659         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
660                        mac_data.entries * ETH_ALEN);
661         if (s != mac_data.entries * ETH_ALEN) {
662             goto error;
663         }
664         in_use += mac_data.entries;
665     } else {
666         multi_overflow = 1;
667     }
668 
669     n->mac_table.in_use = in_use;
670     n->mac_table.first_multi = first_multi;
671     n->mac_table.uni_overflow = uni_overflow;
672     n->mac_table.multi_overflow = multi_overflow;
673     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
674     g_free(macs);
675     rxfilter_notify(nc);
676 
677     return VIRTIO_NET_OK;
678 
679 error:
680     g_free(macs);
681     return VIRTIO_NET_ERR;
682 }
683 
684 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
685                                         struct iovec *iov, unsigned int iov_cnt)
686 {
687     uint16_t vid;
688     size_t s;
689     NetClientState *nc = qemu_get_queue(n->nic);
690 
691     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
692     vid = lduw_p(&vid);
693     if (s != sizeof(vid)) {
694         return VIRTIO_NET_ERR;
695     }
696 
697     if (vid >= MAX_VLAN)
698         return VIRTIO_NET_ERR;
699 
700     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
701         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
702     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
703         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
704     else
705         return VIRTIO_NET_ERR;
706 
707     rxfilter_notify(nc);
708 
709     return VIRTIO_NET_OK;
710 }
711 
712 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
713                                 struct iovec *iov, unsigned int iov_cnt)
714 {
715     VirtIODevice *vdev = VIRTIO_DEVICE(n);
716     struct virtio_net_ctrl_mq mq;
717     size_t s;
718     uint16_t queues;
719 
720     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
721     if (s != sizeof(mq)) {
722         return VIRTIO_NET_ERR;
723     }
724 
725     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
726         return VIRTIO_NET_ERR;
727     }
728 
729     queues = lduw_p(&mq.virtqueue_pairs);
730 
731     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
732         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
733         queues > n->max_queues ||
734         !n->multiqueue) {
735         return VIRTIO_NET_ERR;
736     }
737 
738     n->curr_queues = queues;
739     /* stop the backend before changing the number of queues to avoid handling a
740      * disabled queue */
741     virtio_net_set_status(vdev, vdev->status);
742     virtio_net_set_queues(n);
743 
744     return VIRTIO_NET_OK;
745 }
746 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
747 {
748     VirtIONet *n = VIRTIO_NET(vdev);
749     struct virtio_net_ctrl_hdr ctrl;
750     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
751     VirtQueueElement elem;
752     size_t s;
753     struct iovec *iov;
754     unsigned int iov_cnt;
755 
756     while (virtqueue_pop(vq, &elem)) {
757         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
758             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
759             error_report("virtio-net ctrl missing headers");
760             exit(1);
761         }
762 
763         iov = elem.out_sg;
764         iov_cnt = elem.out_num;
765         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
766         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
767         if (s != sizeof(ctrl)) {
768             status = VIRTIO_NET_ERR;
769         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
770             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
771         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
772             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
773         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
774             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
775         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
776             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
777         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
778             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
779         }
780 
781         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
782         assert(s == sizeof(status));
783 
784         virtqueue_push(vq, &elem, sizeof(status));
785         virtio_notify(vdev, vq);
786     }
787 }
788 
789 /* RX */
790 
791 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
792 {
793     VirtIONet *n = VIRTIO_NET(vdev);
794     int queue_index = vq2q(virtio_get_queue_index(vq));
795 
796     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
797 }
798 
799 static int virtio_net_can_receive(NetClientState *nc)
800 {
801     VirtIONet *n = qemu_get_nic_opaque(nc);
802     VirtIODevice *vdev = VIRTIO_DEVICE(n);
803     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
804 
805     if (!vdev->vm_running) {
806         return 0;
807     }
808 
809     if (nc->queue_index >= n->curr_queues) {
810         return 0;
811     }
812 
813     if (!virtio_queue_ready(q->rx_vq) ||
814         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
815         return 0;
816     }
817 
818     return 1;
819 }
820 
821 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
822 {
823     VirtIONet *n = q->n;
824     if (virtio_queue_empty(q->rx_vq) ||
825         (n->mergeable_rx_bufs &&
826          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
827         virtio_queue_set_notification(q->rx_vq, 1);
828 
829         /* To avoid a race condition where the guest has made some buffers
830          * available after the above check but before notification was
831          * enabled, check for available buffers again.
