xref: /openbmc/qemu/hw/net/virtio-net.c (revision da3e8a23492dbc13c4b70d90b6ae42970624e63a)
1 /*
2  * Virtio Network Device
3  *
4  * Copyright IBM, Corp. 2007
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/iov.h"
15 #include "hw/virtio/virtio.h"
16 #include "net/net.h"
17 #include "net/checksum.h"
18 #include "net/tap.h"
19 #include "qemu/error-report.h"
20 #include "qemu/timer.h"
21 #include "hw/virtio/virtio-net.h"
22 #include "net/vhost_net.h"
23 #include "hw/virtio/virtio-bus.h"
24 #include "qapi/qmp/qjson.h"
25 #include "qapi-event.h"
26 #include "hw/virtio/virtio-access.h"
27 
28 #define VIRTIO_NET_VM_VERSION    11
29 
30 #define MAC_TABLE_ENTRIES    64
31 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
32 
33 /*
34  * Calculate the number of bytes up to and including the given 'field' of
35  * 'container'.
36  */
37 #define endof(container, field) \
38     (offsetof(container, field) + sizeof(((container *)0)->field))
39 
40 typedef struct VirtIOFeature {
41     uint32_t flags;
42     size_t end;
43 } VirtIOFeature;
44 
45 static VirtIOFeature feature_sizes[] = {
46     {.flags = 1 << VIRTIO_NET_F_MAC,
47      .end = endof(struct virtio_net_config, mac)},
48     {.flags = 1 << VIRTIO_NET_F_STATUS,
49      .end = endof(struct virtio_net_config, status)},
50     {.flags = 1 << VIRTIO_NET_F_MQ,
51      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
52     {}
53 };
54 
55 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
56 {
57     VirtIONet *n = qemu_get_nic_opaque(nc);
58 
59     return &n->vqs[nc->queue_index];
60 }
61 
62 static int vq2q(int queue_index)
63 {
64     return queue_index / 2;
65 }
66 
67 /* TODO
68  * - we could suppress RX interrupt if we were so inclined.
69  */
70 
71 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
72 {
73     VirtIONet *n = VIRTIO_NET(vdev);
74     struct virtio_net_config netcfg;
75 
76     virtio_stw_p(vdev, &netcfg.status, n->status);
77     virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues);
78     memcpy(netcfg.mac, n->mac, ETH_ALEN);
79     memcpy(config, &netcfg, n->config_size);
80 }
81 
82 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
83 {
84     VirtIONet *n = VIRTIO_NET(vdev);
85     struct virtio_net_config netcfg = {};
86 
87     memcpy(&netcfg, config, n->config_size);
88 
89     if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
90         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
91         memcpy(n->mac, netcfg.mac, ETH_ALEN);
92         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
93     }
94 }
95 
96 static bool virtio_net_started(VirtIONet *n, uint8_t status)
97 {
98     VirtIODevice *vdev = VIRTIO_DEVICE(n);
99     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
100         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
101 }
102 
103 static void virtio_net_announce_timer(void *opaque)
104 {
105     VirtIONet *n = opaque;
106     VirtIODevice *vdev = VIRTIO_DEVICE(n);
107 
108     n->announce_counter--;
109     n->status |= VIRTIO_NET_S_ANNOUNCE;
110     virtio_notify_config(vdev);
111 }
112 
113 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
114 {
115     VirtIODevice *vdev = VIRTIO_DEVICE(n);
116     NetClientState *nc = qemu_get_queue(n->nic);
117     int queues = n->multiqueue ? n->max_queues : 1;
118 
119     if (!get_vhost_net(nc->peer)) {
120         return;
121     }
122 
123     if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
124         !!n->vhost_started) {
125         return;
126     }
127     if (!n->vhost_started) {
128         int r, i;
129 
130         if (!vhost_net_query(get_vhost_net(nc->peer), vdev)) {
131             return;
132         }
133 
134         /* Any packets outstanding? Purge them to avoid touching rings
135          * when vhost is running.
136          */
137         for (i = 0;  i < queues; i++) {
138             NetClientState *qnc = qemu_get_subqueue(n->nic, i);
139 
140             /* Purge both directions: TX and RX. */
141             qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
142             qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
143         }
144 
145         n->vhost_started = 1;
146         r = vhost_net_start(vdev, n->nic->ncs, queues);
147         if (r < 0) {
148             error_report("unable to start vhost net: %d: "
149                          "falling back on userspace virtio", -r);
150             n->vhost_started = 0;
151         }
152     } else {
153         vhost_net_stop(vdev, n->nic->ncs, queues);
154         n->vhost_started = 0;
155     }
156 }
157 
158 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
159 {
160     VirtIONet *n = VIRTIO_NET(vdev);
161     VirtIONetQueue *q;
162     int i;
163     uint8_t queue_status;
164 
165     virtio_net_vhost_status(n, status);
166 
167     for (i = 0; i < n->max_queues; i++) {
168         q = &n->vqs[i];
169 
170         if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
171             queue_status = 0;
172         } else {
173             queue_status = status;
174         }
175 
176         if (!q->tx_waiting) {
177             continue;
178         }
179 
180         if (virtio_net_started(n, queue_status) && !n->vhost_started) {
181             if (q->tx_timer) {
182                 timer_mod(q->tx_timer,
183                                qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
184             } else {
185                 qemu_bh_schedule(q->tx_bh);
186             }
187         } else {
188             if (q->tx_timer) {
189                 timer_del(q->tx_timer);
190             } else {
191                 qemu_bh_cancel(q->tx_bh);
192             }
