xref: /qemu/hw/net/virtio-net.c (revision 33848cee)
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/osdep.h"
15 #include "qemu/iov.h"
16 #include "hw/virtio/virtio.h"
17 #include "net/net.h"
18 #include "net/checksum.h"
19 #include "net/tap.h"
20 #include "qemu/error-report.h"
21 #include "qemu/timer.h"
22 #include "hw/virtio/virtio-net.h"
23 #include "net/vhost_net.h"
24 #include "hw/virtio/virtio-bus.h"
25 #include "qapi/qmp/qjson.h"
26 #include "qapi-event.h"
27 #include "hw/virtio/virtio-access.h"
28 
29 #define VIRTIO_NET_VM_VERSION    11
30 
31 #define MAC_TABLE_ENTRIES    64
32 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
33 
34 /* previously fixed value */
35 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
36 /* for now, only allow larger queues; with virtio-1, guest can downsize */
37 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
38 
39 /*
40  * Calculate the number of bytes up to and including the given 'field' of
41  * 'container'.
42  */
43 #define endof(container, field) \
44     (offsetof(container, field) + sizeof(((container *)0)->field))
45 
46 typedef struct VirtIOFeature {
47     uint32_t flags;
48     size_t end;
49 } VirtIOFeature;
50 
51 static VirtIOFeature feature_sizes[] = {
52     {.flags = 1 << VIRTIO_NET_F_MAC,
53      .end = endof(struct virtio_net_config, mac)},
54     {.flags = 1 << VIRTIO_NET_F_STATUS,
55      .end = endof(struct virtio_net_config, status)},
56     {.flags = 1 << VIRTIO_NET_F_MQ,
57      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
58     {}
59 };
60 
61 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
62 {
63     VirtIONet *n = qemu_get_nic_opaque(nc);
64 
65     return &n->vqs[nc->queue_index];
66 }
67 
68 static int vq2q(int queue_index)
69 {
70     return queue_index / 2;
71 }
72 
73 /* TODO
74  * - we could suppress RX interrupt if we were so inclined.
75  */
76 
77 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
78 {
79     VirtIONet *n = VIRTIO_NET(vdev);
80     struct virtio_net_config netcfg;
81 
82     virtio_stw_p(vdev, &netcfg.status, n->status);
83     virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues);
84     memcpy(netcfg.mac, n->mac, ETH_ALEN);
85     memcpy(config, &netcfg, n->config_size);
86 }
87 
88 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
89 {
90     VirtIONet *n = VIRTIO_NET(vdev);
91     struct virtio_net_config netcfg = {};
92 
93     memcpy(&netcfg, config, n->config_size);
94 
95     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
96         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
97         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
98         memcpy(n->mac, netcfg.mac, ETH_ALEN);
99         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
100     }
101 }
102 
103 static bool virtio_net_started(VirtIONet *n, uint8_t status)
104 {
105     VirtIODevice *vdev = VIRTIO_DEVICE(n);
106     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
107         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
108 }
109 
110 static void virtio_net_announce_timer(void *opaque)
111 {
112     VirtIONet *n = opaque;
113     VirtIODevice *vdev = VIRTIO_DEVICE(n);
114 
115     n->announce_counter--;
116     n->status |= VIRTIO_NET_S_ANNOUNCE;
117     virtio_notify_config(vdev);
118 }
119 
120 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
121 {
122     VirtIODevice *vdev = VIRTIO_DEVICE(n);
123     NetClientState *nc = qemu_get_queue(n->nic);
124     int queues = n->multiqueue ? n->max_queues : 1;
125 
126     if (!get_vhost_net(nc->peer)) {
127         return;
128     }
129 
130     if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
131         !!n->vhost_started) {
132         return;
133     }
134     if (!n->vhost_started) {
135         int r, i;
136 
137         if (n->needs_vnet_hdr_swap) {
138             error_report("backend does not support %s vnet headers; "
139                          "falling back on userspace virtio",
140                          virtio_is_big_endian(vdev) ? "BE" : "LE");
141             return;
142         }
143 
144         /* Any packets outstanding? Purge them to avoid touching rings
145          * when vhost is running.
146          */
147         for (i = 0;  i < queues; i++) {
148             NetClientState *qnc = qemu_get_subqueue(n->nic, i);
149 
150             /* Purge both directions: TX and RX. */
151             qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
152             qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
153         }
154 
155         n->vhost_started = 1;
156         r = vhost_net_start(vdev, n->nic->ncs, queues);
157         if (r < 0) {
158             error_report("unable to start vhost net: %d: "
159                          "falling back on userspace virtio", -r);
160             n->vhost_started = 0;
161         }
162     } else {
163         vhost_net_stop(vdev, n->nic->ncs, queues);
164         n->vhost_started = 0;
165     }
166 }
167 
168 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
169                                           NetClientState *peer,
170                                           bool enable)
171 {
172     if (virtio_is_big_endian(vdev)) {
173         return qemu_set_vnet_be(peer, enable);
174     } else {
175         return qemu_set_vnet_le(peer, enable);
176     }
177 }
178 
179 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
180                                        int queues, bool enable)
181 {
182     int i;
183 
184     for (i = 0; i < queues; i++) {
185         if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
186             enable) {
187             while (--i >= 0) {
188                 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
189             }
190 
191             return true;
192         }
193     }
194 
195     return false;
196 }
197 
198 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
199 {
200     VirtIODevice *vdev = VIRTIO_DEVICE(n);
201     int queues = n->multiqueue ? n->max_queues : 1;
202 
203     if (virtio_net_started(n, status)) {
204         /* Before using the device, we tell the network backend about the
205          * endianness to use when parsing vnet headers. If the backend
206          * can't do it, we fallback onto fixing the headers in the core
207          * virtio-net code.
208          */
209         n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
210                                                             queues, true);
211     } else if (virtio_net_started(n, vdev->status)) {
212         /* After using the device, we need to reset the network backend to
213          * the default (guest native endianness), otherwise the guest may
214          * lose network connectivity if it is rebooted into a different
215          * endianness.
