xref: /openbmc/qemu/hw/net/virtio-net.c (revision 80adf54e)
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 #include "migration/misc.h"
29 
30 #define VIRTIO_NET_VM_VERSION    11
31 
32 #define MAC_TABLE_ENTRIES    64
33 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
34 
35 /* previously fixed value */
36 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
37 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
38 
39 /* for now, only allow larger queues; with virtio-1, guest can downsize */
40 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
41 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
42 
43 /*
44  * Calculate the number of bytes up to and including the given 'field' of
45  * 'container'.
46  */
47 #define endof(container, field) \
48     (offsetof(container, field) + sizeof(((container *)0)->field))
49 
50 typedef struct VirtIOFeature {
51     uint32_t flags;
52     size_t end;
53 } VirtIOFeature;
54 
55 static VirtIOFeature feature_sizes[] = {
56     {.flags = 1 << VIRTIO_NET_F_MAC,
57      .end = endof(struct virtio_net_config, mac)},
58     {.flags = 1 << VIRTIO_NET_F_STATUS,
59      .end = endof(struct virtio_net_config, status)},
60     {.flags = 1 << VIRTIO_NET_F_MQ,
61      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
62     {.flags = 1 << VIRTIO_NET_F_MTU,
63      .end = endof(struct virtio_net_config, mtu)},
64     {}
65 };
66 
67 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
68 {
69     VirtIONet *n = qemu_get_nic_opaque(nc);
70 
71     return &n->vqs[nc->queue_index];
72 }
73 
74 static int vq2q(int queue_index)
75 {
76     return queue_index / 2;
77 }
78 
79 /* TODO
80  * - we could suppress RX interrupt if we were so inclined.
81  */
82 
83 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
84 {
85     VirtIONet *n = VIRTIO_NET(vdev);
86     struct virtio_net_config netcfg;
87 
88     virtio_stw_p(vdev, &netcfg.status, n->status);
89     virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues);
90     virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
91     memcpy(netcfg.mac, n->mac, ETH_ALEN);
92     memcpy(config, &netcfg, n->config_size);
93 }
94 
95 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
96 {
97     VirtIONet *n = VIRTIO_NET(vdev);
98     struct virtio_net_config netcfg = {};
99 
100     memcpy(&netcfg, config, n->config_size);
101 
102     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
103         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
104         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
105         memcpy(n->mac, netcfg.mac, ETH_ALEN);
106         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
107     }
108 }
109 
110 static bool virtio_net_started(VirtIONet *n, uint8_t status)
111 {
112     VirtIODevice *vdev = VIRTIO_DEVICE(n);
113     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
114         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
115 }
116 
117 static void virtio_net_announce_timer(void *opaque)
118 {
119     VirtIONet *n = opaque;
120     VirtIODevice *vdev = VIRTIO_DEVICE(n);
121 
122     n->announce_counter--;
123     n->status |= VIRTIO_NET_S_ANNOUNCE;
124     virtio_notify_config(vdev);
125 }
126 
127 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
128 {
129     VirtIODevice *vdev = VIRTIO_DEVICE(n);
130     NetClientState *nc = qemu_get_queue(n->nic);
131     int queues = n->multiqueue ? n->max_queues : 1;
132 
133     if (!get_vhost_net(nc->peer)) {
134         return;
135     }
136 
137     if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
138         !!n->vhost_started) {
139         return;
140     }
141     if (!n->vhost_started) {
142         int r, i;
143 
144         if (n->needs_vnet_hdr_swap) {
145             error_report("backend does not support %s vnet headers; "
146                          "falling back on userspace virtio",
147                          virtio_is_big_endian(vdev) ? "BE" : "LE");
148             return;
149         }
150 
151         /* Any packets outstanding? Purge them to avoid touching rings
152          * when vhost is running.
153          */
154         for (i = 0;  i < queues; i++) {
155             NetClientState *qnc = qemu_get_subqueue(n->nic, i);
156 
157             /* Purge both directions: TX and RX. */
158             qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
159             qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
160         }
161 
162         if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
163             r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
164             if (r < 0) {
165                 error_report("%uBytes MTU not supported by the backend",
166                              n->net_conf.mtu);
167 
168                 return;
169             }
170         }
171 
172         n->vhost_started = 1;
173         r = vhost_net_start(vdev, n->nic->ncs, queues);
174         if (r < 0) {
175             error_report("unable to start vhost net: %d: "
176                          "falling back on userspace virtio", -r);
177             n->vhost_started = 0;
178         }
179     } else {
180         vhost_net_stop(vdev, n->nic->ncs, queues);
181         n->vhost_started = 0;
182     }
183 }
184 
185 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
186                                           NetClientState *peer,
187                                           bool enable)
188 {
189     if (virtio_is_big_endian(vdev)) {
190         return qemu_set_vnet_be(peer, enable);
191     } else {
192         return qemu_set_vnet_le(peer, enable);
193     }
194 }
195 
196 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
197                                        int queues, bool enable)
198 {
199     int i;
200 
201     for (i = 0; i < queues; i++) {
202         if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
203             enable) {
204             while (--i >= 0) {
205                 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
206             }
207 
208             return true;
209         }
210     }
211 
212     return false;
213 }
214 
215 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
216 {
217     VirtIODevice *vdev = VIRTIO_DEVICE(n);
218     int queues = n->multiqueue ? n->max_queues : 1;
219 
220     if (virtio_net_started(n, status)) {
221         /* Before using the device, we tell the network backend about the
222          * endianness to use when parsing vnet headers. If the backend
223          * can't do it, we fallback onto fixing the headers in the core
224          * virtio-net code.
225          */
226         n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
227                                                             queues, true);
228     } else if (virtio_net_started(n, vdev->status)) {
229         /* After using the device, we need to reset the network backend to
230          * the default (guest native endianness), otherwise the guest may
231          * lose network connectivity if it is rebooted into a different
232          * endianness.