193         }
194     }
195 }
196 
197 static void virtio_net_set_link_status(NetClientState *nc)
198 {
199     VirtIONet *n = qemu_get_nic_opaque(nc);
200     VirtIODevice *vdev = VIRTIO_DEVICE(n);
201     uint16_t old_status = n->status;
202 
203     if (nc->link_down)
204         n->status &= ~VIRTIO_NET_S_LINK_UP;
205     else
206         n->status |= VIRTIO_NET_S_LINK_UP;
207 
208     if (n->status != old_status)
209         virtio_notify_config(vdev);
210 
211     virtio_net_set_status(vdev, vdev->status);
212 }
213 
214 static void rxfilter_notify(NetClientState *nc)
215 {
216     VirtIONet *n = qemu_get_nic_opaque(nc);
217 
218     if (nc->rxfilter_notify_enabled) {
219         gchar *path = object_get_canonical_path(OBJECT(n->qdev));
220         qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
221                                               n->netclient_name, path, &error_abort);
222         g_free(path);
223 
224         /* disable event notification to avoid events flooding */
225         nc->rxfilter_notify_enabled = 0;
226     }
227 }
228 
229 static char *mac_strdup_printf(const uint8_t *mac)
230 {
231     return g_strdup_printf("%.2x:%.2x:%.2x:%.2x:%.2x:%.2x", mac[0],
232                             mac[1], mac[2], mac[3], mac[4], mac[5]);
233 }
234 
235 static intList *get_vlan_table(VirtIONet *n)
236 {
237     intList *list, *entry;
238     int i, j;
239 
240     list = NULL;
241     for (i = 0; i < MAX_VLAN >> 5; i++) {
242         for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
243             if (n->vlans[i] & (1U << j)) {
244                 entry = g_malloc0(sizeof(*entry));
245                 entry->value = (i << 5) + j;
246                 entry->next = list;
247                 list = entry;
248             }
249         }
250     }
251 
252     return list;
253 }
254 
255 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
256 {
257     VirtIONet *n = qemu_get_nic_opaque(nc);
258     VirtIODevice *vdev = VIRTIO_DEVICE(n);
259     RxFilterInfo *info;
260     strList *str_list, *entry;
261     int i;
262 
263     info = g_malloc0(sizeof(*info));
264     info->name = g_strdup(nc->name);
265     info->promiscuous = n->promisc;
266 
267     if (n->nouni) {
268         info->unicast = RX_STATE_NONE;
269     } else if (n->alluni) {
270         info->unicast = RX_STATE_ALL;
271     } else {
272         info->unicast = RX_STATE_NORMAL;
273     }
274 
275     if (n->nomulti) {
276         info->multicast = RX_STATE_NONE;
277     } else if (n->allmulti) {
278         info->multicast = RX_STATE_ALL;
279     } else {
280         info->multicast = RX_STATE_NORMAL;
281     }
282 
283     info->broadcast_allowed = n->nobcast;
284     info->multicast_overflow = n->mac_table.multi_overflow;
285     info->unicast_overflow = n->mac_table.uni_overflow;
286 
287     info->main_mac = mac_strdup_printf(n->mac);
288 
289     str_list = NULL;
290     for (i = 0; i < n->mac_table.first_multi; i++) {
291         entry = g_malloc0(sizeof(*entry));
292         entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
293         entry->next = str_list;
294         str_list = entry;
295     }
296     info->unicast_table = str_list;
297 
298     str_list = NULL;
299     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
300         entry = g_malloc0(sizeof(*entry));
301         entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
302         entry->next = str_list;
303         str_list = entry;
304     }
305     info->multicast_table = str_list;
306     info->vlan_table = get_vlan_table(n);
307 
308     if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
309         info->vlan = RX_STATE_ALL;
310     } else if (!info->vlan_table) {
311         info->vlan = RX_STATE_NONE;
312     } else {
313         info->vlan = RX_STATE_NORMAL;
314     }
315 
316     /* enable event notification after query */
317     nc->rxfilter_notify_enabled = 1;
318 
319     return info;
320 }
321 
322 static void virtio_net_reset(VirtIODevice *vdev)
323 {
324     VirtIONet *n = VIRTIO_NET(vdev);
325 
326     /* Reset back to compatibility mode */
327     n->promisc = 1;
328     n->allmulti = 0;
329     n->alluni = 0;
330     n->nomulti = 0;
331     n->nouni = 0;
332     n->nobcast = 0;
333     /* multiqueue is disabled by default */
334     n->curr_queues = 1;
335     timer_del(n->announce_timer);
336     n->announce_counter = 0;
337     n->status &= ~VIRTIO_NET_S_ANNOUNCE;
338 
339     /* Flush any MAC and VLAN filter table state */
340     n->mac_table.in_use = 0;
341     n->mac_table.first_multi = 0;
342     n->mac_table.multi_overflow = 0;
343     n->mac_table.uni_overflow = 0;
344     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
345     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
346     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
347     memset(n->vlans, 0, MAX_VLAN >> 3);
348 }
349 
350 static void peer_test_vnet_hdr(VirtIONet *n)
351 {
352     NetClientState *nc = qemu_get_queue(n->nic);
353     if (!nc->peer) {
354         return;
355     }
356 
357     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
358 }
359 
360 static int peer_has_vnet_hdr(VirtIONet *n)
361 {
362     return n->has_vnet_hdr;
363 }
364 
365 static int peer_has_ufo(VirtIONet *n)
366 {
367     if (!peer_has_vnet_hdr(n))
368         return 0;
369 
370     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
371 
372     return n->has_ufo;
373 }
374 
375 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs)
376 {
377     int i;
378     NetClientState *nc;
379 
380     n->mergeable_rx_bufs = mergeable_rx_bufs;
381 
382     n->guest_hdr_len = n->mergeable_rx_bufs ?