832          */
833         if (virtio_queue_empty(q->rx_vq) ||
834             (n->mergeable_rx_bufs &&
835              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
836             return 0;
837         }
838     }
839 
840     virtio_queue_set_notification(q->rx_vq, 0);
841     return 1;
842 }
843 
844 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
845  * it never finds out that the packets don't have valid checksums.  This
846  * causes dhclient to get upset.  Fedora's carried a patch for ages to
847  * fix this with Xen but it hasn't appeared in an upstream release of
848  * dhclient yet.
849  *
850  * To avoid breaking existing guests, we catch udp packets and add
851  * checksums.  This is terrible but it's better than hacking the guest
852  * kernels.
853  *
854  * N.B. if we introduce a zero-copy API, this operation is no longer free so
855  * we should provide a mechanism to disable it to avoid polluting the host
856  * cache.
857  */
858 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
859                                         uint8_t *buf, size_t size)
860 {
861     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
862         (size > 27 && size < 1500) && /* normal sized MTU */
863         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
864         (buf[23] == 17) && /* ip.protocol == UDP */
865         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
866         net_checksum_calculate(buf, size);
867         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
868     }
869 }
870 
871 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
872                            const void *buf, size_t size)
873 {
874     if (n->has_vnet_hdr) {
875         /* FIXME this cast is evil */
876         void *wbuf = (void *)buf;
877         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
878                                     size - n->host_hdr_len);
879         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
880     } else {
881         struct virtio_net_hdr hdr = {
882             .flags = 0,
883             .gso_type = VIRTIO_NET_HDR_GSO_NONE
884         };
885         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
886     }
887 }
888 
889 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
890 {
891     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
892     static const uint8_t vlan[] = {0x81, 0x00};
893     uint8_t *ptr = (uint8_t *)buf;
894     int i;
895 
896     if (n->promisc)
897         return 1;
898 
899     ptr += n->host_hdr_len;
900 
901     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
902         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
903         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
904             return 0;
905     }
906 
907     if (ptr[0] & 1) { // multicast
908         if (!memcmp(ptr, bcast, sizeof(bcast))) {
909             return !n->nobcast;
910         } else if (n->nomulti) {
911             return 0;
912         } else if (n->allmulti || n->mac_table.multi_overflow) {
913             return 1;
914         }
915 
916         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
917             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
918                 return 1;
919             }
920         }
921     } else { // unicast
922         if (n->nouni) {
923             return 0;
924         } else if (n->alluni || n->mac_table.uni_overflow) {
925             return 1;
926         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
927             return 1;
928         }
929 
930         for (i = 0; i < n->mac_table.first_multi; i++) {
931             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
932                 return 1;
933             }
934         }
935     }
936 
937     return 0;
938 }
939 
940 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
941 {
942     VirtIONet *n = qemu_get_nic_opaque(nc);
943     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
944     VirtIODevice *vdev = VIRTIO_DEVICE(n);
945     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
946     struct virtio_net_hdr_mrg_rxbuf mhdr;
947     unsigned mhdr_cnt = 0;
948     size_t offset, i, guest_offset;
949 
950     if (!virtio_net_can_receive(nc)) {
951         return -1;
952     }
953 
954     /* hdr_len refers to the header we supply to the guest */
955     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
956         return 0;
957     }
958 
959     if (!receive_filter(n, buf, size))
960         return size;
961 
962     offset = i = 0;
963 
964     while (offset < size) {
965         VirtQueueElement elem;
966         int len, total;
967         const struct iovec *sg = elem.in_sg;
968 
969         total = 0;
970 
971         if (virtqueue_pop(q->rx_vq, &elem) == 0) {
972             if (i == 0)
973                 return -1;
974             error_report("virtio-net unexpected empty queue: "
975                     "i %zd mergeable %d offset %zd, size %zd, "
976                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
977                     i, n->mergeable_rx_bufs, offset, size,
978                     n->guest_hdr_len, n->host_hdr_len, vdev->guest_features);
979             exit(1);
980         }
981 
982         if (elem.in_num < 1) {
983             error_report("virtio-net receive queue contains no in buffers");
984             exit(1);
985         }
986 
987         if (i == 0) {
988             assert(offset == 0);
989             if (n->mergeable_rx_bufs) {
990                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
991                                     sg, elem.in_num,
992                                     offsetof(typeof(mhdr), num_buffers),
993                                     sizeof(mhdr.num_buffers));
994             }
995 
996             receive_header(n, sg, elem.in_num, buf, size);
997             offset = n->host_hdr_len;
998             total += n->guest_hdr_len;
999             guest_offset = n->guest_hdr_len;
1000         } else {
1001             guest_offset = 0;
1002         }
1003 
1004         /* copy in packet.  ugh */
1005         len = iov_from_buf(sg, elem.in_num, guest_offset,
1006                            buf + offset, size - offset);
1007         total += len;
1008         offset += len;
1009         /* If buffers can't be merged, at this point we
1010          * must have consumed the complete packet.