216          */
217         virtio_net_set_vnet_endian(vdev, n->nic->ncs, queues, false);
218     }
219 }
220 
221 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
222 {
223     VirtIONet *n = VIRTIO_NET(vdev);
224     VirtIONetQueue *q;
225     int i;
226     uint8_t queue_status;
227 
228     virtio_net_vnet_endian_status(n, status);
229     virtio_net_vhost_status(n, status);
230 
231     for (i = 0; i < n->max_queues; i++) {
232         NetClientState *ncs = qemu_get_subqueue(n->nic, i);
233         bool queue_started;
234         q = &n->vqs[i];
235 
236         if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
237             queue_status = 0;
238         } else {
239             queue_status = status;
240         }
241         queue_started =
242             virtio_net_started(n, queue_status) && !n->vhost_started;
243 
244         if (queue_started) {
245             qemu_flush_queued_packets(ncs);
246         }
247 
248         if (!q->tx_waiting) {
249             continue;
250         }
251 
252         if (queue_started) {
253             if (q->tx_timer) {
254                 timer_mod(q->tx_timer,
255                                qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
256             } else {
257                 qemu_bh_schedule(q->tx_bh);
258             }
259         } else {
260             if (q->tx_timer) {
261                 timer_del(q->tx_timer);
262             } else {
263                 qemu_bh_cancel(q->tx_bh);
264             }
265         }
266     }
267 }
268 
269 static void virtio_net_set_link_status(NetClientState *nc)
270 {
271     VirtIONet *n = qemu_get_nic_opaque(nc);
272     VirtIODevice *vdev = VIRTIO_DEVICE(n);
273     uint16_t old_status = n->status;
274 
275     if (nc->link_down)
276         n->status &= ~VIRTIO_NET_S_LINK_UP;
277     else
278         n->status |= VIRTIO_NET_S_LINK_UP;
279 
280     if (n->status != old_status)
281         virtio_notify_config(vdev);
282 
283     virtio_net_set_status(vdev, vdev->status);
284 }
285 
286 static void rxfilter_notify(NetClientState *nc)
287 {
288     VirtIONet *n = qemu_get_nic_opaque(nc);
289 
290     if (nc->rxfilter_notify_enabled) {
291         gchar *path = object_get_canonical_path(OBJECT(n->qdev));
292         qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
293                                               n->netclient_name, path, &error_abort);
294         g_free(path);
295 
296         /* disable event notification to avoid events flooding */
297         nc->rxfilter_notify_enabled = 0;
298     }
299 }
300 
301 static intList *get_vlan_table(VirtIONet *n)
302 {
303     intList *list, *entry;
304     int i, j;
305 
306     list = NULL;
307     for (i = 0; i < MAX_VLAN >> 5; i++) {
308         for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
309             if (n->vlans[i] & (1U << j)) {
310                 entry = g_malloc0(sizeof(*entry));
311                 entry->value = (i << 5) + j;
312                 entry->next = list;
313                 list = entry;
314             }
315         }
316     }
317 
318     return list;
319 }
320 
321 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
322 {
323     VirtIONet *n = qemu_get_nic_opaque(nc);
324     VirtIODevice *vdev = VIRTIO_DEVICE(n);
325     RxFilterInfo *info;
326     strList *str_list, *entry;
327     int i;
328 
329     info = g_malloc0(sizeof(*info));
330     info->name = g_strdup(nc->name);
331     info->promiscuous = n->promisc;
332 
333     if (n->nouni) {
334         info->unicast = RX_STATE_NONE;
335     } else if (n->alluni) {
336         info->unicast = RX_STATE_ALL;
337     } else {
338         info->unicast = RX_STATE_NORMAL;
339     }
340 
341     if (n->nomulti) {
342         info->multicast = RX_STATE_NONE;
343     } else if (n->allmulti) {
344         info->multicast = RX_STATE_ALL;
345     } else {
346         info->multicast = RX_STATE_NORMAL;
347     }
348 
349     info->broadcast_allowed = n->nobcast;
350     info->multicast_overflow = n->mac_table.multi_overflow;
351     info->unicast_overflow = n->mac_table.uni_overflow;
352 
353     info->main_mac = qemu_mac_strdup_printf(n->mac);
354 
355     str_list = NULL;
356     for (i = 0; i < n->mac_table.first_multi; i++) {
357         entry = g_malloc0(sizeof(*entry));
358         entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
359         entry->next = str_list;
360         str_list = entry;
361     }
362     info->unicast_table = str_list;
363 
364     str_list = NULL;
365     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
366         entry = g_malloc0(sizeof(*entry));
367         entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
368         entry->next = str_list;
369         str_list = entry;
370     }
371     info->multicast_table = str_list;
372     info->vlan_table = get_vlan_table(n);
373 
374     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
375         info->vlan = RX_STATE_ALL;
376     } else if (!info->vlan_table) {
377         info->vlan = RX_STATE_NONE;
378     } else {
379         info->vlan = RX_STATE_NORMAL;
380     }
381 
382     /* enable event notification after query */
383     nc->rxfilter_notify_enabled = 1;
384 
385     return info;
386 }
387 
388 static void virtio_net_reset(VirtIODevice *vdev)
389 {
390     VirtIONet *n = VIRTIO_NET(vdev);
391 
392     /* Reset back to compatibility mode */
393     n->promisc = 1;
394     n->allmulti = 0;
395     n->alluni = 0;
396     n->nomulti = 0;
397     n->nouni = 0;
398     n->nobcast = 0;
399     /* multiqueue is disabled by default */
400     n->curr_queues = 1;
401     timer_del(n->announce_timer);
402     n->announce_counter = 0;
403     n->status &= ~VIRTIO_NET_S_ANNOUNCE;
404 
405     /* Flush any MAC and VLAN filter table state */
406     n->mac_table.in_use = 0;
407     n->mac_table.first_multi = 0;
408     n->mac_table.multi_overflow = 0;
409     n->mac_table.uni_overflow = 0;
410     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
411     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
412     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
413     memset(n->vlans, 0, MAX_VLAN >> 3);
414 }
415 
416 static void peer_test_vnet_hdr(VirtIONet *n)
417 {
418     NetClientState *nc = qemu_get_queue(n->nic);
419     if (!nc->peer) {
420         return;
421     }
422 
423     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
424 }
425 
426 static int peer_has_vnet_hdr(VirtIONet *n)
427 {
428     return n->has_vnet_hdr;
429 }
430 
431 static int peer_has_ufo(VirtIONet *n)
432 {
433     if (!peer_has_vnet_hdr(n))
434         return 0;
435 
436     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
437 
438     return n->has_ufo;
439 }
440 
441 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
442                                        int version_1)
443 {
444     int i;
445     NetClientState *nc;
446 
447     n->mergeable_rx_bufs = mergeable_rx_bufs;
448 
449     if (version_1) {
450         n->guest_hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
451     } else {
452         n->guest_hdr_len = n->mergeable_rx_bufs ?