383         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
384 
385     for (i = 0; i < n->max_queues; i++) {
386         nc = qemu_get_subqueue(n->nic, i);
387 
388         if (peer_has_vnet_hdr(n) &&
389             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
390             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
391             n->host_hdr_len = n->guest_hdr_len;
392         }
393     }
394 }
395 
396 static int peer_attach(VirtIONet *n, int index)
397 {
398     NetClientState *nc = qemu_get_subqueue(n->nic, index);
399 
400     if (!nc->peer) {
401         return 0;
402     }
403 
404     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
405         return 0;
406     }
407 
408     return tap_enable(nc->peer);
409 }
410 
411 static int peer_detach(VirtIONet *n, int index)
412 {
413     NetClientState *nc = qemu_get_subqueue(n->nic, index);
414 
415     if (!nc->peer) {
416         return 0;
417     }
418 
419     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
420         return 0;
421     }
422 
423     return tap_disable(nc->peer);
424 }
425 
426 static void virtio_net_set_queues(VirtIONet *n)
427 {
428     int i;
429     int r;
430 
431     for (i = 0; i < n->max_queues; i++) {
432         if (i < n->curr_queues) {
433             r = peer_attach(n, i);
434             assert(!r);
435         } else {
436             r = peer_detach(n, i);
437             assert(!r);
438         }
439     }
440 }
441 
442 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
443 
444 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
445 {
446     VirtIONet *n = VIRTIO_NET(vdev);
447     NetClientState *nc = qemu_get_queue(n->nic);
448 
449     /* Firstly sync all virtio-net possible supported features */
450     features |= n->host_features;
451 
452     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
453 
454     if (!peer_has_vnet_hdr(n)) {
455         virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
456         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
457         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
458         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
459 
460         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
461         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
462         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
463         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
464     }
465 
466     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
467         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
468         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
469     }
470 
471     if (!get_vhost_net(nc->peer)) {
472         return features;
473     }
474     return vhost_net_get_features(get_vhost_net(nc->peer), features);
475 }
476 
477 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
478 {
479     uint32_t features = 0;
480 
481     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
482      * but also these: */
483     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
484     virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
485     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
486     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
487     virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
488 
489     return features;
490 }
491 
492 static void virtio_net_apply_guest_offloads(VirtIONet *n)
493 {
494     qemu_set_offload(qemu_get_queue(n->nic)->peer,
495             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
496             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
497             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
498             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
499             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
500 }
501 
502 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
503 {
504     static const uint64_t guest_offloads_mask =
505         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
506         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
507         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
508         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
509         (1ULL << VIRTIO_NET_F_GUEST_UFO);
510 
511     return guest_offloads_mask & features;
512 }
513 
514 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
515 {
516     VirtIODevice *vdev = VIRTIO_DEVICE(n);
517     return virtio_net_guest_offloads_by_features(vdev->guest_features);
518 }
519 
520 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
521 {
522     VirtIONet *n = VIRTIO_NET(vdev);
523     int i;
524 
525     virtio_net_set_multiqueue(n,
526                               __virtio_has_feature(features, VIRTIO_NET_F_MQ));
527 
528     virtio_net_set_mrg_rx_bufs(n,
529                                __virtio_has_feature(features,
530                                                     VIRTIO_NET_F_MRG_RXBUF));
531 
532     if (n->has_vnet_hdr) {
533         n->curr_guest_offloads =
534             virtio_net_guest_offloads_by_features(features);
535         virtio_net_apply_guest_offloads(n);
536     }
537 
538     for (i = 0;  i < n->max_queues; i++) {
539         NetClientState *nc = qemu_get_subqueue(n->nic, i);
540 
541         if (!get_vhost_net(nc->peer)) {
542             continue;
543         }
544         vhost_net_ack_features(get_vhost_net(nc->peer), features);
545     }
546 
547     if (__virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
548         memset(n->vlans, 0, MAX_VLAN >> 3);
549     } else {
550         memset(n->vlans, 0xff, MAX_VLAN >> 3);
551     }
552 }
553 
554 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
555                                      struct iovec *iov, unsigned int iov_cnt)
556 {
557     uint8_t on;
558     size_t s;
559     NetClientState *nc = qemu_get_queue(n->nic);
560 
561     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
562     if (s != sizeof(on)) {
563         return VIRTIO_NET_ERR;
564     }
565 
566     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
567         n->promisc = on;
568     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
569         n->allmulti = on;
570     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
571         n->alluni = on;
572     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
573         n->nomulti = on;
574     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
575         n->nouni = on;
576     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
577         n->nobcast = on;
578     } else {
579         return VIRTIO_NET_ERR;
580     }
581 
582     rxfilter_notify(nc);
583 
584     return VIRTIO_NET_OK;
585 }
586 
587 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
588                                      struct iovec *iov, unsigned int iov_cnt)
589 {
590     VirtIODevice *vdev = VIRTIO_DEVICE(n);
591     uint64_t offloads;
592     size_t s;
593 
594     if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
595         return VIRTIO_NET_ERR;
596     }
597 
598     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
599     if (s != sizeof(offloads)) {
600         return VIRTIO_NET_ERR;
601     }
602 
603     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
604         uint64_t supported_offloads;
605 
606         if (!n->has_vnet_hdr) {
607             return VIRTIO_NET_ERR;
608         }
609 
610         supported_offloads = virtio_net_supported_guest_offloads(n);
611         if (offloads & ~supported_offloads) {
612             return VIRTIO_NET_ERR;
613         }
614 
615         n->curr_guest_offloads = offloads;
616         virtio_net_apply_guest_offloads(n);
617 
618         return VIRTIO_NET_OK;
619     } else {
620         return VIRTIO_NET_ERR;
621     }
622 }
623 
624 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
625                                  struct iovec *iov, unsigned int iov_cnt)
626 {
627     VirtIODevice *vdev = VIRTIO_DEVICE(n);
628     struct virtio_net_ctrl_mac mac_data;
629     size_t s;
630     NetClientState *nc = qemu_get_queue(n->nic);
631 
632     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
633         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
634             return VIRTIO_NET_ERR;
635         }
636         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
637         assert(s == sizeof(n->mac));
638         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
639         rxfilter_notify(nc);
640 
641         return VIRTIO_NET_OK;
642     }
643 
644     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
645         return VIRTIO_NET_ERR;
646     }
647 
648     int in_use = 0;
649     int first_multi = 0;
650     uint8_t uni_overflow = 0;
651     uint8_t multi_overflow = 0;
652     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
653 
654     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
655                    sizeof(mac_data.entries));
656     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
657     if (s != sizeof(mac_data.entries)) {
658         goto error;