1011          * Otherwise, drop it. */
1012         if (!n->mergeable_rx_bufs && offset < size) {
1013 #if 0
1014             error_report("virtio-net truncated non-mergeable packet: "
1015                          "i %zd mergeable %d offset %zd, size %zd, "
1016                          "guest hdr len %zd, host hdr len %zd",
1017                          i, n->mergeable_rx_bufs,
1018                          offset, size, n->guest_hdr_len, n->host_hdr_len);
1019 #endif
1020             return size;
1021         }
1022 
1023         /* signal other side */
1024         virtqueue_fill(q->rx_vq, &elem, total, i++);
1025     }
1026 
1027     if (mhdr_cnt) {
1028         stw_p(&mhdr.num_buffers, i);
1029         iov_from_buf(mhdr_sg, mhdr_cnt,
1030                      0,
1031                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1032     }
1033 
1034     virtqueue_flush(q->rx_vq, i);
1035     virtio_notify(vdev, q->rx_vq);
1036 
1037     return size;
1038 }
1039 
1040 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1041 
1042 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1043 {
1044     VirtIONet *n = qemu_get_nic_opaque(nc);
1045     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1046     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1047 
1048     virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
1049     virtio_notify(vdev, q->tx_vq);
1050 
1051     q->async_tx.elem.out_num = q->async_tx.len = 0;
1052 
1053     virtio_queue_set_notification(q->tx_vq, 1);
1054     virtio_net_flush_tx(q);
1055 }
1056 
1057 /* TX */
1058 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
1059 {
1060     VirtIONet *n = q->n;
1061     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1062     VirtQueueElement elem;
1063     int32_t num_packets = 0;
1064     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
1065     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1066         return num_packets;
1067     }
1068 
1069     assert(vdev->vm_running);
1070 
1071     if (q->async_tx.elem.out_num) {
1072         virtio_queue_set_notification(q->tx_vq, 0);
1073         return num_packets;
1074     }
1075 
1076     while (virtqueue_pop(q->tx_vq, &elem)) {
1077         ssize_t ret, len;
1078         unsigned int out_num = elem.out_num;
1079         struct iovec *out_sg = &elem.out_sg[0];
1080         struct iovec sg[VIRTQUEUE_MAX_SIZE];
1081 
1082         if (out_num < 1) {
1083             error_report("virtio-net header not in first element");
1084             exit(1);
1085         }
1086 
1087         /*
1088          * If host wants to see the guest header as is, we can
1089          * pass it on unchanged. Otherwise, copy just the parts
1090          * that host is interested in.