453             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
454             sizeof(struct virtio_net_hdr);
455     }
456 
457     for (i = 0; i < n->max_queues; i++) {
458         nc = qemu_get_subqueue(n->nic, i);
459 
460         if (peer_has_vnet_hdr(n) &&
461             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
462             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
463             n->host_hdr_len = n->guest_hdr_len;
464         }
465     }
466 }
467 
468 static int peer_attach(VirtIONet *n, int index)
469 {
470     NetClientState *nc = qemu_get_subqueue(n->nic, index);
471 
472     if (!nc->peer) {
473         return 0;
474     }
475 
476     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
477         vhost_set_vring_enable(nc->peer, 1);
478     }
479 
480     if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
481         return 0;
482     }
483 
484     return tap_enable(nc->peer);
485 }
486 
487 static int peer_detach(VirtIONet *n, int index)
488 {
489     NetClientState *nc = qemu_get_subqueue(n->nic, index);
490 
491     if (!nc->peer) {
492         return 0;
493     }
494 
495     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
496         vhost_set_vring_enable(nc->peer, 0);
497     }
498 
499     if (nc->peer->info->type !=  NET_CLIENT_DRIVER_TAP) {
500         return 0;
501     }
502 
503     return tap_disable(nc->peer);
504 }
505 
506 static void virtio_net_set_queues(VirtIONet *n)
507 {
508     int i;
509     int r;
510 
511     if (n->nic->peer_deleted) {
512         return;
513     }
514 
515     for (i = 0; i < n->max_queues; i++) {
516         if (i < n->curr_queues) {
517             r = peer_attach(n, i);
518             assert(!r);
519         } else {
520             r = peer_detach(n, i);
521             assert(!r);
522         }
523     }
524 }
525 
526 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
527 
528 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
529                                         Error **errp)
530 {
531     VirtIONet *n = VIRTIO_NET(vdev);
532     NetClientState *nc = qemu_get_queue(n->nic);
533 
534     /* Firstly sync all virtio-net possible supported features */
535     features |= n->host_features;
536 
537     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
538 
539     if (!peer_has_vnet_hdr(n)) {
540         virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
541         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
542         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
543         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
544 
545         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
546         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
547         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
548         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
549     }
550 
551     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
552         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
553         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
554     }
555 
556     if (!get_vhost_net(nc->peer)) {
557         return features;
558     }
559     return vhost_net_get_features(get_vhost_net(nc->peer), features);
560 }
561 
562 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
563 {
564     uint64_t features = 0;
565 
566     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
567      * but also these: */
568     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
569     virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
570     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
571     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
572     virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
573 
574     return features;
575 }
576 
577 static void virtio_net_apply_guest_offloads(VirtIONet *n)
578 {
579     qemu_set_offload(qemu_get_queue(n->nic)->peer,
580             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
581             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
582             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
583             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
584             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
585 }
586 
587 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
588 {
589     static const uint64_t guest_offloads_mask =
590         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
591         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
592         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
593         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
594         (1ULL << VIRTIO_NET_F_GUEST_UFO);
595 
596     return guest_offloads_mask & features;
597 }
598 
599 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
600 {
601     VirtIODevice *vdev = VIRTIO_DEVICE(n);
602     return virtio_net_guest_offloads_by_features(vdev->guest_features);
603 }
604 
605 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
606 {
607     VirtIONet *n = VIRTIO_NET(vdev);
608     int i;
609 
610     virtio_net_set_multiqueue(n,
611                               virtio_has_feature(features, VIRTIO_NET_F_MQ));
612 
613     virtio_net_set_mrg_rx_bufs(n,
614                                virtio_has_feature(features,
615                                                   VIRTIO_NET_F_MRG_RXBUF),
616                                virtio_has_feature(features,
617                                                   VIRTIO_F_VERSION_1));
618 
619     if (n->has_vnet_hdr) {
620         n->curr_guest_offloads =
621             virtio_net_guest_offloads_by_features(features);
622         virtio_net_apply_guest_offloads(n);
623     }
624 
625     for (i = 0;  i < n->max_queues; i++) {
626         NetClientState *nc = qemu_get_subqueue(n->nic, i);
627 
628         if (!get_vhost_net(nc->peer)) {
629             continue;
630         }
631         vhost_net_ack_features(get_vhost_net(nc->peer), features);
632     }
633 
634     if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
635         memset(n->vlans, 0, MAX_VLAN >> 3);
636     } else {
637         memset(n->vlans, 0xff, MAX_VLAN >> 3);
638     }
639 }
640 
641 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
642                                      struct iovec *iov, unsigned int iov_cnt)
643 {
644     uint8_t on;
645     size_t s;
646     NetClientState *nc = qemu_get_queue(n->nic);
647 
648     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
649     if (s != sizeof(on)) {
650         return VIRTIO_NET_ERR;
651     }
652 
653     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
654         n->promisc = on;
655     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
656         n->allmulti = on;
657     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
658         n->alluni = on;
659     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
660         n->nomulti = on;
661     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
662         n->nouni = on;
663     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
664         n->nobcast = on;
665     } else {
666         return VIRTIO_NET_ERR;
667     }
668 
669     rxfilter_notify(nc);
670 
671     return VIRTIO_NET_OK;
672 }
673 
674 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
675                                      struct iovec *iov, unsigned int iov_cnt)
676 {
677     VirtIODevice *vdev = VIRTIO_DEVICE(n);
678     uint64_t offloads;
679     size_t s;
680 
681     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
682         return VIRTIO_NET_ERR;
683     }
684 
685     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
686     if (s != sizeof(offloads)) {
687         return VIRTIO_NET_ERR;
688     }
689 
690     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
691         uint64_t supported_offloads;
692 
693         if (!n->has_vnet_hdr) {
694             return VIRTIO_NET_ERR;
695         }
696 
697         supported_offloads = virtio_net_supported_guest_offloads(n);
698         if (offloads & ~supported_offloads) {
699             return VIRTIO_NET_ERR;
700         }
701 
702         n->curr_guest_offloads = offloads;
703         virtio_net_apply_guest_offloads(n);
704 
705         return VIRTIO_NET_OK;
706     } else {
707         return VIRTIO_NET_ERR;
708     }
709 }
710 
711 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
712                                  struct iovec *iov, unsigned int iov_cnt)
713 {
714     VirtIODevice *vdev = VIRTIO_DEVICE(n);
715     struct virtio_net_ctrl_mac mac_data;
716     size_t s;
717     NetClientState *nc = qemu_get_queue(n->nic);
718 
719     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
720         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
721             return VIRTIO_NET_ERR;
722         }
723         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
724         assert(s == sizeof(n->mac));
725         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
726         rxfilter_notify(nc);
727 
728         return VIRTIO_NET_OK;
729     }
730 
731     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
732         return VIRTIO_NET_ERR;
733     }
734 
735     int in_use = 0;
736     int first_multi = 0;
737     uint8_t uni_overflow = 0;
738     uint8_t multi_overflow = 0;
739     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
740 
741     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
742                    sizeof(mac_data.entries));
743     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