233          */
234         virtio_net_set_vnet_endian(vdev, n->nic->ncs, queues, false);
235     }
236 }
237 
238 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq)
239 {
240     unsigned int dropped = virtqueue_drop_all(vq);
241     if (dropped) {
242         virtio_notify(vdev, vq);
243     }
244 }
245 
246 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
247 {
248     VirtIONet *n = VIRTIO_NET(vdev);
249     VirtIONetQueue *q;
250     int i;
251     uint8_t queue_status;
252 
253     virtio_net_vnet_endian_status(n, status);
254     virtio_net_vhost_status(n, status);
255 
256     for (i = 0; i < n->max_queues; i++) {
257         NetClientState *ncs = qemu_get_subqueue(n->nic, i);
258         bool queue_started;
259         q = &n->vqs[i];
260 
261         if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
262             queue_status = 0;
263         } else {
264             queue_status = status;
265         }
266         queue_started =
267             virtio_net_started(n, queue_status) && !n->vhost_started;
268 
269         if (queue_started) {
270             qemu_flush_queued_packets(ncs);
271         }
272 
273         if (!q->tx_waiting) {
274             continue;
275         }
276 
277         if (queue_started) {
278             if (q->tx_timer) {
279                 timer_mod(q->tx_timer,
280                                qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
281             } else {
282                 qemu_bh_schedule(q->tx_bh);
283             }
284         } else {
285             if (q->tx_timer) {
286                 timer_del(q->tx_timer);
287             } else {
288                 qemu_bh_cancel(q->tx_bh);
289             }
290             if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 &&
291                 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK)) {
292                 /* if tx is waiting we are likely have some packets in tx queue
293                  * and disabled notification */
294                 q->tx_waiting = 0;
295                 virtio_queue_set_notification(q->tx_vq, 1);
296                 virtio_net_drop_tx_queue_data(vdev, q->tx_vq);
297             }
298         }
299     }
300 }
301 
302 static void virtio_net_set_link_status(NetClientState *nc)
303 {
304     VirtIONet *n = qemu_get_nic_opaque(nc);
305     VirtIODevice *vdev = VIRTIO_DEVICE(n);
306     uint16_t old_status = n->status;
307 
308     if (nc->link_down)
309         n->status &= ~VIRTIO_NET_S_LINK_UP;
310     else
311         n->status |= VIRTIO_NET_S_LINK_UP;
312 
313     if (n->status != old_status)
314         virtio_notify_config(vdev);
315 
316     virtio_net_set_status(vdev, vdev->status);
317 }
318 
319 static void rxfilter_notify(NetClientState *nc)
320 {
321     VirtIONet *n = qemu_get_nic_opaque(nc);
322 
323     if (nc->rxfilter_notify_enabled) {
324         gchar *path = object_get_canonical_path(OBJECT(n->qdev));
325         qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
326                                               n->netclient_name, path, &error_abort);
327         g_free(path);
328 
329         /* disable event notification to avoid events flooding */
330         nc->rxfilter_notify_enabled = 0;
331     }
332 }
333 
334 static intList *get_vlan_table(VirtIONet *n)
335 {
336     intList *list, *entry;
337     int i, j;
338 
339     list = NULL;
340     for (i = 0; i < MAX_VLAN >> 5; i++) {
341         for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
342             if (n->vlans[i] & (1U << j)) {
343                 entry = g_malloc0(sizeof(*entry));
344                 entry->value = (i << 5) + j;
345                 entry->next = list;
346                 list = entry;
347             }
348         }
349     }
350 
351     return list;
352 }
353 
354 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
355 {
356     VirtIONet *n = qemu_get_nic_opaque(nc);
357     VirtIODevice *vdev = VIRTIO_DEVICE(n);
358     RxFilterInfo *info;
359     strList *str_list, *entry;
360     int i;
361 
362     info = g_malloc0(sizeof(*info));
363     info->name = g_strdup(nc->name);
364     info->promiscuous = n->promisc;
365 
366     if (n->nouni) {
367         info->unicast = RX_STATE_NONE;
368     } else if (n->alluni) {
369         info->unicast = RX_STATE_ALL;
370     } else {
371         info->unicast = RX_STATE_NORMAL;
372     }
373 
374     if (n->nomulti) {
375         info->multicast = RX_STATE_NONE;
376     } else if (n->allmulti) {
377         info->multicast = RX_STATE_ALL;
378     } else {
379         info->multicast = RX_STATE_NORMAL;
380     }
381 
382     info->broadcast_allowed = n->nobcast;
383     info->multicast_overflow = n->mac_table.multi_overflow;
384     info->unicast_overflow = n->mac_table.uni_overflow;
385 
386     info->main_mac = qemu_mac_strdup_printf(n->mac);
387 
388     str_list = NULL;
389     for (i = 0; i < n->mac_table.first_multi; i++) {
390         entry = g_malloc0(sizeof(*entry));
391         entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
392         entry->next = str_list;
393         str_list = entry;
394     }
395     info->unicast_table = str_list;
396 
397     str_list = NULL;
398     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
399         entry = g_malloc0(sizeof(*entry));
400         entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
401         entry->next = str_list;
402         str_list = entry;
403     }
404     info->multicast_table = str_list;
405     info->vlan_table = get_vlan_table(n);
406 
407     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
408         info->vlan = RX_STATE_ALL;
409     } else if (!info->vlan_table) {
410         info->vlan = RX_STATE_NONE;
411     } else {
412         info->vlan = RX_STATE_NORMAL;
413     }
414 
415     /* enable event notification after query */
416     nc->rxfilter_notify_enabled = 1;
417 
418     return info;
419 }
420 
421 static void virtio_net_reset(VirtIODevice *vdev)
422 {
423     VirtIONet *n = VIRTIO_NET(vdev);
424 
425     /* Reset back to compatibility mode */
426     n->promisc = 1;
427     n->allmulti = 0;
428     n->alluni = 0;
429     n->nomulti = 0;
430     n->nouni = 0;
431     n->nobcast = 0;
432     /* multiqueue is disabled by default */
433     n->curr_queues = 1;
434     timer_del(n->announce_timer);
435     n->announce_counter = 0;
436     n->status &= ~VIRTIO_NET_S_ANNOUNCE;
437 
438     /* Flush any MAC and VLAN filter table state */
439     n->mac_table.in_use = 0;
440     n->mac_table.first_multi = 0;
441     n->mac_table.multi_overflow = 0;
442     n->mac_table.uni_overflow = 0;
443     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
444     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
445     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
446     memset(n->vlans, 0, MAX_VLAN >> 3);
447 }
448 
449 static void peer_test_vnet_hdr(VirtIONet *n)
450 {
451     NetClientState *nc = qemu_get_queue(n->nic);
452     if (!nc->peer) {
453         return;
454     }
455 
456     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
457 }
458 
459 static int peer_has_vnet_hdr(VirtIONet *n)
460 {
461     return n->has_vnet_hdr;
462 }
463 
464 static int peer_has_ufo(VirtIONet *n)
465 {
466     if (!peer_has_vnet_hdr(n))
467         return 0;
468 
469     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
470 
471     return n->has_ufo;
472 }
473 
474 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
475                                        int version_1)
476 {
477     int i;
478     NetClientState *nc;
479 
480     n->mergeable_rx_bufs = mergeable_rx_bufs;
481 
482     if (version_1) {
483         n->guest_hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
484     } else {
485         n->guest_hdr_len = n->mergeable_rx_bufs ?
486             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
487             sizeof(struct virtio_net_hdr);
488     }
489 
490     for (i = 0; i < n->max_queues; i++) {
491         nc = qemu_get_subqueue(n->nic, i);
492 
493         if (peer_has_vnet_hdr(n) &&
494             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
495             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
496             n->host_hdr_len = n->guest_hdr_len;
497         }
498     }
499 }
500 
501 static int virtio_net_max_tx_queue_size(VirtIONet *n)
502 {
503     NetClientState *peer = n->nic_conf.peers.ncs[0];
504 
505     /*
506      * Backends other than vhost-user don't support max queue size.
507      */
508     if (!peer) {
509         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
510     }
511 
512     if (peer->info->type != NET_CLIENT_DRIVER_VHOST_USER) {
513         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
514     }
515 
516     return VIRTQUEUE_MAX_SIZE;
517 }
518 
519 static int peer_attach(VirtIONet *n, int index)
520 {
521     NetClientState *nc = qemu_get_subqueue(n->nic, index);
522 
523     if (!nc->peer) {
524         return 0;
525     }
526 
527     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
528         vhost_set_vring_enable(nc->peer, 1);
529     }
530 
531     if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
532         return 0;
533     }
534 
535     if (n->max_queues == 1) {
536         return 0;
537     }
538 
539     return tap_enable(nc->peer);
540 }
541 
542 static int peer_detach(VirtIONet *n, int index)
543 {
544     NetClientState *nc = qemu_get_subqueue(n->nic, index);
545 
546     if (!nc->peer) {
547         return 0;
548     }
549 
550     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
551         vhost_set_vring_enable(nc->peer, 0);
552     }
553 
554     if (nc->peer->info->type !=  NET_CLIENT_DRIVER_TAP) {
555         return 0;
556     }
557 
558     return tap_disable(nc->peer);
559 }
560 
561 static void virtio_net_set_queues(VirtIONet *n)
562 {
563     int i;
564     int r;
565 
566     if (n->nic->peer_deleted) {
567         return;
568     }
569 
570     for (i = 0; i < n->max_queues; i++) {
571         if (i < n->curr_queues) {
572             r = peer_attach(n, i);
573             assert(!r);
574         } else {
575             r = peer_detach(n, i);
576             assert(!r);
577         }
578     }
579 }
580 
581 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
582 
583 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
584                                         Error **errp)
585 {
586     VirtIONet *n = VIRTIO_NET(vdev);
587     NetClientState *nc = qemu_get_queue(n->nic);
588 
589     /* Firstly sync all virtio-net possible supported features */
590     features |= n->host_features;
591 
592     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
593 
594     if (!peer_has_vnet_hdr(n)) {
595         virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
596         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
597         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
598         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
599 
600         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
601         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
602         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
603         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
604     }
605 
606     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