1091          */
1092         assert(n->host_hdr_len <= n->guest_hdr_len);
1093         if (n->host_hdr_len != n->guest_hdr_len) {
1094             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
1095                                        out_sg, out_num,
1096                                        0, n->host_hdr_len);
1097             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
1098                              out_sg, out_num,
1099                              n->guest_hdr_len, -1);
1100             out_num = sg_num;
1101             out_sg = sg;
1102         }
1103 
1104         len = n->guest_hdr_len;
1105 
1106         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
1107                                       out_sg, out_num, virtio_net_tx_complete);
1108         if (ret == 0) {
1109             virtio_queue_set_notification(q->tx_vq, 0);
1110             q->async_tx.elem = elem;
1111             q->async_tx.len  = len;
1112             return -EBUSY;
1113         }
1114 
1115         len += ret;
1116 
1117         virtqueue_push(q->tx_vq, &elem, 0);
1118         virtio_notify(vdev, q->tx_vq);
1119 
1120         if (++num_packets >= n->tx_burst) {
1121             break;
1122         }
1123     }
1124     return num_packets;
1125 }
1126 
1127 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
1128 {
1129     VirtIONet *n = VIRTIO_NET(vdev);
1130     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1131 
1132     /* This happens when device was stopped but VCPU wasn't. */
1133     if (!vdev->vm_running) {
1134         q->tx_waiting = 1;
1135         return;
1136     }
1137 
1138     if (q->tx_waiting) {
1139         virtio_queue_set_notification(vq, 1);
1140         timer_del(q->tx_timer);
1141         q->tx_waiting = 0;
1142         virtio_net_flush_tx(q);
1143     } else {
1144         timer_mod(q->tx_timer,
1145                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
1146         q->tx_waiting = 1;
1147         virtio_queue_set_notification(vq, 0);
1148     }
1149 }
1150 
1151 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1152 {
1153     VirtIONet *n = VIRTIO_NET(vdev);
1154     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1155 
1156     if (unlikely(q->tx_waiting)) {
1157         return;
1158     }
1159     q->tx_waiting = 1;
1160     /* This happens when device was stopped but VCPU wasn't. */
1161     if (!vdev->vm_running) {
1162         return;
1163     }
1164     virtio_queue_set_notification(vq, 0);
1165     qemu_bh_schedule(q->tx_bh);
1166 }
1167 
1168 static void virtio_net_tx_timer(void *opaque)
1169 {
1170     VirtIONetQueue *q = opaque;
1171     VirtIONet *n = q->n;
1172     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1173     assert(vdev->vm_running);
1174 
1175     q->tx_waiting = 0;
1176 
1177     /* Just in case the driver is not ready on more */
1178     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1179         return;
1180     }
1181 
1182     virtio_queue_set_notification(q->tx_vq, 1);
1183     virtio_net_flush_tx(q);
1184 }
1185 
1186 static void virtio_net_tx_bh(void *opaque)
1187 {
1188     VirtIONetQueue *q = opaque;
1189     VirtIONet *n = q->n;
1190     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1191     int32_t ret;
1192 
1193     assert(vdev->vm_running);
1194 
1195     q->tx_waiting = 0;
1196 
1197     /* Just in case the driver is not ready on more */
1198     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1199         return;
1200     }
1201 
1202     ret = virtio_net_flush_tx(q);
1203     if (ret == -EBUSY) {
1204         return; /* Notification re-enable handled by tx_complete */
1205     }
1206 
1207     /* If we flush a full burst of packets, assume there are
1208      * more coming and immediately reschedule */
1209     if (ret >= n->tx_burst) {
1210         qemu_bh_schedule(q->tx_bh);
1211         q->tx_waiting = 1;
1212         return;
1213     }
1214 
1215     /* If less than a full burst, re-enable notification and flush
1216      * anything that may have come in while we weren't looking.  If
1217      * we find something, assume the guest is still active and reschedule */
1218     virtio_queue_set_notification(q->tx_vq, 1);
1219     if (virtio_net_flush_tx(q) > 0) {
1220         virtio_queue_set_notification(q->tx_vq, 0);
1221         qemu_bh_schedule(q->tx_bh);
1222         q->tx_waiting = 1;
1223     }
1224 }
1225 
1226 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1227 {
1228     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1229     int i, max = multiqueue ? n->max_queues : 1;
1230 
1231     n->multiqueue = multiqueue;
1232 
1233     for (i = 2; i <= n->max_queues * 2 + 1; i++) {
1234         virtio_del_queue(vdev, i);
1235     }
1236 
1237     for (i = 1; i < max; i++) {
1238         n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1239         if (n->vqs[i].tx_timer) {
1240             n->vqs[i].tx_vq =
1241                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1242             n->vqs[i].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
1243                                                    virtio_net_tx_timer,
1244                                                    &n->vqs[i]);
1245         } else {
1246             n->vqs[i].tx_vq =
1247                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1248             n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1249         }
1250 
1251         n->vqs[i].tx_waiting = 0;
1252         n->vqs[i].n = n;
1253     }
1254 
1255     /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack
1256      * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid
1257      * breaking them.