744     if (s != sizeof(mac_data.entries)) {
745         goto error;
746     }
747     iov_discard_front(&iov, &iov_cnt, s);
748 
749     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
750         goto error;
751     }
752 
753     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
754         s = iov_to_buf(iov, iov_cnt, 0, macs,
755                        mac_data.entries * ETH_ALEN);
756         if (s != mac_data.entries * ETH_ALEN) {
757             goto error;
758         }
759         in_use += mac_data.entries;
760     } else {
761         uni_overflow = 1;
762     }
763 
764     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
765 
766     first_multi = in_use;
767 
768     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
769                    sizeof(mac_data.entries));
770     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
771     if (s != sizeof(mac_data.entries)) {
772         goto error;
773     }
774 
775     iov_discard_front(&iov, &iov_cnt, s);
776 
777     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
778         goto error;
779     }
780 
781     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
782         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
783                        mac_data.entries * ETH_ALEN);
784         if (s != mac_data.entries * ETH_ALEN) {
785             goto error;
786         }
787         in_use += mac_data.entries;
788     } else {
789         multi_overflow = 1;
790     }
791 
792     n->mac_table.in_use = in_use;
793     n->mac_table.first_multi = first_multi;
794     n->mac_table.uni_overflow = uni_overflow;
795     n->mac_table.multi_overflow = multi_overflow;
796     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
797     g_free(macs);
798     rxfilter_notify(nc);
799 
800     return VIRTIO_NET_OK;
801 
802 error:
803     g_free(macs);
804     return VIRTIO_NET_ERR;
805 }
806 
807 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
808                                         struct iovec *iov, unsigned int iov_cnt)
809 {
810     VirtIODevice *vdev = VIRTIO_DEVICE(n);
811     uint16_t vid;
812     size_t s;
813     NetClientState *nc = qemu_get_queue(n->nic);
814 
815     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
816     vid = virtio_lduw_p(vdev, &vid);
817     if (s != sizeof(vid)) {
818         return VIRTIO_NET_ERR;
819     }
820 
821     if (vid >= MAX_VLAN)
822         return VIRTIO_NET_ERR;
823 
824     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
825         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
826     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
827         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
828     else
829         return VIRTIO_NET_ERR;
830 
831     rxfilter_notify(nc);
832 
833     return VIRTIO_NET_OK;
834 }
835 
836 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
837                                       struct iovec *iov, unsigned int iov_cnt)
838 {
839     if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
840         n->status & VIRTIO_NET_S_ANNOUNCE) {
841         n->status &= ~VIRTIO_NET_S_ANNOUNCE;
842         if (n->announce_counter) {
843             timer_mod(n->announce_timer,
844                       qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
845                       self_announce_delay(n->announce_counter));
846         }
847         return VIRTIO_NET_OK;
848     } else {
849         return VIRTIO_NET_ERR;
850     }
851 }
852 
853 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
854                                 struct iovec *iov, unsigned int iov_cnt)
855 {
856     VirtIODevice *vdev = VIRTIO_DEVICE(n);
857     struct virtio_net_ctrl_mq mq;
858     size_t s;
859     uint16_t queues;
860 
861     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
862     if (s != sizeof(mq)) {
863         return VIRTIO_NET_ERR;
864     }
865 
866     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
867         return VIRTIO_NET_ERR;
868     }
869 
870     queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
871 
872     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
873         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
874         queues > n->max_queues ||
875         !n->multiqueue) {
876         return VIRTIO_NET_ERR;
877     }
878 
879     n->curr_queues = queues;
880     /* stop the backend before changing the number of queues to avoid handling a
881      * disabled queue */
882     virtio_net_set_status(vdev, vdev->status);
883     virtio_net_set_queues(n);
884 
885     return VIRTIO_NET_OK;
886 }
887 
888 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
889 {
890     VirtIONet *n = VIRTIO_NET(vdev);
891     struct virtio_net_ctrl_hdr ctrl;
892     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
893     VirtQueueElement *elem;
894     size_t s;
895     struct iovec *iov, *iov2;
896     unsigned int iov_cnt;
897 
898     for (;;) {
899         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
900         if (!elem) {
901             break;
902         }
903         if (iov_size(elem->in_sg, elem->in_num) < sizeof(status) ||
904             iov_size(elem->out_sg, elem->out_num) < sizeof(ctrl)) {
905             virtio_error(vdev, "virtio-net ctrl missing headers");
906             virtqueue_detach_element(vq, elem, 0);
907             g_free(elem);
908             break;
909         }
910 
911         iov_cnt = elem->out_num;
912         iov2 = iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num);
913         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
914         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
915         if (s != sizeof(ctrl)) {
916             status = VIRTIO_NET_ERR;
917         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
918             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
919         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
920             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
921         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
922             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
923         } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
924             status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
925         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
926             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
927         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
928             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
929         }
930 
931         s = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, sizeof(status));
932         assert(s == sizeof(status));
933 
934         virtqueue_push(vq, elem, sizeof(status));
935         virtio_notify(vdev, vq);
936         g_free(iov2);
937         g_free(elem);
938     }
939 }
940 
941 /* RX */
942 
943 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
944 {
945     VirtIONet *n = VIRTIO_NET(vdev);
946     int queue_index = vq2q(virtio_get_queue_index(vq));
947 
948     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
949 }
950 
951 static int virtio_net_can_receive(NetClientState *nc)
952 {
953     VirtIONet *n = qemu_get_nic_opaque(nc);
954     VirtIODevice *vdev = VIRTIO_DEVICE(n);
955     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
956 
957     if (!vdev->vm_running) {
958         return 0;
959     }
960 
961     if (nc->queue_index >= n->curr_queues) {
962         return 0;
963     }
964 
965     if (!virtio_queue_ready(q->rx_vq) ||
966         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
967         return 0;
968     }
969 
970     return 1;
971 }
972 
973 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
974 {
975     VirtIONet *n = q->n;
976     if (virtio_queue_empty(q->rx_vq) ||
977         (n->mergeable_rx_bufs &&
978          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
979         virtio_queue_set_notification(q->rx_vq, 1);
980 
981         /* To avoid a race condition where the guest has made some buffers
982          * available after the above check but before notification was
983          * enabled, check for available buffers again.
984          */
985         if (virtio_queue_empty(q->rx_vq) ||
986             (n->mergeable_rx_bufs &&
987              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
988             return 0;
989         }
990     }
991 
992     virtio_queue_set_notification(q->rx_vq, 0);
993     return 1;
994 }
995 
996 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
997 {
998     virtio_tswap16s(vdev, &hdr->hdr_len);
999     virtio_tswap16s(vdev, &hdr->gso_size);
1000     virtio_tswap16s(vdev, &hdr->csum_start);
1001     virtio_tswap16s(vdev, &hdr->csum_offset);
1002 }
1003 
1004 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1005  * it never finds out that the packets don't have valid checksums.  This
1006  * causes dhclient to get upset.  Fedora's carried a patch for ages to
1007  * fix this with Xen but it hasn't appeared in an upstream release of
1008  * dhclient yet.
1009  *
1010  * To avoid breaking existing guests, we catch udp packets and add
1011  * checksums.  This is terrible but it's better than hacking the guest
1012  * kernels.
1013  *
1014  * N.B. if we introduce a zero-copy API, this operation is no longer free so
1015  * we should provide a mechanism to disable it to avoid polluting the host
1016  * cache.
1017  */
1018 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1019                                         uint8_t *buf, size_t size)
1020 {