659     }
660     iov_discard_front(&iov, &iov_cnt, s);
661 
662     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
663         goto error;
664     }
665 
666     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
667         s = iov_to_buf(iov, iov_cnt, 0, macs,
668                        mac_data.entries * ETH_ALEN);
669         if (s != mac_data.entries * ETH_ALEN) {
670             goto error;
671         }
672         in_use += mac_data.entries;
673     } else {
674         uni_overflow = 1;
675     }
676 
677     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
678 
679     first_multi = in_use;
680 
681     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
682                    sizeof(mac_data.entries));
683     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
684     if (s != sizeof(mac_data.entries)) {
685         goto error;
686     }
687 
688     iov_discard_front(&iov, &iov_cnt, s);
689 
690     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
691         goto error;
692     }
693 
694     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
695         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
696                        mac_data.entries * ETH_ALEN);
697         if (s != mac_data.entries * ETH_ALEN) {
698             goto error;
699         }
700         in_use += mac_data.entries;
701     } else {
702         multi_overflow = 1;
703     }
704 
705     n->mac_table.in_use = in_use;
706     n->mac_table.first_multi = first_multi;
707     n->mac_table.uni_overflow = uni_overflow;
708     n->mac_table.multi_overflow = multi_overflow;
709     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
710     g_free(macs);
711     rxfilter_notify(nc);
712 
713     return VIRTIO_NET_OK;
714 
715 error:
716     g_free(macs);
717     return VIRTIO_NET_ERR;
718 }
719 
720 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
721                                         struct iovec *iov, unsigned int iov_cnt)
722 {
723     VirtIODevice *vdev = VIRTIO_DEVICE(n);
724     uint16_t vid;
725     size_t s;
726     NetClientState *nc = qemu_get_queue(n->nic);
727 
728     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
729     vid = virtio_lduw_p(vdev, &vid);
730     if (s != sizeof(vid)) {
731         return VIRTIO_NET_ERR;
732     }
733 
734     if (vid >= MAX_VLAN)
735         return VIRTIO_NET_ERR;
736 
737     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
738         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
739     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
740         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
741     else
742         return VIRTIO_NET_ERR;
743 
744     rxfilter_notify(nc);
745 
746     return VIRTIO_NET_OK;
747 }
748 
749 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
750                                       struct iovec *iov, unsigned int iov_cnt)
751 {
752     if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
753         n->status & VIRTIO_NET_S_ANNOUNCE) {
754         n->status &= ~VIRTIO_NET_S_ANNOUNCE;
755         if (n->announce_counter) {
756             timer_mod(n->announce_timer,
757                       qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
758                       self_announce_delay(n->announce_counter));
759         }
760         return VIRTIO_NET_OK;
761     } else {
762         return VIRTIO_NET_ERR;
763     }
764 }
765 
766 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
767                                 struct iovec *iov, unsigned int iov_cnt)
768 {
769     VirtIODevice *vdev = VIRTIO_DEVICE(n);
770     struct virtio_net_ctrl_mq mq;
771     size_t s;
772     uint16_t queues;
773 
774     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
775     if (s != sizeof(mq)) {
776         return VIRTIO_NET_ERR;
777     }
778 
779     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
780         return VIRTIO_NET_ERR;
781     }
782 
783     queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
784 
785     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
786         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
787         queues > n->max_queues ||
788         !n->multiqueue) {
789         return VIRTIO_NET_ERR;
790     }
791 
792     n->curr_queues = queues;
793     /* stop the backend before changing the number of queues to avoid handling a
794      * disabled queue */
795     virtio_net_set_status(vdev, vdev->status);
796     virtio_net_set_queues(n);
797 
798     return VIRTIO_NET_OK;
799 }
800 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
801 {
802     VirtIONet *n = VIRTIO_NET(vdev);
803     struct virtio_net_ctrl_hdr ctrl;
804     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
805     VirtQueueElement elem;
806     size_t s;
807     struct iovec *iov, *iov2;
808     unsigned int iov_cnt;
809 
810     while (virtqueue_pop(vq, &elem)) {
811         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
812             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
813             error_report("virtio-net ctrl missing headers");
814             exit(1);
815         }
816 
817         iov_cnt = elem.out_num;
818         iov2 = iov = g_memdup(elem.out_sg, sizeof(struct iovec) * elem.out_num);
819         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
820         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
821         if (s != sizeof(ctrl)) {
822             status = VIRTIO_NET_ERR;
823         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
824             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
825         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
826             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
827         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
828             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
829         } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
830             status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
831         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
832             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
833         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
834             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
835         }
836 
837         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
838         assert(s == sizeof(status));
839 
840         virtqueue_push(vq, &elem, sizeof(status));
841         virtio_notify(vdev, vq);
842         g_free(iov2);
843     }
844 }
845 
846 /* RX */
847 
848 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
849 {
850     VirtIONet *n = VIRTIO_NET(vdev);
851     int queue_index = vq2q(virtio_get_queue_index(vq));
852 
853     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
854 }
855 
856 static int virtio_net_can_receive(NetClientState *nc)
857 {
858     VirtIONet *n = qemu_get_nic_opaque(nc);
859     VirtIODevice *vdev = VIRTIO_DEVICE(n);
860     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
861 
862     if (!vdev->vm_running) {
863         return 0;
864     }
865 
866     if (nc->queue_index >= n->curr_queues) {
867         return 0;
868     }
869 
870     if (!virtio_queue_ready(q->rx_vq) ||
871         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
872         return 0;
873     }
874 
875     return 1;
876 }
877 
878 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
879 {
880     VirtIONet *n = q->n;
881     if (virtio_queue_empty(q->rx_vq) ||
882         (n->mergeable_rx_bufs &&
883          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
884         virtio_queue_set_notification(q->rx_vq, 1);
885 
886         /* To avoid a race condition where the guest has made some buffers
887          * available after the above check but before notification was
888          * enabled, check for available buffers again.
889          */
890         if (virtio_queue_empty(q->rx_vq) ||
891             (n->mergeable_rx_bufs &&
892              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
893             return 0;
894         }
895     }
896 
897     virtio_queue_set_notification(q->rx_vq, 0);
898     return 1;
899 }
900 
901 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
902 {
903     virtio_tswap16s(vdev, &hdr->hdr_len);
904     virtio_tswap16s(vdev, &hdr->gso_size);
905     virtio_tswap16s(vdev, &hdr->csum_start);
906     virtio_tswap16s(vdev, &hdr->csum_offset);
907 }
908 
909 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
910  * it never finds out that the packets don't have valid checksums.  This
911  * causes dhclient to get upset.  Fedora's carried a patch for ages to
912  * fix this with Xen but it hasn't appeared in an upstream release of
913  * dhclient yet.
914  *
915  * To avoid breaking existing guests, we catch udp packets and add
916  * checksums.  This is terrible but it's better than hacking the guest
917  * kernels.
918  *
919  * N.B. if we introduce a zero-copy API, this operation is no longer free so
920  * we should provide a mechanism to disable it to avoid polluting the host
921  * cache.