607         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
608         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
609     }
610 
611     if (!get_vhost_net(nc->peer)) {
612         return features;
613     }
614     features = vhost_net_get_features(get_vhost_net(nc->peer), features);
615     vdev->backend_features = features;
616 
617     if (n->mtu_bypass_backend &&
618             (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) {
619         features |= (1ULL << VIRTIO_NET_F_MTU);
620     }
621 
622     return features;
623 }
624 
625 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
626 {
627     uint64_t features = 0;
628 
629     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
630      * but also these: */
631     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
632     virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
633     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
634     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
635     virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
636 
637     return features;
638 }
639 
640 static void virtio_net_apply_guest_offloads(VirtIONet *n)
641 {
642     qemu_set_offload(qemu_get_queue(n->nic)->peer,
643             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
644             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
645             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
646             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
647             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
648 }
649 
650 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
651 {
652     static const uint64_t guest_offloads_mask =
653         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
654         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
655         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
656         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
657         (1ULL << VIRTIO_NET_F_GUEST_UFO);
658 
659     return guest_offloads_mask & features;
660 }
661 
662 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
663 {
664     VirtIODevice *vdev = VIRTIO_DEVICE(n);
665     return virtio_net_guest_offloads_by_features(vdev->guest_features);
666 }
667 
668 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
669 {
670     VirtIONet *n = VIRTIO_NET(vdev);
671     int i;
672 
673     if (n->mtu_bypass_backend &&
674             !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) {
675         features &= ~(1ULL << VIRTIO_NET_F_MTU);
676     }
677 
678     virtio_net_set_multiqueue(n,
679                               virtio_has_feature(features, VIRTIO_NET_F_MQ));
680 
681     virtio_net_set_mrg_rx_bufs(n,
682                                virtio_has_feature(features,
683                                                   VIRTIO_NET_F_MRG_RXBUF),
684                                virtio_has_feature(features,
685                                                   VIRTIO_F_VERSION_1));
686 
687     if (n->has_vnet_hdr) {
688         n->curr_guest_offloads =
689             virtio_net_guest_offloads_by_features(features);
690         virtio_net_apply_guest_offloads(n);
691     }
692 
693     for (i = 0;  i < n->max_queues; i++) {
694         NetClientState *nc = qemu_get_subqueue(n->nic, i);
695 
696         if (!get_vhost_net(nc->peer)) {
697             continue;
698         }
699         vhost_net_ack_features(get_vhost_net(nc->peer), features);
700     }
701 
702     if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
703         memset(n->vlans, 0, MAX_VLAN >> 3);
704     } else {
705         memset(n->vlans, 0xff, MAX_VLAN >> 3);
706     }
707 }
708 
709 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
710                                      struct iovec *iov, unsigned int iov_cnt)
711 {
712     uint8_t on;
713     size_t s;
714     NetClientState *nc = qemu_get_queue(n->nic);
715 
716     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
717     if (s != sizeof(on)) {
718         return VIRTIO_NET_ERR;
719     }
720 
721     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
722         n->promisc = on;
723     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
724         n->allmulti = on;
725     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
726         n->alluni = on;
727     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
728         n->nomulti = on;
729     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
730         n->nouni = on;
731     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
732         n->nobcast = on;
733     } else {
734         return VIRTIO_NET_ERR;
735     }
736 
737     rxfilter_notify(nc);
738 
739     return VIRTIO_NET_OK;
740 }
741 
742 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
743                                      struct iovec *iov, unsigned int iov_cnt)
744 {
745     VirtIODevice *vdev = VIRTIO_DEVICE(n);
746     uint64_t offloads;
747     size_t s;
748 
749     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
750         return VIRTIO_NET_ERR;
751     }
752 
753     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
754     if (s != sizeof(offloads)) {
755         return VIRTIO_NET_ERR;
756     }
757 
758     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
759         uint64_t supported_offloads;
760 
761         offloads = virtio_ldq_p(vdev, &offloads);
762 
763         if (!n->has_vnet_hdr) {
764             return VIRTIO_NET_ERR;
765         }
766 
767         supported_offloads = virtio_net_supported_guest_offloads(n);
768         if (offloads & ~supported_offloads) {
769             return VIRTIO_NET_ERR;
770         }
771 
772         n->curr_guest_offloads = offloads;
773         virtio_net_apply_guest_offloads(n);
774 
775         return VIRTIO_NET_OK;
776     } else {
777         return VIRTIO_NET_ERR;
778     }
779 }
780 
781 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
782                                  struct iovec *iov, unsigned int iov_cnt)
783 {
784     VirtIODevice *vdev = VIRTIO_DEVICE(n);
785     struct virtio_net_ctrl_mac mac_data;
786     size_t s;
787     NetClientState *nc = qemu_get_queue(n->nic);
788 
789     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
790         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
791             return VIRTIO_NET_ERR;
792         }
793         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
794         assert(s == sizeof(n->mac));
795         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
796         rxfilter_notify(nc);
797 
798         return VIRTIO_NET_OK;
799     }
800 
801     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
802         return VIRTIO_NET_ERR;
803     }
804 
805     int in_use = 0;
806     int first_multi = 0;
807     uint8_t uni_overflow = 0;
808     uint8_t multi_overflow = 0;
809     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
810 
811     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
812                    sizeof(mac_data.entries));
813     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
814     if (s != sizeof(mac_data.entries)) {
815         goto error;
816     }
817     iov_discard_front(&iov, &iov_cnt, s);
818 
819     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
820         goto error;
821     }
822 
823     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
824         s = iov_to_buf(iov, iov_cnt, 0, macs,
825                        mac_data.entries * ETH_ALEN);
826         if (s != mac_data.entries * ETH_ALEN) {
827             goto error;
828         }
829         in_use += mac_data.entries;
830     } else {
831         uni_overflow = 1;
832     }
833 
834     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
835 
836     first_multi = in_use;
837 
838     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
839                    sizeof(mac_data.entries));
840     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
841     if (s != sizeof(mac_data.entries)) {
842         goto error;
843     }
844 
845     iov_discard_front(&iov, &iov_cnt, s);
846 
847     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
848         goto error;
849     }
850 
851     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
852         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
853                        mac_data.entries * ETH_ALEN);
854         if (s != mac_data.entries * ETH_ALEN) {
855             goto error;
856         }
857         in_use += mac_data.entries;
858     } else {
859         multi_overflow = 1;
860     }
861 
862     n->mac_table.in_use = in_use;
863     n->mac_table.first_multi = first_multi;
864     n->mac_table.uni_overflow = uni_overflow;
865     n->mac_table.multi_overflow = multi_overflow;
866     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
867     g_free(macs);
868     rxfilter_notify(nc);
869 
870     return VIRTIO_NET_OK;
871 
872 error:
873     g_free(macs);
874     return VIRTIO_NET_ERR;
875 }
876 
877 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
878                                         struct iovec *iov, unsigned int iov_cnt)
879 {
880     VirtIODevice *vdev = VIRTIO_DEVICE(n);
881     uint16_t vid;
882     size_t s;
883     NetClientState *nc = qemu_get_queue(n->nic);
884 
885     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
886     vid = virtio_lduw_p(vdev, &vid);
887     if (s != sizeof(vid)) {
888         return VIRTIO_NET_ERR;
889     }
890 
891     if (vid >= MAX_VLAN)
892         return VIRTIO_NET_ERR;
893 
894     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
895         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
896     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
897         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
898     else
899         return VIRTIO_NET_ERR;
900 
901     rxfilter_notify(nc);
902 
903     return VIRTIO_NET_OK;
904 }
905 
906 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
907                                       struct iovec *iov, unsigned int iov_cnt)
908 {
909     if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
910         n->status & VIRTIO_NET_S_ANNOUNCE) {
911         n->status &= ~VIRTIO_NET_S_ANNOUNCE;
912         if (n->announce_counter) {
913             timer_mod(n->announce_timer,
914                       qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
915                       self_announce_delay(n->announce_counter));
916         }
917         return VIRTIO_NET_OK;
918     } else {
919         return VIRTIO_NET_ERR;
920     }
921 }
922 
923 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
924                                 struct iovec *iov, unsigned int iov_cnt)
925 {
926     VirtIODevice *vdev = VIRTIO_DEVICE(n);
927     struct virtio_net_ctrl_mq mq;
928     size_t s;
929     uint16_t queues;
930 
931     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
932     if (s != sizeof(mq)) {
933         return VIRTIO_NET_ERR;
934     }
935 
936     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
937         return VIRTIO_NET_ERR;
938     }
939 
940     queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
941 
942     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
943         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
944         queues > n->max_queues ||