1021     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1022         (size > 27 && size < 1500) && /* normal sized MTU */
1023         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1024         (buf[23] == 17) && /* ip.protocol == UDP */
1025         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1026         net_checksum_calculate(buf, size);
1027         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1028     }
1029 }
1030 
1031 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1032                            const void *buf, size_t size)
1033 {
1034     if (n->has_vnet_hdr) {
1035         /* FIXME this cast is evil */
1036         void *wbuf = (void *)buf;
1037         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1038                                     size - n->host_hdr_len);
1039 
1040         if (n->needs_vnet_hdr_swap) {
1041             virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1042         }
1043         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1044     } else {
1045         struct virtio_net_hdr hdr = {
1046             .flags = 0,
1047             .gso_type = VIRTIO_NET_HDR_GSO_NONE
1048         };
1049         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1050     }
1051 }
1052 
1053 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1054 {
1055     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1056     static const uint8_t vlan[] = {0x81, 0x00};
1057     uint8_t *ptr = (uint8_t *)buf;
1058     int i;
1059 
1060     if (n->promisc)
1061         return 1;
1062 
1063     ptr += n->host_hdr_len;
1064 
1065     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1066         int vid = lduw_be_p(ptr + 14) & 0xfff;
1067         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1068             return 0;
1069     }
1070 
1071     if (ptr[0] & 1) { // multicast
1072         if (!memcmp(ptr, bcast, sizeof(bcast))) {
1073             return !n->nobcast;
1074         } else if (n->nomulti) {
1075             return 0;
1076         } else if (n->allmulti || n->mac_table.multi_overflow) {
1077             return 1;
1078         }
1079 
1080         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1081             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1082                 return 1;
1083             }
1084         }
1085     } else { // unicast
1086         if (n->nouni) {
1087             return 0;
1088         } else if (n->alluni || n->mac_table.uni_overflow) {
1089             return 1;
1090         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1091             return 1;
1092         }
1093 
1094         for (i = 0; i < n->mac_table.first_multi; i++) {
1095             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1096                 return 1;
1097             }
1098         }
1099     }
1100 
1101     return 0;
1102 }
1103 
1104 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
1105 {
1106     VirtIONet *n = qemu_get_nic_opaque(nc);
1107     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1108     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1109     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1110     struct virtio_net_hdr_mrg_rxbuf mhdr;
1111     unsigned mhdr_cnt = 0;
1112     size_t offset, i, guest_offset;
1113 
1114     if (!virtio_net_can_receive(nc)) {
1115         return -1;
1116     }
1117 
1118     /* hdr_len refers to the header we supply to the guest */
1119     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1120         return 0;
1121     }
1122 
1123     if (!receive_filter(n, buf, size))
1124         return size;
1125 
1126     offset = i = 0;
1127 
1128     while (offset < size) {
1129         VirtQueueElement *elem;
1130         int len, total;
1131         const struct iovec *sg;
1132 
1133         total = 0;
1134 
1135         elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1136         if (!elem) {
1137             if (i) {
1138                 virtio_error(vdev, "virtio-net unexpected empty queue: "
1139                              "i %zd mergeable %d offset %zd, size %zd, "
1140                              "guest hdr len %zd, host hdr len %zd "
1141                              "guest features 0x%" PRIx64,
1142                              i, n->mergeable_rx_bufs, offset, size,
1143                              n->guest_hdr_len, n->host_hdr_len,
1144                              vdev->guest_features);
1145             }
1146             return -1;
1147         }
1148 
1149         if (elem->in_num < 1) {
1150             virtio_error(vdev,
1151                          "virtio-net receive queue contains no in buffers");
1152             virtqueue_detach_element(q->rx_vq, elem, 0);
1153             g_free(elem);
1154             return -1;
1155         }
1156 
1157         sg = elem->in_sg;
1158         if (i == 0) {
1159             assert(offset == 0);
1160             if (n->mergeable_rx_bufs) {
1161                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1162                                     sg, elem->in_num,
1163                                     offsetof(typeof(mhdr), num_buffers),
1164                                     sizeof(mhdr.num_buffers));
1165             }
1166 
1167             receive_header(n, sg, elem->in_num, buf, size);
1168             offset = n->host_hdr_len;
1169             total += n->guest_hdr_len;
1170             guest_offset = n->guest_hdr_len;
1171         } else {
1172             guest_offset = 0;
1173         }
1174 
1175         /* copy in packet.  ugh */
1176         len = iov_from_buf(sg, elem->in_num, guest_offset,
1177                            buf + offset, size - offset);
1178         total += len;
1179         offset += len;
1180         /* If buffers can't be merged, at this point we
1181          * must have consumed the complete packet.
1182          * Otherwise, drop it. */
1183         if (!n->mergeable_rx_bufs && offset < size) {
1184             virtqueue_unpop(q->rx_vq, elem, total);
1185             g_free(elem);
1186             return size;
1187         }
1188 
1189         /* signal other side */
1190         virtqueue_fill(q->rx_vq, elem, total, i++);
1191         g_free(elem);
1192     }
1193 
1194     if (mhdr_cnt) {
1195         virtio_stw_p(vdev, &mhdr.num_buffers, i);
1196         iov_from_buf(mhdr_sg, mhdr_cnt,
1197                      0,
1198                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1199     }
1200 
1201     virtqueue_flush(q->rx_vq, i);
1202     virtio_notify(vdev, q->rx_vq);
1203 
1204     return size;
1205 }
1206 
1207 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1208 
1209 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1210 {
1211     VirtIONet *n = qemu_get_nic_opaque(nc);
1212     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1213     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1214 
1215     virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
1216     virtio_notify(vdev, q->tx_vq);
1217 
1218     g_free(q->async_tx.elem);
1219     q->async_tx.elem = NULL;
1220 
1221     virtio_queue_set_notification(q->tx_vq, 1);
1222     virtio_net_flush_tx(q);
1223 }
1224 
1225 /* TX */
1226 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
1227 {
1228     VirtIONet *n = q->n;
1229     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1230     VirtQueueElement *elem;
1231     int32_t num_packets = 0;
1232     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
1233     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1234         return num_packets;
1235     }
1236 
1237     if (q->async_tx.elem) {
1238         virtio_queue_set_notification(q->tx_vq, 0);
1239         return num_packets;
1240     }
1241 
1242     for (;;) {
1243         ssize_t ret;
1244         unsigned int out_num;
1245         struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
1246         struct virtio_net_hdr_mrg_rxbuf mhdr;
1247 
1248         elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
1249         if (!elem) {
1250             break;
1251         }
1252 
1253         out_num = elem->out_num;
1254         out_sg = elem->out_sg;
1255         if (out_num < 1) {
1256             virtio_error(vdev, "virtio-net header not in first element");
1257             virtqueue_detach_element(q->tx_vq, elem, 0);
1258             g_free(elem);
1259             return -EINVAL;
1260         }
1261 
1262         if (n->has_vnet_hdr) {
1263             if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
1264                 n->guest_hdr_len) {
1265                 virtio_error(vdev, "virtio-net header incorrect");
1266                 virtqueue_detach_element(q->tx_vq, elem, 0);
1267                 g_free(elem);
1268                 return -EINVAL;
1269             }
1270             if (n->needs_vnet_hdr_swap) {
1271                 virtio_net_hdr_swap(vdev, (void *) &mhdr);
1272                 sg2[0].iov_base = &mhdr;
1273                 sg2[0].iov_len = n->guest_hdr_len;
1274                 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
1275                                    out_sg, out_num,
1276                                    n->guest_hdr_len, -1);
1277                 if (out_num == VIRTQUEUE_MAX_SIZE) {
1278                     goto drop;
1279 		}
1280                 out_num += 1;
1281                 out_sg = sg2;
1282 	    }
1283         }
1284         /*
1285          * If host wants to see the guest header as is, we can
1286          * pass it on unchanged. Otherwise, copy just the parts
1287          * that host is interested in.