922  */
923 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
924                                         uint8_t *buf, size_t size)
925 {
926     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
927         (size > 27 && size < 1500) && /* normal sized MTU */
928         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
929         (buf[23] == 17) && /* ip.protocol == UDP */
930         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
931         net_checksum_calculate(buf, size);
932         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
933     }
934 }
935 
936 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
937                            const void *buf, size_t size)
938 {
939     if (n->has_vnet_hdr) {
940         /* FIXME this cast is evil */
941         void *wbuf = (void *)buf;
942         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
943                                     size - n->host_hdr_len);
944         virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
945         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
946     } else {
947         struct virtio_net_hdr hdr = {
948             .flags = 0,
949             .gso_type = VIRTIO_NET_HDR_GSO_NONE
950         };
951         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
952     }
953 }
954 
955 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
956 {
957     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
958     static const uint8_t vlan[] = {0x81, 0x00};
959     uint8_t *ptr = (uint8_t *)buf;
960     int i;
961 
962     if (n->promisc)
963         return 1;
964 
965     ptr += n->host_hdr_len;
966 
967     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
968         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
969         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
970             return 0;
971     }
972 
973     if (ptr[0] & 1) { // multicast
974         if (!memcmp(ptr, bcast, sizeof(bcast))) {
975             return !n->nobcast;
976         } else if (n->nomulti) {
977             return 0;
978         } else if (n->allmulti || n->mac_table.multi_overflow) {
979             return 1;
980         }
981 
982         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
983             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
984                 return 1;
985             }
986         }
987     } else { // unicast
988         if (n->nouni) {
989             return 0;
990         } else if (n->alluni || n->mac_table.uni_overflow) {
991             return 1;
992         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
993             return 1;
994         }
995 
996         for (i = 0; i < n->mac_table.first_multi; i++) {
997             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
998                 return 1;
999             }
1000         }
1001     }
1002 
1003     return 0;
1004 }
1005 
1006 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
1007 {
1008     VirtIONet *n = qemu_get_nic_opaque(nc);
1009     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1010     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1011     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1012     struct virtio_net_hdr_mrg_rxbuf mhdr;
1013     unsigned mhdr_cnt = 0;
1014     size_t offset, i, guest_offset;
1015 
1016     if (!virtio_net_can_receive(nc)) {
1017         return -1;
1018     }
1019 
1020     /* hdr_len refers to the header we supply to the guest */
1021     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1022         return 0;
1023     }
1024 
1025     if (!receive_filter(n, buf, size))
1026         return size;
1027 
1028     offset = i = 0;
1029 
1030     while (offset < size) {
1031         VirtQueueElement elem;
1032         int len, total;
1033         const struct iovec *sg = elem.in_sg;
1034 
1035         total = 0;
1036 
1037         if (virtqueue_pop(q->rx_vq, &elem) == 0) {
1038             if (i == 0)
1039                 return -1;
1040             error_report("virtio-net unexpected empty queue: "
1041                     "i %zd mergeable %d offset %zd, size %zd, "
1042                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
1043                     i, n->mergeable_rx_bufs, offset, size,
1044                     n->guest_hdr_len, n->host_hdr_len, vdev->guest_features);
1045             exit(1);
1046         }
1047 
1048         if (elem.in_num < 1) {
1049             error_report("virtio-net receive queue contains no in buffers");
1050             exit(1);
1051         }
1052 
1053         if (i == 0) {
1054             assert(offset == 0);
1055             if (n->mergeable_rx_bufs) {
1056                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1057                                     sg, elem.in_num,
1058                                     offsetof(typeof(mhdr), num_buffers),
1059                                     sizeof(mhdr.num_buffers));
1060             }
1061 
1062             receive_header(n, sg, elem.in_num, buf, size);
1063             offset = n->host_hdr_len;
1064             total += n->guest_hdr_len;
1065             guest_offset = n->guest_hdr_len;
1066         } else {
1067             guest_offset = 0;
1068         }
1069 
1070         /* copy in packet.  ugh */
1071         len = iov_from_buf(sg, elem.in_num, guest_offset,
1072                            buf + offset, size - offset);
1073         total += len;
1074         offset += len;
1075         /* If buffers can't be merged, at this point we
1076          * must have consumed the complete packet.
1077          * Otherwise, drop it. */
1078         if (!n->mergeable_rx_bufs && offset < size) {
1079 #if 0
1080             error_report("virtio-net truncated non-mergeable packet: "
1081                          "i %zd mergeable %d offset %zd, size %zd, "
1082                          "guest hdr len %zd, host hdr len %zd",
1083                          i, n->mergeable_rx_bufs,
1084                          offset, size, n->guest_hdr_len, n->host_hdr_len);
1085 #endif
1086             return size;
1087         }
1088 
1089         /* signal other side */
1090         virtqueue_fill(q->rx_vq, &elem, total, i++);
1091     }
1092 
1093     if (mhdr_cnt) {
1094         virtio_stw_p(vdev, &mhdr.num_buffers, i);
1095         iov_from_buf(mhdr_sg, mhdr_cnt,
1096                      0,
1097                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1098     }
1099 
1100     virtqueue_flush(q->rx_vq, i);
1101     virtio_notify(vdev, q->rx_vq);
1102 
1103     return size;
1104 }
1105 
1106 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1107 
1108 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1109 {
1110     VirtIONet *n = qemu_get_nic_opaque(nc);
1111     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1112     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1113 
1114     virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
1115     virtio_notify(vdev, q->tx_vq);
1116 
1117     q->async_tx.elem.out_num = q->async_tx.len = 0;
1118 
1119     virtio_queue_set_notification(q->tx_vq, 1);
1120     virtio_net_flush_tx(q);
1121 }
1122 
1123 /* TX */
1124 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
1125 {
1126     VirtIONet *n = q->n;
1127     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1128     VirtQueueElement elem;
1129     int32_t num_packets = 0;
1130     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
1131     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1132         return num_packets;
1133     }
1134 
1135     if (q->async_tx.elem.out_num) {
1136         virtio_queue_set_notification(q->tx_vq, 0);
1137         return num_packets;
1138     }
1139 
1140     while (virtqueue_pop(q->tx_vq, &elem)) {
1141         ssize_t ret, len;
1142         unsigned int out_num = elem.out_num;
1143         struct iovec *out_sg = &elem.out_sg[0];
1144         struct iovec sg[VIRTQUEUE_MAX_SIZE];
1145 
1146         if (out_num < 1) {
1147             error_report("virtio-net header not in first element");
1148             exit(1);
1149         }
1150 
1151         if (n->has_vnet_hdr) {
1152             if (out_sg[0].iov_len < n->guest_hdr_len) {
1153                 error_report("virtio-net header incorrect");
1154                 exit(1);
1155             }
1156             virtio_net_hdr_swap(vdev, (void *) out_sg[0].iov_base);
1157         }
1158 
1159         /*
1160          * If host wants to see the guest header as is, we can
1161          * pass it on unchanged. Otherwise, copy just the parts
1162          * that host is interested in.