945         !n->multiqueue) {
946         return VIRTIO_NET_ERR;
947     }
948 
949     n->curr_queues = queues;
950     /* stop the backend before changing the number of queues to avoid handling a
951      * disabled queue */
952     virtio_net_set_status(vdev, vdev->status);
953     virtio_net_set_queues(n);
954 
955     return VIRTIO_NET_OK;
956 }
957 
958 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
959 {
960     VirtIONet *n = VIRTIO_NET(vdev);
961     struct virtio_net_ctrl_hdr ctrl;
962     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
963     VirtQueueElement *elem;
964     size_t s;
965     struct iovec *iov, *iov2;
966     unsigned int iov_cnt;
967 
968     for (;;) {
969         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
970         if (!elem) {
971             break;
972         }
973         if (iov_size(elem->in_sg, elem->in_num) < sizeof(status) ||
974             iov_size(elem->out_sg, elem->out_num) < sizeof(ctrl)) {
975             virtio_error(vdev, "virtio-net ctrl missing headers");
976             virtqueue_detach_element(vq, elem, 0);
977             g_free(elem);
978             break;
979         }
980 
981         iov_cnt = elem->out_num;
982         iov2 = iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num);
983         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
984         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
985         if (s != sizeof(ctrl)) {
986             status = VIRTIO_NET_ERR;
987         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
988             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
989         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
990             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
991         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
992             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
993         } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
994             status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
995         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
996             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
997         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
998             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
999         }
1000 
1001         s = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, sizeof(status));
1002         assert(s == sizeof(status));
1003 
1004         virtqueue_push(vq, elem, sizeof(status));
1005         virtio_notify(vdev, vq);
1006         g_free(iov2);
1007         g_free(elem);
1008     }
1009 }
1010 
1011 /* RX */
1012 
1013 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
1014 {
1015     VirtIONet *n = VIRTIO_NET(vdev);
1016     int queue_index = vq2q(virtio_get_queue_index(vq));
1017 
1018     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
1019 }
1020 
1021 static int virtio_net_can_receive(NetClientState *nc)
1022 {
1023     VirtIONet *n = qemu_get_nic_opaque(nc);
1024     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1025     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1026 
1027     if (!vdev->vm_running) {
1028         return 0;
1029     }
1030 
1031     if (nc->queue_index >= n->curr_queues) {
1032         return 0;
1033     }
1034 
1035     if (!virtio_queue_ready(q->rx_vq) ||
1036         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1037         return 0;
1038     }
1039 
1040     return 1;
1041 }
1042 
1043 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1044 {
1045     VirtIONet *n = q->n;
1046     if (virtio_queue_empty(q->rx_vq) ||
1047         (n->mergeable_rx_bufs &&
1048          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1049         virtio_queue_set_notification(q->rx_vq, 1);
1050 
1051         /* To avoid a race condition where the guest has made some buffers
1052          * available after the above check but before notification was
1053          * enabled, check for available buffers again.
1054          */
1055         if (virtio_queue_empty(q->rx_vq) ||
1056             (n->mergeable_rx_bufs &&
1057              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1058             return 0;
1059         }
1060     }
1061 
1062     virtio_queue_set_notification(q->rx_vq, 0);
1063     return 1;
1064 }
1065 
1066 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1067 {
1068     virtio_tswap16s(vdev, &hdr->hdr_len);
1069     virtio_tswap16s(vdev, &hdr->gso_size);
1070     virtio_tswap16s(vdev, &hdr->csum_start);
1071     virtio_tswap16s(vdev, &hdr->csum_offset);
1072 }
1073 
1074 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1075  * it never finds out that the packets don't have valid checksums.  This
1076  * causes dhclient to get upset.  Fedora's carried a patch for ages to
1077  * fix this with Xen but it hasn't appeared in an upstream release of
1078  * dhclient yet.
1079  *
1080  * To avoid breaking existing guests, we catch udp packets and add
1081  * checksums.  This is terrible but it's better than hacking the guest
1082  * kernels.
1083  *
1084  * N.B. if we introduce a zero-copy API, this operation is no longer free so
1085  * we should provide a mechanism to disable it to avoid polluting the host
1086  * cache.
1087  */
1088 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1089                                         uint8_t *buf, size_t size)
1090 {
1091     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1092         (size > 27 && size < 1500) && /* normal sized MTU */
1093         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1094         (buf[23] == 17) && /* ip.protocol == UDP */
1095         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1096         net_checksum_calculate(buf, size);
1097         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1098     }
1099 }
1100 
1101 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1102                            const void *buf, size_t size)
1103 {
1104     if (n->has_vnet_hdr) {
1105         /* FIXME this cast is evil */
1106         void *wbuf = (void *)buf;
1107         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1108                                     size - n->host_hdr_len);
1109 
1110         if (n->needs_vnet_hdr_swap) {
1111             virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1112         }
1113         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1114     } else {
1115         struct virtio_net_hdr hdr = {
1116             .flags = 0,
1117             .gso_type = VIRTIO_NET_HDR_GSO_NONE
1118         };
1119         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1120     }
1121 }
1122 
1123 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1124 {
1125     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1126     static const uint8_t vlan[] = {0x81, 0x00};
1127     uint8_t *ptr = (uint8_t *)buf;
1128     int i;
1129 
1130     if (n->promisc)
1131         return 1;
1132 
1133     ptr += n->host_hdr_len;
1134 
1135     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1136         int vid = lduw_be_p(ptr + 14) & 0xfff;
1137         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1138             return 0;
1139     }
1140 
1141     if (ptr[0] & 1) { // multicast
1142         if (!memcmp(ptr, bcast, sizeof(bcast))) {
1143             return !n->nobcast;
1144         } else if (n->nomulti) {
1145             return 0;
1146         } else if (n->allmulti || n->mac_table.multi_overflow) {
1147             return 1;
1148         }
1149 
1150         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1151             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1152                 return 1;
1153             }
1154         }
1155     } else { // unicast
1156         if (n->nouni) {
1157             return 0;
1158         } else if (n->alluni || n->mac_table.uni_overflow) {
1159             return 1;
1160         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1161             return 1;
1162         }
1163 
1164         for (i = 0; i < n->mac_table.first_multi; i++) {
1165             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1166                 return 1;
1167             }
1168         }
1169     }
1170 
1171     return 0;
1172 }
1173 
1174 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
1175                                       size_t size)
1176 {
1177     VirtIONet *n = qemu_get_nic_opaque(nc);
1178     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1179     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1180     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1181     struct virtio_net_hdr_mrg_rxbuf mhdr;
1182     unsigned mhdr_cnt = 0;
1183     size_t offset, i, guest_offset;
1184 
1185     if (!virtio_net_can_receive(nc)) {
1186         return -1;
1187     }
1188 
1189     /* hdr_len refers to the header we supply to the guest */
1190     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1191         return 0;
1192     }
1193 
1194     if (!receive_filter(n, buf, size))
1195         return size;
1196 
1197     offset = i = 0;
1198 
1199     while (offset < size) {
1200         VirtQueueElement *elem;
1201         int len, total;
1202         const struct iovec *sg;
1203 
1204         total = 0;
1205 
1206         elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1207         if (!elem) {
1208             if (i) {
1209                 virtio_error(vdev, "virtio-net unexpected empty queue: "
1210                              "i %zd mergeable %d offset %zd, size %zd, "
1211                              "guest hdr len %zd, host hdr len %zd "
1212                              "guest features 0x%" PRIx64,
1213                              i, n->mergeable_rx_bufs, offset, size,
1214                              n->guest_hdr_len, n->host_hdr_len,
1215                              vdev->guest_features);
1216             }
1217             return -1;
1218         }
1219 
1220         if (elem->in_num < 1) {
1221             virtio_error(vdev,
1222                          "virtio-net receive queue contains no in buffers");
1223             virtqueue_detach_element(q->rx_vq, elem, 0);
1224             g_free(elem);
1225             return -1;
1226         }
1227 
1228         sg = elem->in_sg;
1229         if (i == 0) {
1230             assert(offset == 0);
1231             if (n->mergeable_rx_bufs) {
1232                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1233                                     sg, elem->in_num,
1234                                     offsetof(typeof(mhdr), num_buffers),
1235                                     sizeof(mhdr.num_buffers));
1236             }
1237 
1238             receive_header(n, sg, elem->in_num, buf, size);
1239             offset = n->host_hdr_len;
1240             total += n->guest_hdr_len;
1241             guest_offset = n->guest_hdr_len;
1242         } else {
1243             guest_offset = 0;
1244         }
1245 
1246         /* copy in packet.  ugh */
1247         len = iov_from_buf(sg, elem->in_num, guest_offset,
1248                            buf + offset, size - offset);
1249         total += len;
1250         offset += len;
1251         /* If buffers can't be merged, at this point we
1252          * must have consumed the complete packet.