1288          */
1289         assert(n->host_hdr_len <= n->guest_hdr_len);
1290         if (n->host_hdr_len != n->guest_hdr_len) {
1291             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
1292                                        out_sg, out_num,
1293                                        0, n->host_hdr_len);
1294             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
1295                              out_sg, out_num,
1296                              n->guest_hdr_len, -1);
1297             out_num = sg_num;
1298             out_sg = sg;
1299         }
1300 
1301         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
1302                                       out_sg, out_num, virtio_net_tx_complete);
1303         if (ret == 0) {
1304             virtio_queue_set_notification(q->tx_vq, 0);
1305             q->async_tx.elem = elem;
1306             return -EBUSY;
1307         }
1308 
1309 drop:
1310         virtqueue_push(q->tx_vq, elem, 0);
1311         virtio_notify(vdev, q->tx_vq);
1312         g_free(elem);
1313 
1314         if (++num_packets >= n->tx_burst) {
1315             break;
1316         }
1317     }
1318     return num_packets;
1319 }
1320 
1321 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
1322 {
1323     VirtIONet *n = VIRTIO_NET(vdev);
1324     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1325 
1326     /* This happens when device was stopped but VCPU wasn't. */
1327     if (!vdev->vm_running) {
1328         q->tx_waiting = 1;
1329         return;
1330     }
1331 
1332     if (q->tx_waiting) {
1333         virtio_queue_set_notification(vq, 1);
1334         timer_del(q->tx_timer);
1335         q->tx_waiting = 0;
1336         if (virtio_net_flush_tx(q) == -EINVAL) {
1337             return;
1338         }
1339     } else {
1340         timer_mod(q->tx_timer,
1341                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
1342         q->tx_waiting = 1;
1343         virtio_queue_set_notification(vq, 0);
1344     }
1345 }
1346 
1347 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1348 {
1349     VirtIONet *n = VIRTIO_NET(vdev);
1350     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1351 
1352     if (unlikely(q->tx_waiting)) {
1353         return;
1354     }
1355     q->tx_waiting = 1;
1356     /* This happens when device was stopped but VCPU wasn't. */
1357     if (!vdev->vm_running) {
1358         return;
1359     }
1360     virtio_queue_set_notification(vq, 0);
1361     qemu_bh_schedule(q->tx_bh);
1362 }
1363 
1364 static void virtio_net_tx_timer(void *opaque)
1365 {
1366     VirtIONetQueue *q = opaque;
1367     VirtIONet *n = q->n;
1368     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1369     /* This happens when device was stopped but BH wasn't. */
1370     if (!vdev->vm_running) {
1371         /* Make sure tx waiting is set, so we'll run when restarted. */
1372         assert(q->tx_waiting);
1373         return;
1374     }
1375 
1376     q->tx_waiting = 0;
1377 
1378     /* Just in case the driver is not ready on more */
1379     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1380         return;
1381     }
1382 
1383     virtio_queue_set_notification(q->tx_vq, 1);
1384     virtio_net_flush_tx(q);
1385 }
1386 
1387 static void virtio_net_tx_bh(void *opaque)
1388 {
1389     VirtIONetQueue *q = opaque;
1390     VirtIONet *n = q->n;
1391     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1392     int32_t ret;
1393 
1394     /* This happens when device was stopped but BH wasn't. */
1395     if (!vdev->vm_running) {
1396         /* Make sure tx waiting is set, so we'll run when restarted. */
1397         assert(q->tx_waiting);
1398         return;
1399     }
1400 
1401     q->tx_waiting = 0;
1402 
1403     /* Just in case the driver is not ready on more */
1404     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1405         return;
1406     }
1407 
1408     ret = virtio_net_flush_tx(q);
1409     if (ret == -EBUSY || ret == -EINVAL) {
1410         return; /* Notification re-enable handled by tx_complete or device
1411                  * broken */
1412     }
1413 
1414     /* If we flush a full burst of packets, assume there are
1415      * more coming and immediately reschedule */
1416     if (ret >= n->tx_burst) {
1417         qemu_bh_schedule(q->tx_bh);
1418         q->tx_waiting = 1;
1419         return;
1420     }
1421 
1422     /* If less than a full burst, re-enable notification and flush
1423      * anything that may have come in while we weren't looking.  If
1424      * we find something, assume the guest is still active and reschedule */
1425     virtio_queue_set_notification(q->tx_vq, 1);
1426     ret = virtio_net_flush_tx(q);
1427     if (ret == -EINVAL) {
1428         return;
1429     } else if (ret > 0) {
1430         virtio_queue_set_notification(q->tx_vq, 0);
1431         qemu_bh_schedule(q->tx_bh);
1432         q->tx_waiting = 1;
1433     }
1434 }
1435 
1436 static void virtio_net_add_queue(VirtIONet *n, int index)
1437 {
1438     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1439 
1440     n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
1441                                            virtio_net_handle_rx);
1442     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1443         n->vqs[index].tx_vq =
1444             virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1445         n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
1446                                               virtio_net_tx_timer,
1447                                               &n->vqs[index]);
1448     } else {
1449         n->vqs[index].tx_vq =
1450             virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1451         n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
1452     }
1453 
1454     n->vqs[index].tx_waiting = 0;
1455     n->vqs[index].n = n;
1456 }
1457 
1458 static void virtio_net_del_queue(VirtIONet *n, int index)
1459 {
1460     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1461     VirtIONetQueue *q = &n->vqs[index];
1462     NetClientState *nc = qemu_get_subqueue(n->nic, index);
1463 
1464     qemu_purge_queued_packets(nc);
1465 
1466     virtio_del_queue(vdev, index * 2);
1467     if (q->tx_timer) {
1468         timer_del(q->tx_timer);
1469         timer_free(q->tx_timer);
1470     } else {
1471         qemu_bh_delete(q->tx_bh);
1472     }
1473     virtio_del_queue(vdev, index * 2 + 1);
1474 }
1475 
1476 static void virtio_net_change_num_queues(VirtIONet *n, int new_max_queues)
1477 {
1478     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1479     int old_num_queues = virtio_get_num_queues(vdev);
1480     int new_num_queues = new_max_queues * 2 + 1;
1481     int i;
1482 
1483     assert(old_num_queues >= 3);
1484     assert(old_num_queues % 2 == 1);
1485 
1486     if (old_num_queues == new_num_queues) {
1487         return;
1488     }
1489 
1490     /*
1491      * We always need to remove and add ctrl vq if
1492      * old_num_queues != new_num_queues. Remove ctrl_vq first,
1493      * and then we only enter one of the following too loops.
1494      */
1495     virtio_del_queue(vdev, old_num_queues - 1);
1496 
1497     for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
1498         /* new_num_queues < old_num_queues */
1499         virtio_net_del_queue(n, i / 2);
1500     }
1501 
1502     for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
1503         /* new_num_queues > old_num_queues */
1504         virtio_net_add_queue(n, i / 2);
1505     }
1506 
1507     /* add ctrl_vq last */
1508     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1509 }
1510 
1511 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1512 {
1513     int max = multiqueue ? n->max_queues : 1;
1514 
1515     n->multiqueue = multiqueue;
1516     virtio_net_change_num_queues(n, max);
1517 
1518     virtio_net_set_queues(n);
1519 }
1520 
1521 static void virtio_net_save_device(VirtIODevice *vdev, QEMUFile *f)
1522 {
1523     VirtIONet *n = VIRTIO_NET(vdev);
1524     int i;
1525 
1526     qemu_put_buffer(f, n->mac, ETH_ALEN);
1527     qemu_put_be32(f, n->vqs[0].tx_waiting);
1528     qemu_put_be32(f, n->mergeable_rx_bufs);
1529     qemu_put_be16(f, n->status);
1530     qemu_put_byte(f, n->promisc);
1531     qemu_put_byte(f, n->allmulti);
1532     qemu_put_be32(f, n->mac_table.in_use);
1533     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1534     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1535     qemu_put_be32(f, n->has_vnet_hdr);
1536     qemu_put_byte(f, n->mac_table.multi_overflow);
1537     qemu_put_byte(f, n->mac_table.uni_overflow);
1538     qemu_put_byte(f, n->alluni);
1539     qemu_put_byte(f, n->nomulti);
1540     qemu_put_byte(f, n->nouni);
1541     qemu_put_byte(f, n->nobcast);
1542     qemu_put_byte(f, n->has_ufo);
1543     if (n->max_queues > 1) {
1544         qemu_put_be16(f, n->max_queues);
1545         qemu_put_be16(f, n->curr_queues);
1546         for (i = 1; i < n->curr_queues; i++) {
1547             qemu_put_be32(f, n->vqs[i].tx_waiting);
1548         }
1549     }
1550 
1551     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1552         qemu_put_be64(f, n->curr_guest_offloads);
1553     }
1554 }
1555 
1556 static int virtio_net_load_device(VirtIODevice *vdev, QEMUFile *f,
1557                                   int version_id)
1558 {
1559     VirtIONet *n = VIRTIO_NET(vdev);
1560     int i, link_down;
1561 
1562     qemu_get_buffer(f, n->mac, ETH_ALEN);
1563     n->vqs[0].tx_waiting = qemu_get_be32(f);
1564 
1565     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f),
1566                                virtio_vdev_has_feature(vdev,
1567                                                        VIRTIO_F_VERSION_1));
1568 
1569     n->status = qemu_get_be16(f);
1570 
1571     n->promisc = qemu_get_byte(f);
1572     n->allmulti = qemu_get_byte(f);
1573 
1574     n->mac_table.in_use = qemu_get_be32(f);
1575     /* MAC_TABLE_ENTRIES may be different from the saved image */
1576     if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1577         qemu_get_buffer(f, n->mac_table.macs,
1578                         n->mac_table.in_use * ETH_ALEN);
1579     } else {
1580         int64_t i;
1581 
1582         /* Overflow detected - can happen if source has a larger MAC table.