1258      */
1259     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1260 
1261     virtio_net_set_queues(n);
1262 }
1263 
1264 static void virtio_net_save(QEMUFile *f, void *opaque)
1265 {
1266     int i;
1267     VirtIONet *n = opaque;
1268     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1269 
1270     /* At this point, backend must be stopped, otherwise
1271      * it might keep writing to memory. */
1272     assert(!n->vhost_started);
1273     virtio_save(vdev, f);
1274 
1275     qemu_put_buffer(f, n->mac, ETH_ALEN);
1276     qemu_put_be32(f, n->vqs[0].tx_waiting);
1277     qemu_put_be32(f, n->mergeable_rx_bufs);
1278     qemu_put_be16(f, n->status);
1279     qemu_put_byte(f, n->promisc);
1280     qemu_put_byte(f, n->allmulti);
1281     qemu_put_be32(f, n->mac_table.in_use);
1282     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1283     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1284     qemu_put_be32(f, n->has_vnet_hdr);
1285     qemu_put_byte(f, n->mac_table.multi_overflow);
1286     qemu_put_byte(f, n->mac_table.uni_overflow);
1287     qemu_put_byte(f, n->alluni);
1288     qemu_put_byte(f, n->nomulti);
1289     qemu_put_byte(f, n->nouni);
1290     qemu_put_byte(f, n->nobcast);
1291     qemu_put_byte(f, n->has_ufo);
1292     if (n->max_queues > 1) {
1293         qemu_put_be16(f, n->max_queues);
1294         qemu_put_be16(f, n->curr_queues);
1295         for (i = 1; i < n->curr_queues; i++) {
1296             qemu_put_be32(f, n->vqs[i].tx_waiting);
1297         }
1298     }
1299 
1300     if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) {
1301         qemu_put_be64(f, n->curr_guest_offloads);
1302     }
1303 }
1304 
1305 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1306 {
1307     VirtIONet *n = opaque;
1308     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1309     int ret, i, link_down;
1310 
1311     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1312         return -EINVAL;
1313 
1314     ret = virtio_load(vdev, f);
1315     if (ret) {
1316         return ret;
1317     }
1318 
1319     qemu_get_buffer(f, n->mac, ETH_ALEN);
1320     n->vqs[0].tx_waiting = qemu_get_be32(f);
1321 
1322     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1323 
1324     if (version_id >= 3)
1325         n->status = qemu_get_be16(f);
1326 
1327     if (version_id >= 4) {
1328         if (version_id < 8) {
1329             n->promisc = qemu_get_be32(f);
1330             n->allmulti = qemu_get_be32(f);
1331         } else {
1332             n->promisc = qemu_get_byte(f);
1333             n->allmulti = qemu_get_byte(f);
1334         }
1335     }
1336 
1337     if (version_id >= 5) {
1338         n->mac_table.in_use = qemu_get_be32(f);
1339         /* MAC_TABLE_ENTRIES may be different from the saved image */
1340         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1341             qemu_get_buffer(f, n->mac_table.macs,
1342                             n->mac_table.in_use * ETH_ALEN);
1343         } else if (n->mac_table.in_use) {
1344             uint8_t *buf = g_malloc0(n->mac_table.in_use);
1345             qemu_get_buffer(f, buf, n->mac_table.in_use * ETH_ALEN);
1346             g_free(buf);
1347             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1348             n->mac_table.in_use = 0;
1349         }
1350     }
1351 
1352     if (version_id >= 6)
1353         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1354 
1355     if (version_id >= 7) {
1356         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1357             error_report("virtio-net: saved image requires vnet_hdr=on");
1358             return -1;
1359         }
1360     }
1361 
1362     if (version_id >= 9) {
1363         n->mac_table.multi_overflow = qemu_get_byte(f);
1364         n->mac_table.uni_overflow = qemu_get_byte(f);
1365     }
1366 
1367     if (version_id >= 10) {
1368         n->alluni = qemu_get_byte(f);
1369         n->nomulti = qemu_get_byte(f);
1370         n->nouni = qemu_get_byte(f);
1371         n->nobcast = qemu_get_byte(f);
1372     }
1373 
1374     if (version_id >= 11) {
1375         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1376             error_report("virtio-net: saved image requires TUN_F_UFO support");
1377             return -1;
1378         }
1379     }
1380 
1381     if (n->max_queues > 1) {
1382         if (n->max_queues != qemu_get_be16(f)) {
1383             error_report("virtio-net: different max_queues ");
1384             return -1;
1385         }
1386 
1387         n->curr_queues = qemu_get_be16(f);
1388         for (i = 1; i < n->curr_queues; i++) {
1389             n->vqs[i].tx_waiting = qemu_get_be32(f);
1390         }
1391     }
1392 
1393     if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) {
1394         n->curr_guest_offloads = qemu_get_be64(f);
1395     } else {
1396         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
1397     }
1398 
1399     if (peer_has_vnet_hdr(n)) {