1253          * Otherwise, drop it. */
1254         if (!n->mergeable_rx_bufs && offset < size) {
1255             virtqueue_unpop(q->rx_vq, elem, total);
1256             g_free(elem);
1257             return size;
1258         }
1259 
1260         /* signal other side */
1261         virtqueue_fill(q->rx_vq, elem, total, i++);
1262         g_free(elem);
1263     }
1264 
1265     if (mhdr_cnt) {
1266         virtio_stw_p(vdev, &mhdr.num_buffers, i);
1267         iov_from_buf(mhdr_sg, mhdr_cnt,
1268                      0,
1269                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1270     }
1271 
1272     virtqueue_flush(q->rx_vq, i);
1273     virtio_notify(vdev, q->rx_vq);
1274 
1275     return size;
1276 }
1277 
1278 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf,
1279                                   size_t size)
1280 {
1281     ssize_t r;
1282 
1283     rcu_read_lock();
1284     r = virtio_net_receive_rcu(nc, buf, size);
1285     rcu_read_unlock();
1286     return r;
1287 }
1288 
1289 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1290 
1291 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1292 {
1293     VirtIONet *n = qemu_get_nic_opaque(nc);
1294     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1295     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1296 
1297     virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
1298     virtio_notify(vdev, q->tx_vq);
1299 
1300     g_free(q->async_tx.elem);
1301     q->async_tx.elem = NULL;
1302 
1303     virtio_queue_set_notification(q->tx_vq, 1);
1304     virtio_net_flush_tx(q);
1305 }
1306 
1307 /* TX */
1308 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
1309 {
1310     VirtIONet *n = q->n;
1311     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1312     VirtQueueElement *elem;
1313     int32_t num_packets = 0;
1314     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
1315     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1316         return num_packets;
1317     }
1318 
1319     if (q->async_tx.elem) {
1320         virtio_queue_set_notification(q->tx_vq, 0);
1321         return num_packets;
1322     }
1323 
1324     for (;;) {
1325         ssize_t ret;
1326         unsigned int out_num;
1327         struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
1328         struct virtio_net_hdr_mrg_rxbuf mhdr;
1329 
1330         elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
1331         if (!elem) {
1332             break;
1333         }
1334 
1335         out_num = elem->out_num;
1336         out_sg = elem->out_sg;
1337         if (out_num < 1) {
1338             virtio_error(vdev, "virtio-net header not in first element");
1339             virtqueue_detach_element(q->tx_vq, elem, 0);
1340             g_free(elem);
1341             return -EINVAL;
1342         }
1343 
1344         if (n->has_vnet_hdr) {
1345             if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
1346                 n->guest_hdr_len) {
1347                 virtio_error(vdev, "virtio-net header incorrect");
1348                 virtqueue_detach_element(q->tx_vq, elem, 0);
1349                 g_free(elem);
1350                 return -EINVAL;
1351             }
1352             if (n->needs_vnet_hdr_swap) {
1353                 virtio_net_hdr_swap(vdev, (void *) &mhdr);
1354                 sg2[0].iov_base = &mhdr;
1355                 sg2[0].iov_len = n->guest_hdr_len;
1356                 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
1357                                    out_sg, out_num,
1358                                    n->guest_hdr_len, -1);
1359                 if (out_num == VIRTQUEUE_MAX_SIZE) {
1360                     goto drop;
1361 		}
1362                 out_num += 1;
1363                 out_sg = sg2;
1364 	    }
1365         }
1366         /*
1367          * If host wants to see the guest header as is, we can
1368          * pass it on unchanged. Otherwise, copy just the parts
1369          * that host is interested in.
1370          */
1371         assert(n->host_hdr_len <= n->guest_hdr_len);
1372         if (n->host_hdr_len != n->guest_hdr_len) {
1373             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
1374                                        out_sg, out_num,
1375                                        0, n->host_hdr_len);
1376             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
1377                              out_sg, out_num,
1378                              n->guest_hdr_len, -1);
1379             out_num = sg_num;
1380             out_sg = sg;
1381         }
1382 
1383         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
1384                                       out_sg, out_num, virtio_net_tx_complete);
1385         if (ret == 0) {
1386             virtio_queue_set_notification(q->tx_vq, 0);
1387             q->async_tx.elem = elem;
1388             return -EBUSY;
1389         }
1390 
1391 drop:
1392         virtqueue_push(q->tx_vq, elem, 0);
1393         virtio_notify(vdev, q->tx_vq);
1394         g_free(elem);
1395 
1396         if (++num_packets >= n->tx_burst) {
1397             break;
1398         }
1399     }
1400     return num_packets;
1401 }
1402 
1403 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
1404 {
1405     VirtIONet *n = VIRTIO_NET(vdev);
1406     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1407 
1408     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
1409         virtio_net_drop_tx_queue_data(vdev, vq);
1410         return;
1411     }
1412 
1413     /* This happens when device was stopped but VCPU wasn't. */
1414     if (!vdev->vm_running) {
1415         q->tx_waiting = 1;
1416         return;
1417     }
1418 
1419     if (q->tx_waiting) {
1420         virtio_queue_set_notification(vq, 1);
1421         timer_del(q->tx_timer);
1422         q->tx_waiting = 0;
1423         if (virtio_net_flush_tx(q) == -EINVAL) {
1424             return;
1425         }
1426     } else {
1427         timer_mod(q->tx_timer,
1428                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
1429         q->tx_waiting = 1;
1430         virtio_queue_set_notification(vq, 0);
1431     }
1432 }
1433 
1434 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1435 {
1436     VirtIONet *n = VIRTIO_NET(vdev);
1437     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1438 
1439     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
1440         virtio_net_drop_tx_queue_data(vdev, vq);
1441         return;
1442     }
1443 
1444     if (unlikely(q->tx_waiting)) {
1445         return;
1446     }
1447     q->tx_waiting = 1;
1448     /* This happens when device was stopped but VCPU wasn't. */
1449     if (!vdev->vm_running) {
1450         return;
1451     }
1452     virtio_queue_set_notification(vq, 0);
1453     qemu_bh_schedule(q->tx_bh);
1454 }
1455 
1456 static void virtio_net_tx_timer(void *opaque)
1457 {
1458     VirtIONetQueue *q = opaque;
1459     VirtIONet *n = q->n;
1460     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1461     /* This happens when device was stopped but BH wasn't. */
1462     if (!vdev->vm_running) {
1463         /* Make sure tx waiting is set, so we'll run when restarted. */
1464         assert(q->tx_waiting);
1465         return;
1466     }
1467 
1468     q->tx_waiting = 0;
1469 
1470     /* Just in case the driver is not ready on more */
1471     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1472         return;
1473     }
1474 
1475     virtio_queue_set_notification(q->tx_vq, 1);
1476     virtio_net_flush_tx(q);
1477 }
1478 
1479 static void virtio_net_tx_bh(void *opaque)
1480 {
1481     VirtIONetQueue *q = opaque;
1482     VirtIONet *n = q->n;
1483     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1484     int32_t ret;
1485 
1486     /* This happens when device was stopped but BH wasn't. */
1487     if (!vdev->vm_running) {
1488         /* Make sure tx waiting is set, so we'll run when restarted. */