1163          */
1164         assert(n->host_hdr_len <= n->guest_hdr_len);
1165         if (n->host_hdr_len != n->guest_hdr_len) {
1166             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
1167                                        out_sg, out_num,
1168                                        0, n->host_hdr_len);
1169             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
1170                              out_sg, out_num,
1171                              n->guest_hdr_len, -1);
1172             out_num = sg_num;
1173             out_sg = sg;
1174         }
1175 
1176         len = n->guest_hdr_len;
1177 
1178         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
1179                                       out_sg, out_num, virtio_net_tx_complete);
1180         if (ret == 0) {
1181             virtio_queue_set_notification(q->tx_vq, 0);
1182             q->async_tx.elem = elem;
1183             q->async_tx.len  = len;
1184             return -EBUSY;
1185         }
1186 
1187         len += ret;
1188 
1189         virtqueue_push(q->tx_vq, &elem, 0);
1190         virtio_notify(vdev, q->tx_vq);
1191 
1192         if (++num_packets >= n->tx_burst) {
1193             break;
1194         }
1195     }
1196     return num_packets;
1197 }
1198 
1199 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
1200 {
1201     VirtIONet *n = VIRTIO_NET(vdev);
1202     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1203 
1204     /* This happens when device was stopped but VCPU wasn't. */
1205     if (!vdev->vm_running) {
1206         q->tx_waiting = 1;
1207         return;
1208     }
1209 
1210     if (q->tx_waiting) {
1211         virtio_queue_set_notification(vq, 1);
1212         timer_del(q->tx_timer);
1213         q->tx_waiting = 0;
1214         virtio_net_flush_tx(q);
1215     } else {
1216         timer_mod(q->tx_timer,
1217                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
1218         q->tx_waiting = 1;
1219         virtio_queue_set_notification(vq, 0);
1220     }
1221 }
1222 
1223 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1224 {
1225     VirtIONet *n = VIRTIO_NET(vdev);
1226     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1227 
1228     if (unlikely(q->tx_waiting)) {
1229         return;
1230     }
1231     q->tx_waiting = 1;
1232     /* This happens when device was stopped but VCPU wasn't. */
1233     if (!vdev->vm_running) {
1234         return;
1235     }
1236     virtio_queue_set_notification(vq, 0);
1237     qemu_bh_schedule(q->tx_bh);
1238 }
1239 
1240 static void virtio_net_tx_timer(void *opaque)
1241 {
1242     VirtIONetQueue *q = opaque;
1243     VirtIONet *n = q->n;
1244     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1245     /* This happens when device was stopped but BH wasn't. */
1246     if (!vdev->vm_running) {
1247         /* Make sure tx waiting is set, so we'll run when restarted. */
1248         assert(q->tx_waiting);
1249         return;
1250     }
1251 
1252     q->tx_waiting = 0;
1253 
1254     /* Just in case the driver is not ready on more */
1255     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1256         return;
1257     }
1258 
1259     virtio_queue_set_notification(q->tx_vq, 1);
1260     virtio_net_flush_tx(q);
1261 }
1262 
1263 static void virtio_net_tx_bh(void *opaque)
1264 {
1265     VirtIONetQueue *q = opaque;
1266     VirtIONet *n = q->n;
1267     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1268     int32_t ret;
1269 
1270     /* This happens when device was stopped but BH wasn't. */
1271     if (!vdev->vm_running) {
1272         /* Make sure tx waiting is set, so we'll run when restarted. */
1273         assert(q->tx_waiting);
1274         return;
1275     }
1276 
1277     q->tx_waiting = 0;
1278 
1279     /* Just in case the driver is not ready on more */
1280     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1281         return;
1282     }
1283 
1284     ret = virtio_net_flush_tx(q);
1285     if (ret == -EBUSY) {
1286         return; /* Notification re-enable handled by tx_complete */
1287     }
1288 
1289     /* If we flush a full burst of packets, assume there are
1290      * more coming and immediately reschedule */
1291     if (ret >= n->tx_burst) {
1292         qemu_bh_schedule(q->tx_bh);
1293         q->tx_waiting = 1;
1294         return;
1295     }
1296 
1297     /* If less than a full burst, re-enable notification and flush
1298      * anything that may have come in while we weren't looking.  If
1299      * we find something, assume the guest is still active and reschedule */
1300     virtio_queue_set_notification(q->tx_vq, 1);
1301     if (virtio_net_flush_tx(q) > 0) {
1302         virtio_queue_set_notification(q->tx_vq, 0);
1303         qemu_bh_schedule(q->tx_bh);
1304         q->tx_waiting = 1;
1305     }
1306 }
1307 
1308 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1309 {
1310     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1311     int i, max = multiqueue ? n->max_queues : 1;
1312 
1313     n->multiqueue = multiqueue;
1314 
1315     for (i = 2; i < n->max_queues * 2 + 1; i++) {
1316         virtio_del_queue(vdev, i);
1317     }
1318 
1319     for (i = 1; i < max; i++) {
1320         n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1321         if (n->vqs[i].tx_timer) {
1322             n->vqs[i].tx_vq =
1323                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1324             n->vqs[i].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
1325                                                    virtio_net_tx_timer,
1326                                                    &n->vqs[i]);
1327         } else {
1328             n->vqs[i].tx_vq =
1329                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1330             n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1331         }
1332 
1333         n->vqs[i].tx_waiting = 0;
1334         n->vqs[i].n = n;
1335     }
1336 
1337     /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack
1338      * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid
1339      * breaking them.