1400         virtio_net_apply_guest_offloads(n);
1401     }
1402 
1403     virtio_net_set_queues(n);
1404 
1405     /* Find the first multicast entry in the saved MAC filter */
1406     for (i = 0; i < n->mac_table.in_use; i++) {
1407         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1408             break;
1409         }
1410     }
1411     n->mac_table.first_multi = i;
1412 
1413     /* nc.link_down can't be migrated, so infer link_down according
1414      * to link status bit in n->status */
1415     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1416     for (i = 0; i < n->max_queues; i++) {
1417         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1418     }
1419 
1420     return 0;
1421 }
1422 
1423 static void virtio_net_cleanup(NetClientState *nc)
1424 {
1425     VirtIONet *n = qemu_get_nic_opaque(nc);
1426 
1427     n->nic = NULL;
1428 }
1429 
1430 static NetClientInfo net_virtio_info = {
1431     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1432     .size = sizeof(NICState),
1433     .can_receive = virtio_net_can_receive,
1434     .receive = virtio_net_receive,
1435     .cleanup = virtio_net_cleanup,
1436     .link_status_changed = virtio_net_set_link_status,
1437     .query_rx_filter = virtio_net_query_rxfilter,
1438 };
1439 
1440 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1441 {
1442     VirtIONet *n = VIRTIO_NET(vdev);
1443     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1444     assert(n->vhost_started);
1445     return vhost_net_virtqueue_pending(tap_get_vhost_net(nc->peer), idx);
1446 }
1447 
1448 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1449                                            bool mask)
1450 {
1451     VirtIONet *n = VIRTIO_NET(vdev);
1452     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1453     assert(n->vhost_started);
1454     vhost_net_virtqueue_mask(tap_get_vhost_net(nc->peer),
1455                              vdev, idx, mask);
1456 }
1457 
1458 void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features)
1459 {
1460     int i, config_size = 0;
1461     host_features |= (1 << VIRTIO_NET_F_MAC);
1462     for (i = 0; feature_sizes[i].flags != 0; i++) {
1463         if (host_features & feature_sizes[i].flags) {
1464             config_size = MAX(feature_sizes[i].end, config_size);
1465         }
1466     }
1467     n->config_size = config_size;
1468 }
1469 
1470 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
1471                                    const char *type)
1472 {
1473     /*
1474      * The name can be NULL, the netclient name will be type.x.
1475      */
1476     assert(type != NULL);
1477 
1478     if (n->netclient_name) {
1479         g_free(n->netclient_name);
1480         n->netclient_name = NULL;
1481     }
1482     if (n->netclient_type) {
1483         g_free(n->netclient_type);
1484         n->netclient_type = NULL;
1485     }
1486 
1487     if (name != NULL) {
1488         n->netclient_name = g_strdup(name);
1489     }
1490     n->netclient_type = g_strdup(type);
1491 }
1492 
1493 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
1494 {
1495     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1496     VirtIONet *n = VIRTIO_NET(dev);
1497     NetClientState *nc;
1498     int i;
1499 
1500     virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
1501 
1502     n->max_queues = MAX(n->nic_conf.queues, 1);
1503     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1504     n->vqs[0].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1505     n->curr_queues = 1;
1506     n->vqs[0].n = n;
1507     n->tx_timeout = n->net_conf.txtimer;
1508 
1509     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1510                        && strcmp(n->net_conf.tx, "bh")) {
1511         error_report("virtio-net: "
1512                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1513                      n->net_conf.tx);
1514         error_report("Defaulting to \"bh\"");
1515     }
1516 
1517     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1518         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1519                                            virtio_net_handle_tx_timer);
1520         n->vqs[0].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, virtio_net_tx_timer,
1521                                                &n->vqs[0]);
1522     } else {
1523         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1524                                            virtio_net_handle_tx_bh);
1525         n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]);
1526     }
1527     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1528     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1529     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1530     n->status = VIRTIO_NET_S_LINK_UP;
1531 
1532     if (n->netclient_type) {
1533         /*
1534          * Happen when virtio_net_set_netclient_name has been called.