1489         assert(q->tx_waiting);
1490         return;
1491     }
1492 
1493     q->tx_waiting = 0;
1494 
1495     /* Just in case the driver is not ready on more */
1496     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1497         return;
1498     }
1499 
1500     ret = virtio_net_flush_tx(q);
1501     if (ret == -EBUSY || ret == -EINVAL) {
1502         return; /* Notification re-enable handled by tx_complete or device
1503                  * broken */
1504     }
1505 
1506     /* If we flush a full burst of packets, assume there are
1507      * more coming and immediately reschedule */
1508     if (ret >= n->tx_burst) {
1509         qemu_bh_schedule(q->tx_bh);
1510         q->tx_waiting = 1;
1511         return;
1512     }
1513 
1514     /* If less than a full burst, re-enable notification and flush
1515      * anything that may have come in while we weren't looking.  If
1516      * we find something, assume the guest is still active and reschedule */
1517     virtio_queue_set_notification(q->tx_vq, 1);
1518     ret = virtio_net_flush_tx(q);
1519     if (ret == -EINVAL) {
1520         return;
1521     } else if (ret > 0) {
1522         virtio_queue_set_notification(q->tx_vq, 0);
1523         qemu_bh_schedule(q->tx_bh);
1524         q->tx_waiting = 1;
1525     }
1526 }
1527 
1528 static void virtio_net_add_queue(VirtIONet *n, int index)
1529 {
1530     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1531 
1532     n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
1533                                            virtio_net_handle_rx);
1534 
1535     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1536         n->vqs[index].tx_vq =
1537             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
1538                              virtio_net_handle_tx_timer);
1539         n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
1540                                               virtio_net_tx_timer,
1541                                               &n->vqs[index]);
1542     } else {
1543         n->vqs[index].tx_vq =
1544             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
1545                              virtio_net_handle_tx_bh);
1546         n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
1547     }
1548 
1549     n->vqs[index].tx_waiting = 0;
1550     n->vqs[index].n = n;
1551 }
1552 
1553 static void virtio_net_del_queue(VirtIONet *n, int index)
1554 {
1555     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1556     VirtIONetQueue *q = &n->vqs[index];
1557     NetClientState *nc = qemu_get_subqueue(n->nic, index);
1558 
1559     qemu_purge_queued_packets(nc);
1560 
1561     virtio_del_queue(vdev, index * 2);
1562     if (q->tx_timer) {
1563         timer_del(q->tx_timer);
1564         timer_free(q->tx_timer);
1565         q->tx_timer = NULL;
1566     } else {
1567         qemu_bh_delete(q->tx_bh);
1568         q->tx_bh = NULL;
1569     }
1570     q->tx_waiting = 0;
1571     virtio_del_queue(vdev, index * 2 + 1);
1572 }
1573 
1574 static void virtio_net_change_num_queues(VirtIONet *n, int new_max_queues)
1575 {
1576     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1577     int old_num_queues = virtio_get_num_queues(vdev);
1578     int new_num_queues = new_max_queues * 2 + 1;
1579     int i;
1580 
1581     assert(old_num_queues >= 3);
1582     assert(old_num_queues % 2 == 1);
1583 
1584     if (old_num_queues == new_num_queues) {
1585         return;
1586     }
1587 
1588     /*
1589      * We always need to remove and add ctrl vq if
1590      * old_num_queues != new_num_queues. Remove ctrl_vq first,
1591      * and then we only enter one of the following too loops.
1592      */
1593     virtio_del_queue(vdev, old_num_queues - 1);
1594 
1595     for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
1596         /* new_num_queues < old_num_queues */
1597         virtio_net_del_queue(n, i / 2);
1598     }
1599 
1600     for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
1601         /* new_num_queues > old_num_queues */
1602         virtio_net_add_queue(n, i / 2);
1603     }
1604 
1605     /* add ctrl_vq last */
1606     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1607 }
1608 
1609 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1610 {
1611     int max = multiqueue ? n->max_queues : 1;
1612 
1613     n->multiqueue = multiqueue;
1614     virtio_net_change_num_queues(n, max);
1615 
1616     virtio_net_set_queues(n);
1617 }
1618 
1619 static int virtio_net_post_load_device(void *opaque, int version_id)
1620 {
1621     VirtIONet *n = opaque;
1622     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1623     int i, link_down;
1624 
1625     virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
1626                                virtio_vdev_has_feature(vdev,
1627                                                        VIRTIO_F_VERSION_1));
1628 
1629     /* MAC_TABLE_ENTRIES may be different from the saved image */
1630     if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
1631         n->mac_table.in_use = 0;
1632     }
1633 
1634     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1635         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
1636     }
1637 
1638     if (peer_has_vnet_hdr(n)) {
1639         virtio_net_apply_guest_offloads(n);
1640     }
1641 
1642     virtio_net_set_queues(n);
1643 
1644     /* Find the first multicast entry in the saved MAC filter */
1645     for (i = 0; i < n->mac_table.in_use; i++) {
1646         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1647             break;
1648         }
1649     }
1650     n->mac_table.first_multi = i;
1651 
1652     /* nc.link_down can't be migrated, so infer link_down according
1653      * to link status bit in n->status */
1654     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1655     for (i = 0; i < n->max_queues; i++) {
1656         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1657     }
1658 
1659     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
1660         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
1661         n->announce_counter = SELF_ANNOUNCE_ROUNDS;
1662         timer_mod(n->announce_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL));
1663     }
1664 
1665     return 0;
1666 }
1667 
1668 /* tx_waiting field of a VirtIONetQueue */
1669 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = {
1670     .name = "virtio-net-queue-tx_waiting",
1671     .fields = (VMStateField[]) {
1672         VMSTATE_UINT32(tx_waiting, VirtIONetQueue),
1673         VMSTATE_END_OF_LIST()
1674    },
1675 };
1676 
1677 static bool max_queues_gt_1(void *opaque, int version_id)
1678 {
1679     return VIRTIO_NET(opaque)->max_queues > 1;
1680 }
1681 
1682 static bool has_ctrl_guest_offloads(void *opaque, int version_id)
1683 {
1684     return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque),
1685                                    VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
1686 }
1687 
1688 static bool mac_table_fits(void *opaque, int version_id)
1689 {
1690     return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES;
1691 }
1692 
1693 static bool mac_table_doesnt_fit(void *opaque, int version_id)
1694 {
1695     return !mac_table_fits(opaque, version_id);
1696 }
1697 
1698 /* This temporary type is shared by all the WITH_TMP methods
1699  * although only some fields are used by each.
1700  */
1701 struct VirtIONetMigTmp {
1702     VirtIONet      *parent;
1703     VirtIONetQueue *vqs_1;
1704     uint16_t        curr_queues_1;
1705     uint8_t         has_ufo;
1706     uint32_t        has_vnet_hdr;
1707 };
1708 
1709 /* The 2nd and subsequent tx_waiting flags are loaded later than
1710  * the 1st entry in the queues and only if there's more than one
1711  * entry.  We use the tmp mechanism to calculate a temporary
1712  * pointer and count and also validate the count.