1340      */
1341     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1342 
1343     virtio_net_set_queues(n);
1344 }
1345 
1346 static void virtio_net_save(QEMUFile *f, void *opaque)
1347 {
1348     VirtIONet *n = opaque;
1349     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1350 
1351     /* At this point, backend must be stopped, otherwise
1352      * it might keep writing to memory. */
1353     assert(!n->vhost_started);
1354     virtio_save(vdev, f);
1355 }
1356 
1357 static void virtio_net_save_device(VirtIODevice *vdev, QEMUFile *f)
1358 {
1359     VirtIONet *n = VIRTIO_NET(vdev);
1360     int i;
1361 
1362     qemu_put_buffer(f, n->mac, ETH_ALEN);
1363     qemu_put_be32(f, n->vqs[0].tx_waiting);
1364     qemu_put_be32(f, n->mergeable_rx_bufs);
1365     qemu_put_be16(f, n->status);
1366     qemu_put_byte(f, n->promisc);
1367     qemu_put_byte(f, n->allmulti);
1368     qemu_put_be32(f, n->mac_table.in_use);
1369     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1370     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1371     qemu_put_be32(f, n->has_vnet_hdr);
1372     qemu_put_byte(f, n->mac_table.multi_overflow);
1373     qemu_put_byte(f, n->mac_table.uni_overflow);
1374     qemu_put_byte(f, n->alluni);
1375     qemu_put_byte(f, n->nomulti);
1376     qemu_put_byte(f, n->nouni);
1377     qemu_put_byte(f, n->nobcast);
1378     qemu_put_byte(f, n->has_ufo);
1379     if (n->max_queues > 1) {
1380         qemu_put_be16(f, n->max_queues);
1381         qemu_put_be16(f, n->curr_queues);
1382         for (i = 1; i < n->curr_queues; i++) {
1383             qemu_put_be32(f, n->vqs[i].tx_waiting);
1384         }
1385     }
1386 
1387     if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1388         qemu_put_be64(f, n->curr_guest_offloads);
1389     }
1390 }
1391 
1392 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1393 {
1394     VirtIONet *n = opaque;
1395     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1396 
1397     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1398         return -EINVAL;
1399 
1400     return virtio_load(vdev, f, version_id);
1401 }
1402 
1403 static int virtio_net_load_device(VirtIODevice *vdev, QEMUFile *f,
1404                                   int version_id)
1405 {
1406     VirtIONet *n = VIRTIO_NET(vdev);
1407     int i, link_down;
1408 
1409     qemu_get_buffer(f, n->mac, ETH_ALEN);
1410     n->vqs[0].tx_waiting = qemu_get_be32(f);
1411 
1412     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1413 
1414     if (version_id >= 3)
1415         n->status = qemu_get_be16(f);
1416 
1417     if (version_id >= 4) {
1418         if (version_id < 8) {
1419             n->promisc = qemu_get_be32(f);
1420             n->allmulti = qemu_get_be32(f);
1421         } else {
1422             n->promisc = qemu_get_byte(f);
1423             n->allmulti = qemu_get_byte(f);
1424         }
1425     }
1426 
1427     if (version_id >= 5) {
1428         n->mac_table.in_use = qemu_get_be32(f);
1429         /* MAC_TABLE_ENTRIES may be different from the saved image */
1430         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1431             qemu_get_buffer(f, n->mac_table.macs,
1432                             n->mac_table.in_use * ETH_ALEN);
1433         } else {
1434             int64_t i;
1435 
1436             /* Overflow detected - can happen if source has a larger MAC table.
1437              * We simply set overflow flag so there's no need to maintain the
1438              * table of addresses, discard them all.
1439              * Note: 64 bit math to avoid integer overflow.
1440              */
1441             for (i = 0; i < (int64_t)n->mac_table.in_use * ETH_ALEN; ++i) {
1442                 qemu_get_byte(f);
1443             }
1444             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1445             n->mac_table.in_use = 0;
1446         }
1447     }
1448 
1449     if (version_id >= 6)
1450         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1451 
1452     if (version_id >= 7) {
1453         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1454             error_report("virtio-net: saved image requires vnet_hdr=on");
1455             return -1;
1456         }
1457     }
1458 
1459     if (version_id >= 9) {
1460         n->mac_table.multi_overflow = qemu_get_byte(f);
1461         n->mac_table.uni_overflow = qemu_get_byte(f);
1462     }
1463 
1464     if (version_id >= 10) {
1465         n->alluni = qemu_get_byte(f);
1466         n->nomulti = qemu_get_byte(f);
1467         n->nouni = qemu_get_byte(f);
1468         n->nobcast = qemu_get_byte(f);
1469     }
1470 
1471     if (version_id >= 11) {
1472         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1473             error_report("virtio-net: saved image requires TUN_F_UFO support");
1474             return -1;
1475         }
1476     }
1477 
1478     if (n->max_queues > 1) {
1479         if (n->max_queues != qemu_get_be16(f)) {
1480             error_report("virtio-net: different max_queues ");
1481             return -1;
1482         }
1483 
1484         n->curr_queues = qemu_get_be16(f);
1485         if (n->curr_queues > n->max_queues) {
1486             error_report("virtio-net: curr_queues %x > max_queues %x",
1487                          n->curr_queues, n->max_queues);
1488             return -1;
1489         }
1490         for (i = 1; i < n->curr_queues; i++) {
1491             n->vqs[i].tx_waiting = qemu_get_be32(f);
1492         }
1493     }
1494 
1495     if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1496         n->curr_guest_offloads = qemu_get_be64(f);
1497     } else {
1498         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
1499     }
1500 
1501     if (peer_has_vnet_hdr(n)) {
1502         virtio_net_apply_guest_offloads(n);
1503     }
1504 
1505     virtio_net_set_queues(n);
1506 
1507     /* Find the first multicast entry in the saved MAC filter */
1508     for (i = 0; i < n->mac_table.in_use; i++) {
1509         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1510             break;
1511         }
1512     }
1513     n->mac_table.first_multi = i;
1514 
1515     /* nc.link_down can't be migrated, so infer link_down according
1516      * to link status bit in n->status */
1517     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1518     for (i = 0; i < n->max_queues; i++) {
1519         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1520     }
1521 
1522     if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
1523         virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
1524         n->announce_counter = SELF_ANNOUNCE_ROUNDS;
1525         timer_mod(n->announce_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL));
1526     }
1527 
1528     return 0;
1529 }
1530 
1531 static NetClientInfo net_virtio_info = {
1532     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1533     .size = sizeof(NICState),
1534     .can_receive = virtio_net_can_receive,
1535     .receive = virtio_net_receive,
1536     .link_status_changed = virtio_net_set_link_status,
1537     .query_rx_filter = virtio_net_query_rxfilter,
1538 };
1539 
1540 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1541 {
1542     VirtIONet *n = VIRTIO_NET(vdev);
1543     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1544     assert(n->vhost_started);
1545     return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
1546 }
1547 
1548 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1549                                            bool mask)
1550 {
1551     VirtIONet *n = VIRTIO_NET(vdev);
1552     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1553     assert(n->vhost_started);
1554     vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
1555                              vdev, idx, mask);
1556 }
1557 
1558 static void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features)
1559 {
1560     int i, config_size = 0;
1561     virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
1562     for (i = 0; feature_sizes[i].flags != 0; i++) {
1563         if (host_features & feature_sizes[i].flags) {
1564             config_size = MAX(feature_sizes[i].end, config_size);
1565         }
1566     }
1567     n->config_size = config_size;
1568 }
1569 
1570 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
1571                                    const char *type)
1572 {
1573     /*
1574      * The name can be NULL, the netclient name will be type.x.