1535          */
1536         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1537                               n->netclient_type, n->netclient_name, n);
1538     } else {
1539         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1540                               object_get_typename(OBJECT(dev)), dev->id, n);
1541     }
1542 
1543     peer_test_vnet_hdr(n);
1544     if (peer_has_vnet_hdr(n)) {
1545         for (i = 0; i < n->max_queues; i++) {
1546             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1547         }
1548         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1549     } else {
1550         n->host_hdr_len = 0;
1551     }
1552 
1553     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1554 
1555     n->vqs[0].tx_waiting = 0;
1556     n->tx_burst = n->net_conf.txburst;
1557     virtio_net_set_mrg_rx_bufs(n, 0);
1558     n->promisc = 1; /* for compatibility */
1559 
1560     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1561 
1562     n->vlans = g_malloc0(MAX_VLAN >> 3);
1563 
1564     nc = qemu_get_queue(n->nic);
1565     nc->rxfilter_notify_enabled = 1;
1566 
1567     n->qdev = dev;
1568     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1569                     virtio_net_save, virtio_net_load, n);
1570 
1571     add_boot_device_path(n->nic_conf.bootindex, dev, "/ethernet-phy@0");
1572 }
1573 
1574 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
1575 {
1576     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1577     VirtIONet *n = VIRTIO_NET(dev);
1578     int i;
1579 
1580     /* This will stop vhost backend if appropriate. */
1581     virtio_net_set_status(vdev, 0);
1582 
1583     unregister_savevm(dev, "virtio-net", n);
1584 
1585     if (n->netclient_name) {
1586         g_free(n->netclient_name);
1587         n->netclient_name = NULL;
1588     }
1589     if (n->netclient_type) {
1590         g_free(n->netclient_type);
1591         n->netclient_type = NULL;
1592     }
1593 
1594     g_free(n->mac_table.macs);
1595     g_free(n->vlans);
1596 
1597     for (i = 0; i < n->max_queues; i++) {
1598         VirtIONetQueue *q = &n->vqs[i];
1599         NetClientState *nc = qemu_get_subqueue(n->nic, i);
1600 
1601         qemu_purge_queued_packets(nc);
1602 
1603         if (q->tx_timer) {
1604             timer_del(q->tx_timer);
1605             timer_free(q->tx_timer);
1606         } else if (q->tx_bh) {
1607             qemu_bh_delete(q->tx_bh);
1608         }
1609     }
1610 
1611     g_free(n->vqs);
1612     qemu_del_nic(n->nic);
1613     virtio_cleanup(vdev);
1614 }
1615 
1616 static void virtio_net_instance_init(Object *obj)
1617 {
1618     VirtIONet *n = VIRTIO_NET(obj);
1619 
1620     /*
1621      * The default config_size is sizeof(struct virtio_net_config).
1622      * Can be overriden with virtio_net_set_config_size.
1623      */
1624     n->config_size = sizeof(struct virtio_net_config);
1625 }
1626 
1627 static Property virtio_net_properties[] = {
1628     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
1629     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
1630                                                TX_TIMER_INTERVAL),
1631     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
1632     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
1633     DEFINE_PROP_END_OF_LIST(),
1634 };
1635 
1636 static void virtio_net_class_init(ObjectClass *klass, void *data)
1637 {
1638     DeviceClass *dc = DEVICE_CLASS(klass);
1639     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1640 
1641     dc->props = virtio_net_properties;
1642     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
1643     vdc->realize = virtio_net_device_realize;
1644     vdc->unrealize = virtio_net_device_unrealize;
1645     vdc->get_config = virtio_net_get_config;
1646     vdc->set_config = virtio_net_set_config;
1647     vdc->get_features = virtio_net_get_features;
1648     vdc->set_features = virtio_net_set_features;
1649     vdc->bad_features = virtio_net_bad_features;
1650     vdc->reset = virtio_net_reset;
1651     vdc->set_status = virtio_net_set_status;
1652     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
1653     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
1654 }
1655 
1656 static const TypeInfo virtio_net_info = {
1657     .name = TYPE_VIRTIO_NET,
1658     .parent = TYPE_VIRTIO_DEVICE,
1659     .instance_size = sizeof(VirtIONet),
1660     .instance_init = virtio_net_instance_init,
1661     .class_init = virtio_net_class_init,
1662 };
1663 
1664 static void virtio_register_types(void)
1665 {
1666     type_register_static(&virtio_net_info);
1667 }
1668 
1669 type_init(virtio_register_types)
1670