1583          * We simply set overflow flag so there's no need to maintain the
1584          * table of addresses, discard them all.
1585          * Note: 64 bit math to avoid integer overflow.
1586          */
1587         for (i = 0; i < (int64_t)n->mac_table.in_use * ETH_ALEN; ++i) {
1588             qemu_get_byte(f);
1589         }
1590         n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1591         n->mac_table.in_use = 0;
1592     }
1593 
1594     qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1595 
1596     if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1597         error_report("virtio-net: saved image requires vnet_hdr=on");
1598         return -1;
1599     }
1600 
1601     n->mac_table.multi_overflow = qemu_get_byte(f);
1602     n->mac_table.uni_overflow = qemu_get_byte(f);
1603 
1604     n->alluni = qemu_get_byte(f);
1605     n->nomulti = qemu_get_byte(f);
1606     n->nouni = qemu_get_byte(f);
1607     n->nobcast = qemu_get_byte(f);
1608 
1609     if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1610         error_report("virtio-net: saved image requires TUN_F_UFO support");
1611         return -1;
1612     }
1613 
1614     if (n->max_queues > 1) {
1615         if (n->max_queues != qemu_get_be16(f)) {
1616             error_report("virtio-net: different max_queues ");
1617             return -1;
1618         }
1619 
1620         n->curr_queues = qemu_get_be16(f);
1621         if (n->curr_queues > n->max_queues) {
1622             error_report("virtio-net: curr_queues %x > max_queues %x",
1623                          n->curr_queues, n->max_queues);
1624             return -1;
1625         }
1626         for (i = 1; i < n->curr_queues; i++) {
1627             n->vqs[i].tx_waiting = qemu_get_be32(f);
1628         }
1629     }
1630 
1631     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1632         n->curr_guest_offloads = qemu_get_be64(f);
1633     } else {
1634         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
1635     }
1636 
1637     if (peer_has_vnet_hdr(n)) {
1638         virtio_net_apply_guest_offloads(n);
1639     }
1640 
1641     virtio_net_set_queues(n);
1642 
1643     /* Find the first multicast entry in the saved MAC filter */
1644     for (i = 0; i < n->mac_table.in_use; i++) {
1645         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1646             break;
1647         }
1648     }
1649     n->mac_table.first_multi = i;
1650 
1651     /* nc.link_down can't be migrated, so infer link_down according
1652      * to link status bit in n->status */
1653     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1654     for (i = 0; i < n->max_queues; i++) {
1655         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1656     }
1657 
1658     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
1659         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
1660         n->announce_counter = SELF_ANNOUNCE_ROUNDS;
1661         timer_mod(n->announce_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL));
1662     }
1663 
1664     return 0;
1665 }
1666 
1667 static NetClientInfo net_virtio_info = {
1668     .type = NET_CLIENT_DRIVER_NIC,
1669     .size = sizeof(NICState),
1670     .can_receive = virtio_net_can_receive,
1671     .receive = virtio_net_receive,
1672     .link_status_changed = virtio_net_set_link_status,
1673     .query_rx_filter = virtio_net_query_rxfilter,
1674 };
1675 
1676 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1677 {
1678     VirtIONet *n = VIRTIO_NET(vdev);
1679     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1680     assert(n->vhost_started);
1681     return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
1682 }
1683 
1684 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1685                                            bool mask)
1686 {
1687     VirtIONet *n = VIRTIO_NET(vdev);
1688     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1689     assert(n->vhost_started);
1690     vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
1691                              vdev, idx, mask);
1692 }
1693 
1694 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
1695 {
1696     int i, config_size = 0;
1697     virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
1698     for (i = 0; feature_sizes[i].flags != 0; i++) {
1699         if (host_features & feature_sizes[i].flags) {
1700             config_size = MAX(feature_sizes[i].end, config_size);
1701         }
1702     }
1703     n->config_size = config_size;
1704 }
1705 
1706 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
1707                                    const char *type)
1708 {
1709     /*
1710      * The name can be NULL, the netclient name will be type.x.
1711      */
1712     assert(type != NULL);
1713 
1714     g_free(n->netclient_name);
1715     g_free(n->netclient_type);
1716     n->netclient_name = g_strdup(name);
1717     n->netclient_type = g_strdup(type);
1718 }
1719 
1720 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
1721 {
1722     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1723     VirtIONet *n = VIRTIO_NET(dev);
1724     NetClientState *nc;
1725     int i;
1726 
1727     virtio_net_set_config_size(n, n->host_features);
1728     virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
1729 
1730     /*
1731      * We set a lower limit on RX queue size to what it always was.
1732      * Guests that want a smaller ring can always resize it without
1733      * help from us (using virtio 1 and up).
1734      */
1735     if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
1736         n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
1737         (n->net_conf.rx_queue_size & (n->net_conf.rx_queue_size - 1))) {
1738         error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
1739                    "must be a power of 2 between %d and %d.",
1740                    n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
1741                    VIRTQUEUE_MAX_SIZE);
1742         virtio_cleanup(vdev);
1743         return;
1744     }
1745 
1746     n->max_queues = MAX(n->nic_conf.peers.queues, 1);
1747     if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) {
1748         error_setg(errp, "Invalid number of queues (= %" PRIu32 "), "
1749                    "must be a positive integer less than %d.",
1750                    n->max_queues, (VIRTIO_QUEUE_MAX - 1) / 2);
1751         virtio_cleanup(vdev);
1752         return;
1753     }
1754     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1755     n->curr_queues = 1;
1756     n->tx_timeout = n->net_conf.txtimer;
1757 
1758     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1759                        && strcmp(n->net_conf.tx, "bh")) {
1760         error_report("virtio-net: "
1761                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1762                      n->net_conf.tx);
1763         error_report("Defaulting to \"bh\"");
1764     }
1765 
1766     for (i = 0; i < n->max_queues; i++) {
1767         virtio_net_add_queue(n, i);
1768     }
1769 
1770     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1771     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1772     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1773     n->status = VIRTIO_NET_S_LINK_UP;
1774     n->announce_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
1775                                      virtio_net_announce_timer, n);
1776 
1777     if (n->netclient_type) {
1778         /*
1779          * Happen when virtio_net_set_netclient_name has been called.