1713  */
1714 
1715 static void virtio_net_tx_waiting_pre_save(void *opaque)
1716 {
1717     struct VirtIONetMigTmp *tmp = opaque;
1718 
1719     tmp->vqs_1 = tmp->parent->vqs + 1;
1720     tmp->curr_queues_1 = tmp->parent->curr_queues - 1;
1721     if (tmp->parent->curr_queues == 0) {
1722         tmp->curr_queues_1 = 0;
1723     }
1724 }
1725 
1726 static int virtio_net_tx_waiting_pre_load(void *opaque)
1727 {
1728     struct VirtIONetMigTmp *tmp = opaque;
1729 
1730     /* Reuse the pointer setup from save */
1731     virtio_net_tx_waiting_pre_save(opaque);
1732 
1733     if (tmp->parent->curr_queues > tmp->parent->max_queues) {
1734         error_report("virtio-net: curr_queues %x > max_queues %x",
1735             tmp->parent->curr_queues, tmp->parent->max_queues);
1736 
1737         return -EINVAL;
1738     }
1739 
1740     return 0; /* all good */
1741 }
1742 
1743 static const VMStateDescription vmstate_virtio_net_tx_waiting = {
1744     .name      = "virtio-net-tx_waiting",
1745     .pre_load  = virtio_net_tx_waiting_pre_load,
1746     .pre_save  = virtio_net_tx_waiting_pre_save,
1747     .fields    = (VMStateField[]) {
1748         VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp,
1749                                      curr_queues_1,
1750                                      vmstate_virtio_net_queue_tx_waiting,
1751                                      struct VirtIONetQueue),
1752         VMSTATE_END_OF_LIST()
1753     },
1754 };
1755 
1756 /* the 'has_ufo' flag is just tested; if the incoming stream has the
1757  * flag set we need to check that we have it
1758  */
1759 static int virtio_net_ufo_post_load(void *opaque, int version_id)
1760 {
1761     struct VirtIONetMigTmp *tmp = opaque;
1762 
1763     if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) {
1764         error_report("virtio-net: saved image requires TUN_F_UFO support");
1765         return -EINVAL;
1766     }
1767 
1768     return 0;
1769 }
1770 
1771 static void virtio_net_ufo_pre_save(void *opaque)
1772 {
1773     struct VirtIONetMigTmp *tmp = opaque;
1774 
1775     tmp->has_ufo = tmp->parent->has_ufo;
1776 }
1777 
1778 static const VMStateDescription vmstate_virtio_net_has_ufo = {
1779     .name      = "virtio-net-ufo",
1780     .post_load = virtio_net_ufo_post_load,
1781     .pre_save  = virtio_net_ufo_pre_save,
1782     .fields    = (VMStateField[]) {
1783         VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp),
1784         VMSTATE_END_OF_LIST()
1785     },
1786 };
1787 
1788 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
1789  * flag set we need to check that we have it
1790  */
1791 static int virtio_net_vnet_post_load(void *opaque, int version_id)
1792 {
1793     struct VirtIONetMigTmp *tmp = opaque;
1794 
1795     if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) {
1796         error_report("virtio-net: saved image requires vnet_hdr=on");
1797         return -EINVAL;
1798     }
1799 
1800     return 0;
1801 }
1802 
1803 static void virtio_net_vnet_pre_save(void *opaque)
1804 {
1805     struct VirtIONetMigTmp *tmp = opaque;
1806 
1807     tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr;
1808 }
1809 
1810 static const VMStateDescription vmstate_virtio_net_has_vnet = {
1811     .name      = "virtio-net-vnet",
1812     .post_load = virtio_net_vnet_post_load,
1813     .pre_save  = virtio_net_vnet_pre_save,
1814     .fields    = (VMStateField[]) {
1815         VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp),
1816         VMSTATE_END_OF_LIST()
1817     },
1818 };
1819 
1820 static const VMStateDescription vmstate_virtio_net_device = {
1821     .name = "virtio-net-device",
1822     .version_id = VIRTIO_NET_VM_VERSION,
1823     .minimum_version_id = VIRTIO_NET_VM_VERSION,
1824     .post_load = virtio_net_post_load_device,
1825     .fields = (VMStateField[]) {
1826         VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN),
1827         VMSTATE_STRUCT_POINTER(vqs, VirtIONet,
1828                                vmstate_virtio_net_queue_tx_waiting,
1829                                VirtIONetQueue),
1830         VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet),
1831         VMSTATE_UINT16(status, VirtIONet),
1832         VMSTATE_UINT8(promisc, VirtIONet),
1833         VMSTATE_UINT8(allmulti, VirtIONet),
1834         VMSTATE_UINT32(mac_table.in_use, VirtIONet),
1835 
1836         /* Guarded pair: If it fits we load it, else we throw it away
1837          * - can happen if source has a larger MAC table.; post-load
1838          *  sets flags in this case.
1839          */
1840         VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet,
1841                                 0, mac_table_fits, mac_table.in_use,
1842                                  ETH_ALEN),
1843         VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0,
1844                                      mac_table.in_use, ETH_ALEN),
1845 
1846         /* Note: This is an array of uint32's that's always been saved as a
1847          * buffer; hold onto your endiannesses; it's actually used as a bitmap
1848          * but based on the uint.
1849          */
1850         VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3),
1851         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
1852                          vmstate_virtio_net_has_vnet),
1853         VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet),
1854         VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet),
1855         VMSTATE_UINT8(alluni, VirtIONet),
1856         VMSTATE_UINT8(nomulti, VirtIONet),
1857         VMSTATE_UINT8(nouni, VirtIONet),
1858         VMSTATE_UINT8(nobcast, VirtIONet),
1859         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
1860                          vmstate_virtio_net_has_ufo),
1861         VMSTATE_SINGLE_TEST(max_queues, VirtIONet, max_queues_gt_1, 0,
1862                             vmstate_info_uint16_equal, uint16_t),
1863         VMSTATE_UINT16_TEST(curr_queues, VirtIONet, max_queues_gt_1),
1864         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
1865                          vmstate_virtio_net_tx_waiting),
1866         VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet,
1867                             has_ctrl_guest_offloads),
1868         VMSTATE_END_OF_LIST()
1869    },
1870 };
1871 
1872 static NetClientInfo net_virtio_info = {
1873     .type = NET_CLIENT_DRIVER_NIC,
1874     .size = sizeof(NICState),
1875     .can_receive = virtio_net_can_receive,
1876     .receive = virtio_net_receive,
1877     .link_status_changed = virtio_net_set_link_status,
1878     .query_rx_filter = virtio_net_query_rxfilter,
1879 };
1880 
1881 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1882 {
1883     VirtIONet *n = VIRTIO_NET(vdev);
1884     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1885     assert(n->vhost_started);
1886     return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
1887 }
1888 
1889 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1890                                            bool mask)
1891 {
1892     VirtIONet *n = VIRTIO_NET(vdev);
1893     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1894     assert(n->vhost_started);
1895     vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
1896                              vdev, idx, mask);
1897 }
1898 
1899 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
1900 {
1901     int i, config_size = 0;
1902     virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
1903 
1904     for (i = 0; feature_sizes[i].flags != 0; i++) {
1905         if (host_features & feature_sizes[i].flags) {
1906             config_size = MAX(feature_sizes[i].end, config_size);
1907         }
1908     }
1909     n->config_size = config_size;
1910 }
1911 
1912 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
1913                                    const char *type)
1914 {
1915     /*
1916      * The name can be NULL, the netclient name will be type.x.
1917      */
1918     assert(type != NULL);
1919 
1920     g_free(n->netclient_name);
1921     g_free(n->netclient_type);
1922     n->netclient_name = g_strdup(name);
1923     n->netclient_type = g_strdup(type);
1924 }
1925 
1926 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
1927 {
1928     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1929     VirtIONet *n = VIRTIO_NET(dev);
1930     NetClientState *nc;
1931     int i;
1932 
1933     if (n->net_conf.mtu) {
1934         n->host_features |= (0x1 << VIRTIO_NET_F_MTU);
1935     }
1936 
1937     virtio_net_set_config_size(n, n->host_features);
1938     virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
1939 
1940     /*
1941      * We set a lower limit on RX queue size to what it always was.
1942      * Guests that want a smaller ring can always resize it without
1943      * help from us (using virtio 1 and up).
1944      */
1945     if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
1946         n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
1947         !is_power_of_2(n->net_conf.rx_queue_size)) {
1948         error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
1949                    "must be a power of 2 between %d and %d.",
1950                    n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
1951                    VIRTQUEUE_MAX_SIZE);
1952         virtio_cleanup(vdev);
1953         return;
1954     }
1955 
1956     if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE ||
1957         n->net_conf.tx_queue_size > VIRTQUEUE_MAX_SIZE ||
1958         !is_power_of_2(n->net_conf.tx_queue_size)) {
1959         error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), "
1960                    "must be a power of 2 between %d and %d",
1961                    n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE,
1962                    VIRTQUEUE_MAX_SIZE);
1963         virtio_cleanup(vdev);
1964         return;
1965     }
1966 
1967     n->max_queues = MAX(n->nic_conf.peers.queues, 1);
1968     if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) {
1969         error_setg(errp, "Invalid number of queues (= %" PRIu32 "), "
1970                    "must be a positive integer less than %d.",
1971                    n->max_queues, (VIRTIO_QUEUE_MAX - 1) / 2);
1972         virtio_cleanup(vdev);
1973         return;
1974     }
1975     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1976     n->curr_queues = 1;
1977     n->tx_timeout = n->net_conf.txtimer;
1978 
1979     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1980                        && strcmp(n->net_conf.tx, "bh")) {
1981         error_report("virtio-net: "
1982                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1983                      n->net_conf.tx);
1984         error_report("Defaulting to \"bh\"");
1985     }
1986 
1987     n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n),
1988                                     n->net_conf.tx_queue_size);
1989 
1990     for (i = 0; i < n->max_queues; i++) {
1991         virtio_net_add_queue(n, i);
1992     }
1993 
1994     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1995     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1996     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1997     n->status = VIRTIO_NET_S_LINK_UP;
1998     n->announce_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
1999                                      virtio_net_announce_timer, n);
2000 
2001     if (n->netclient_type) {
2002         /*
2003          * Happen when virtio_net_set_netclient_name has been called.