1575      */
1576     assert(type != NULL);
1577 
1578     g_free(n->netclient_name);
1579     g_free(n->netclient_type);
1580     n->netclient_name = g_strdup(name);
1581     n->netclient_type = g_strdup(type);
1582 }
1583 
1584 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
1585 {
1586     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1587     VirtIONet *n = VIRTIO_NET(dev);
1588     NetClientState *nc;
1589     int i;
1590 
1591     virtio_net_set_config_size(n, n->host_features);
1592     virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
1593 
1594     n->max_queues = MAX(n->nic_conf.peers.queues, 1);
1595     if (n->max_queues * 2 + 1 > VIRTIO_PCI_QUEUE_MAX) {
1596         error_setg(errp, "Invalid number of queues (= %" PRIu32 "), "
1597                    "must be a postive integer less than %d.",
1598                    n->max_queues, (VIRTIO_PCI_QUEUE_MAX - 1) / 2);
1599         virtio_cleanup(vdev);
1600         return;
1601     }
1602     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1603     n->vqs[0].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1604     n->curr_queues = 1;
1605     n->vqs[0].n = n;
1606     n->tx_timeout = n->net_conf.txtimer;
1607 
1608     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1609                        && strcmp(n->net_conf.tx, "bh")) {
1610         error_report("virtio-net: "
1611                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1612                      n->net_conf.tx);
1613         error_report("Defaulting to \"bh\"");
1614     }
1615 
1616     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1617         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1618                                            virtio_net_handle_tx_timer);
1619         n->vqs[0].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, virtio_net_tx_timer,
1620                                                &n->vqs[0]);
1621     } else {
1622         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1623                                            virtio_net_handle_tx_bh);
1624         n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]);
1625     }
1626     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1627     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1628     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1629     n->status = VIRTIO_NET_S_LINK_UP;
1630     n->announce_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
1631                                      virtio_net_announce_timer, n);
1632 
1633     if (n->netclient_type) {
1634         /*
1635          * Happen when virtio_net_set_netclient_name has been called.
1636          */
1637         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1638                               n->netclient_type, n->netclient_name, n);
1639     } else {
1640         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1641                               object_get_typename(OBJECT(dev)), dev->id, n);
1642     }
1643 
1644     peer_test_vnet_hdr(n);
1645     if (peer_has_vnet_hdr(n)) {
1646         for (i = 0; i < n->max_queues; i++) {
1647             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1648         }
1649         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1650     } else {
1651         n->host_hdr_len = 0;
1652     }
1653 
1654     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1655 
1656     n->vqs[0].tx_waiting = 0;
1657     n->tx_burst = n->net_conf.txburst;
1658     virtio_net_set_mrg_rx_bufs(n, 0);
1659     n->promisc = 1; /* for compatibility */
1660 
1661     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1662 
1663     n->vlans = g_malloc0(MAX_VLAN >> 3);
1664 
1665     nc = qemu_get_queue(n->nic);
1666     nc->rxfilter_notify_enabled = 1;
1667 
1668     n->qdev = dev;
1669     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1670                     virtio_net_save, virtio_net_load, n);
1671 }
1672 
1673 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
1674 {
1675     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1676     VirtIONet *n = VIRTIO_NET(dev);
1677     int i;
1678 
1679     /* This will stop vhost backend if appropriate. */
1680     virtio_net_set_status(vdev, 0);
1681 
1682     unregister_savevm(dev, "virtio-net", n);
1683 
1684     g_free(n->netclient_name);
1685     n->netclient_name = NULL;
1686     g_free(n->netclient_type);
1687     n->netclient_type = NULL;
1688 
1689     g_free(n->mac_table.macs);
1690     g_free(n->vlans);
1691 
1692     for (i = 0; i < n->max_queues; i++) {
1693         VirtIONetQueue *q = &n->vqs[i];
1694         NetClientState *nc = qemu_get_subqueue(n->nic, i);
1695 
1696         qemu_purge_queued_packets(nc);
1697 
1698         if (q->tx_timer) {
1699             timer_del(q->tx_timer);
1700             timer_free(q->tx_timer);
1701         } else if (q->tx_bh) {
1702             qemu_bh_delete(q->tx_bh);
1703         }
1704     }
1705 
1706     timer_del(n->announce_timer);
1707     timer_free(n->announce_timer);
1708     g_free(n->vqs);
1709     qemu_del_nic(n->nic);
1710     virtio_cleanup(vdev);
1711 }
1712 
1713 static void virtio_net_instance_init(Object *obj)
1714 {
1715     VirtIONet *n = VIRTIO_NET(obj);
1716 
1717     /*
1718      * The default config_size is sizeof(struct virtio_net_config).
1719      * Can be overriden with virtio_net_set_config_size.
1720      */
1721     n->config_size = sizeof(struct virtio_net_config);
1722     device_add_bootindex_property(obj, &n->nic_conf.bootindex,
1723                                   "bootindex", "/ethernet-phy@0",
1724                                   DEVICE(n), NULL);
1725 }
1726 
1727 static Property virtio_net_properties[] = {
1728     DEFINE_VIRTIO_NET_FEATURES(VirtIONet, host_features),
1729     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
1730     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
1731                                                TX_TIMER_INTERVAL),
1732     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
1733     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
1734     DEFINE_PROP_END_OF_LIST(),
1735 };
1736 
1737 static void virtio_net_class_init(ObjectClass *klass, void *data)
1738 {
1739     DeviceClass *dc = DEVICE_CLASS(klass);
1740     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1741 
1742     dc->props = virtio_net_properties;
1743     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
1744     vdc->realize = virtio_net_device_realize;
1745     vdc->unrealize = virtio_net_device_unrealize;
1746     vdc->get_config = virtio_net_get_config;
1747     vdc->set_config = virtio_net_set_config;
1748     vdc->get_features = virtio_net_get_features;
1749     vdc->set_features = virtio_net_set_features;
1750     vdc->bad_features = virtio_net_bad_features;
1751     vdc->reset = virtio_net_reset;
1752     vdc->set_status = virtio_net_set_status;
1753     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
1754     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
1755     vdc->load = virtio_net_load_device;
1756     vdc->save = virtio_net_save_device;
1757 }
1758 
1759 static const TypeInfo virtio_net_info = {
1760     .name = TYPE_VIRTIO_NET,
1761     .parent = TYPE_VIRTIO_DEVICE,
1762     .instance_size = sizeof(VirtIONet),
1763     .instance_init = virtio_net_instance_init,
1764     .class_init = virtio_net_class_init,
1765 };
1766 
1767 static void virtio_register_types(void)
1768 {
1769     type_register_static(&virtio_net_info);
1770 }
1771 
1772 type_init(virtio_register_types)
1773