1780          */
1781         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1782                               n->netclient_type, n->netclient_name, n);
1783     } else {
1784         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1785                               object_get_typename(OBJECT(dev)), dev->id, n);
1786     }
1787 
1788     peer_test_vnet_hdr(n);
1789     if (peer_has_vnet_hdr(n)) {
1790         for (i = 0; i < n->max_queues; i++) {
1791             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1792         }
1793         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1794     } else {
1795         n->host_hdr_len = 0;
1796     }
1797 
1798     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1799 
1800     n->vqs[0].tx_waiting = 0;
1801     n->tx_burst = n->net_conf.txburst;
1802     virtio_net_set_mrg_rx_bufs(n, 0, 0);
1803     n->promisc = 1; /* for compatibility */
1804 
1805     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1806 
1807     n->vlans = g_malloc0(MAX_VLAN >> 3);
1808 
1809     nc = qemu_get_queue(n->nic);
1810     nc->rxfilter_notify_enabled = 1;
1811 
1812     n->qdev = dev;
1813 }
1814 
1815 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
1816 {
1817     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1818     VirtIONet *n = VIRTIO_NET(dev);
1819     int i, max_queues;
1820 
1821     /* This will stop vhost backend if appropriate. */
1822     virtio_net_set_status(vdev, 0);
1823 
1824     g_free(n->netclient_name);
1825     n->netclient_name = NULL;
1826     g_free(n->netclient_type);
1827     n->netclient_type = NULL;
1828 
1829     g_free(n->mac_table.macs);
1830     g_free(n->vlans);
1831 
1832     max_queues = n->multiqueue ? n->max_queues : 1;
1833     for (i = 0; i < max_queues; i++) {
1834         virtio_net_del_queue(n, i);
1835     }
1836 
1837     timer_del(n->announce_timer);
1838     timer_free(n->announce_timer);
1839     g_free(n->vqs);
1840     qemu_del_nic(n->nic);
1841     virtio_cleanup(vdev);
1842 }
1843 
1844 static void virtio_net_instance_init(Object *obj)
1845 {
1846     VirtIONet *n = VIRTIO_NET(obj);
1847 
1848     /*
1849      * The default config_size is sizeof(struct virtio_net_config).
1850      * Can be overriden with virtio_net_set_config_size.
1851      */
1852     n->config_size = sizeof(struct virtio_net_config);
1853     device_add_bootindex_property(obj, &n->nic_conf.bootindex,
1854                                   "bootindex", "/ethernet-phy@0",
1855                                   DEVICE(n), NULL);
1856 }
1857 
1858 static void virtio_net_pre_save(void *opaque)
1859 {
1860     VirtIONet *n = opaque;
1861 
1862     /* At this point, backend must be stopped, otherwise
1863      * it might keep writing to memory. */
1864     assert(!n->vhost_started);
1865 }
1866 
1867 static const VMStateDescription vmstate_virtio_net = {
1868     .name = "virtio-net",
1869     .minimum_version_id = VIRTIO_NET_VM_VERSION,
1870     .version_id = VIRTIO_NET_VM_VERSION,
1871     .fields = (VMStateField[]) {
1872         VMSTATE_VIRTIO_DEVICE,
1873         VMSTATE_END_OF_LIST()
1874     },
1875     .pre_save = virtio_net_pre_save,
1876 };
1877 
1878 static Property virtio_net_properties[] = {
1879     DEFINE_PROP_BIT("csum", VirtIONet, host_features, VIRTIO_NET_F_CSUM, true),
1880     DEFINE_PROP_BIT("guest_csum", VirtIONet, host_features,
1881                     VIRTIO_NET_F_GUEST_CSUM, true),
1882     DEFINE_PROP_BIT("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
1883     DEFINE_PROP_BIT("guest_tso4", VirtIONet, host_features,
1884                     VIRTIO_NET_F_GUEST_TSO4, true),
1885     DEFINE_PROP_BIT("guest_tso6", VirtIONet, host_features,
1886                     VIRTIO_NET_F_GUEST_TSO6, true),
1887     DEFINE_PROP_BIT("guest_ecn", VirtIONet, host_features,
1888                     VIRTIO_NET_F_GUEST_ECN, true),
1889     DEFINE_PROP_BIT("guest_ufo", VirtIONet, host_features,
1890                     VIRTIO_NET_F_GUEST_UFO, true),
1891     DEFINE_PROP_BIT("guest_announce", VirtIONet, host_features,
1892                     VIRTIO_NET_F_GUEST_ANNOUNCE, true),
1893     DEFINE_PROP_BIT("host_tso4", VirtIONet, host_features,
1894                     VIRTIO_NET_F_HOST_TSO4, true),
1895     DEFINE_PROP_BIT("host_tso6", VirtIONet, host_features,
1896                     VIRTIO_NET_F_HOST_TSO6, true),
1897     DEFINE_PROP_BIT("host_ecn", VirtIONet, host_features,
1898                     VIRTIO_NET_F_HOST_ECN, true),
1899     DEFINE_PROP_BIT("host_ufo", VirtIONet, host_features,
1900                     VIRTIO_NET_F_HOST_UFO, true),
1901     DEFINE_PROP_BIT("mrg_rxbuf", VirtIONet, host_features,
1902                     VIRTIO_NET_F_MRG_RXBUF, true),
1903     DEFINE_PROP_BIT("status", VirtIONet, host_features,
1904                     VIRTIO_NET_F_STATUS, true),
1905     DEFINE_PROP_BIT("ctrl_vq", VirtIONet, host_features,
1906                     VIRTIO_NET_F_CTRL_VQ, true),
1907     DEFINE_PROP_BIT("ctrl_rx", VirtIONet, host_features,
1908                     VIRTIO_NET_F_CTRL_RX, true),
1909     DEFINE_PROP_BIT("ctrl_vlan", VirtIONet, host_features,
1910                     VIRTIO_NET_F_CTRL_VLAN, true),
1911     DEFINE_PROP_BIT("ctrl_rx_extra", VirtIONet, host_features,
1912                     VIRTIO_NET_F_CTRL_RX_EXTRA, true),
1913     DEFINE_PROP_BIT("ctrl_mac_addr", VirtIONet, host_features,
1914                     VIRTIO_NET_F_CTRL_MAC_ADDR, true),
1915     DEFINE_PROP_BIT("ctrl_guest_offloads", VirtIONet, host_features,
1916                     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
1917     DEFINE_PROP_BIT("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
1918     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
1919     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
1920                        TX_TIMER_INTERVAL),
1921     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
1922     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
1923     DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
1924                        VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
1925     DEFINE_PROP_END_OF_LIST(),
1926 };
1927 
1928 static void virtio_net_class_init(ObjectClass *klass, void *data)
1929 {
1930     DeviceClass *dc = DEVICE_CLASS(klass);
1931     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1932 
1933     dc->props = virtio_net_properties;
1934     dc->vmsd = &vmstate_virtio_net;
1935     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
1936     vdc->realize = virtio_net_device_realize;
1937     vdc->unrealize = virtio_net_device_unrealize;
1938     vdc->get_config = virtio_net_get_config;
1939     vdc->set_config = virtio_net_set_config;
1940     vdc->get_features = virtio_net_get_features;
1941     vdc->set_features = virtio_net_set_features;
1942     vdc->bad_features = virtio_net_bad_features;
1943     vdc->reset = virtio_net_reset;
1944     vdc->set_status = virtio_net_set_status;
1945     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
1946     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
1947     vdc->load = virtio_net_load_device;
1948     vdc->save = virtio_net_save_device;
1949     vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
1950 }
1951 
1952 static const TypeInfo virtio_net_info = {
1953     .name = TYPE_VIRTIO_NET,
1954     .parent = TYPE_VIRTIO_DEVICE,
1955     .instance_size = sizeof(VirtIONet),
1956     .instance_init = virtio_net_instance_init,
1957     .class_init = virtio_net_class_init,
1958 };
1959 
1960 static void virtio_register_types(void)
1961 {
1962     type_register_static(&virtio_net_info);
1963 }
1964 
1965 type_init(virtio_register_types)
1966