2004          */
2005         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
2006                               n->netclient_type, n->netclient_name, n);
2007     } else {
2008         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
2009                               object_get_typename(OBJECT(dev)), dev->id, n);
2010     }
2011 
2012     peer_test_vnet_hdr(n);
2013     if (peer_has_vnet_hdr(n)) {
2014         for (i = 0; i < n->max_queues; i++) {
2015             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
2016         }
2017         n->host_hdr_len = sizeof(struct virtio_net_hdr);
2018     } else {
2019         n->host_hdr_len = 0;
2020     }
2021 
2022     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
2023 
2024     n->vqs[0].tx_waiting = 0;
2025     n->tx_burst = n->net_conf.txburst;
2026     virtio_net_set_mrg_rx_bufs(n, 0, 0);
2027     n->promisc = 1; /* for compatibility */
2028 
2029     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
2030 
2031     n->vlans = g_malloc0(MAX_VLAN >> 3);
2032 
2033     nc = qemu_get_queue(n->nic);
2034     nc->rxfilter_notify_enabled = 1;
2035 
2036     n->qdev = dev;
2037 }
2038 
2039 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
2040 {
2041     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2042     VirtIONet *n = VIRTIO_NET(dev);
2043     int i, max_queues;
2044 
2045     /* This will stop vhost backend if appropriate. */
2046     virtio_net_set_status(vdev, 0);
2047 
2048     g_free(n->netclient_name);
2049     n->netclient_name = NULL;
2050     g_free(n->netclient_type);
2051     n->netclient_type = NULL;
2052 
2053     g_free(n->mac_table.macs);
2054     g_free(n->vlans);
2055 
2056     max_queues = n->multiqueue ? n->max_queues : 1;
2057     for (i = 0; i < max_queues; i++) {
2058         virtio_net_del_queue(n, i);
2059     }
2060 
2061     timer_del(n->announce_timer);
2062     timer_free(n->announce_timer);
2063     g_free(n->vqs);
2064     qemu_del_nic(n->nic);
2065     virtio_cleanup(vdev);
2066 }
2067 
2068 static void virtio_net_instance_init(Object *obj)
2069 {
2070     VirtIONet *n = VIRTIO_NET(obj);
2071 
2072     /*
2073      * The default config_size is sizeof(struct virtio_net_config).
2074      * Can be overriden with virtio_net_set_config_size.
2075      */
2076     n->config_size = sizeof(struct virtio_net_config);
2077     device_add_bootindex_property(obj, &n->nic_conf.bootindex,
2078                                   "bootindex", "/ethernet-phy@0",
2079                                   DEVICE(n), NULL);
2080 }
2081 
2082 static void virtio_net_pre_save(void *opaque)
2083 {
2084     VirtIONet *n = opaque;
2085 
2086     /* At this point, backend must be stopped, otherwise
2087      * it might keep writing to memory. */
2088     assert(!n->vhost_started);
2089 }
2090 
2091 static const VMStateDescription vmstate_virtio_net = {
2092     .name = "virtio-net",
2093     .minimum_version_id = VIRTIO_NET_VM_VERSION,
2094     .version_id = VIRTIO_NET_VM_VERSION,
2095     .fields = (VMStateField[]) {
2096         VMSTATE_VIRTIO_DEVICE,
2097         VMSTATE_END_OF_LIST()
2098     },
2099     .pre_save = virtio_net_pre_save,
2100 };
2101 
2102 static Property virtio_net_properties[] = {
2103     DEFINE_PROP_BIT("csum", VirtIONet, host_features, VIRTIO_NET_F_CSUM, true),
2104     DEFINE_PROP_BIT("guest_csum", VirtIONet, host_features,
2105                     VIRTIO_NET_F_GUEST_CSUM, true),
2106     DEFINE_PROP_BIT("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
2107     DEFINE_PROP_BIT("guest_tso4", VirtIONet, host_features,
2108                     VIRTIO_NET_F_GUEST_TSO4, true),
2109     DEFINE_PROP_BIT("guest_tso6", VirtIONet, host_features,
2110                     VIRTIO_NET_F_GUEST_TSO6, true),
2111     DEFINE_PROP_BIT("guest_ecn", VirtIONet, host_features,
2112                     VIRTIO_NET_F_GUEST_ECN, true),
2113     DEFINE_PROP_BIT("guest_ufo", VirtIONet, host_features,
2114                     VIRTIO_NET_F_GUEST_UFO, true),
2115     DEFINE_PROP_BIT("guest_announce", VirtIONet, host_features,
2116                     VIRTIO_NET_F_GUEST_ANNOUNCE, true),
2117     DEFINE_PROP_BIT("host_tso4", VirtIONet, host_features,
2118                     VIRTIO_NET_F_HOST_TSO4, true),
2119     DEFINE_PROP_BIT("host_tso6", VirtIONet, host_features,
2120                     VIRTIO_NET_F_HOST_TSO6, true),
2121     DEFINE_PROP_BIT("host_ecn", VirtIONet, host_features,
2122                     VIRTIO_NET_F_HOST_ECN, true),
2123     DEFINE_PROP_BIT("host_ufo", VirtIONet, host_features,
2124                     VIRTIO_NET_F_HOST_UFO, true),
2125     DEFINE_PROP_BIT("mrg_rxbuf", VirtIONet, host_features,
2126                     VIRTIO_NET_F_MRG_RXBUF, true),
2127     DEFINE_PROP_BIT("status", VirtIONet, host_features,
2128                     VIRTIO_NET_F_STATUS, true),
2129     DEFINE_PROP_BIT("ctrl_vq", VirtIONet, host_features,
2130                     VIRTIO_NET_F_CTRL_VQ, true),
2131     DEFINE_PROP_BIT("ctrl_rx", VirtIONet, host_features,
2132                     VIRTIO_NET_F_CTRL_RX, true),
2133     DEFINE_PROP_BIT("ctrl_vlan", VirtIONet, host_features,
2134                     VIRTIO_NET_F_CTRL_VLAN, true),
2135     DEFINE_PROP_BIT("ctrl_rx_extra", VirtIONet, host_features,
2136                     VIRTIO_NET_F_CTRL_RX_EXTRA, true),
2137     DEFINE_PROP_BIT("ctrl_mac_addr", VirtIONet, host_features,
2138                     VIRTIO_NET_F_CTRL_MAC_ADDR, true),
2139     DEFINE_PROP_BIT("ctrl_guest_offloads", VirtIONet, host_features,
2140                     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
2141     DEFINE_PROP_BIT("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
2142     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
2143     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
2144                        TX_TIMER_INTERVAL),
2145     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
2146     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
2147     DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
2148                        VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
2149     DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size,
2150                        VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE),
2151     DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
2152     DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend,
2153                      true),
2154     DEFINE_PROP_END_OF_LIST(),
2155 };
2156 
2157 static void virtio_net_class_init(ObjectClass *klass, void *data)
2158 {
2159     DeviceClass *dc = DEVICE_CLASS(klass);
2160     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
2161 
2162     dc->props = virtio_net_properties;
2163     dc->vmsd = &vmstate_virtio_net;
2164     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
2165     vdc->realize = virtio_net_device_realize;
2166     vdc->unrealize = virtio_net_device_unrealize;
2167     vdc->get_config = virtio_net_get_config;
2168     vdc->set_config = virtio_net_set_config;
2169     vdc->get_features = virtio_net_get_features;
2170     vdc->set_features = virtio_net_set_features;
2171     vdc->bad_features = virtio_net_bad_features;
2172     vdc->reset = virtio_net_reset;
2173     vdc->set_status = virtio_net_set_status;
2174     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
2175     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
2176     vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
2177     vdc->vmsd = &vmstate_virtio_net_device;
2178 }
2179 
2180 static const TypeInfo virtio_net_info = {
2181     .name = TYPE_VIRTIO_NET,
2182     .parent = TYPE_VIRTIO_DEVICE,
2183     .instance_size = sizeof(VirtIONet),
2184     .instance_init = virtio_net_instance_init,
2185     .class_init = virtio_net_class_init,
2186 };
2187 
2188 static void virtio_register_types(void)
2189 {
2190     type_register_static(&virtio_net_info);
2191 }
2192 
2193 type_init(virtio_register_types)
2194