xref: /openbmc/qemu/hw/net/virtio-net.c (revision 116694c3)
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 intList *get_vlan_table(VirtIONet *n)
230 {
231     intList *list, *entry;
232     int i, j;
233 
234     list = NULL;
235     for (i = 0; i < MAX_VLAN >> 5; i++) {
236         for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
237             if (n->vlans[i] & (1U << j)) {
238                 entry = g_malloc0(sizeof(*entry));
239                 entry->value = (i << 5) + j;
240                 entry->next = list;
241                 list = entry;
242             }
243         }
244     }
245 
246     return list;
247 }
248 
249 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
250 {
251     VirtIONet *n = qemu_get_nic_opaque(nc);
252     VirtIODevice *vdev = VIRTIO_DEVICE(n);
253     RxFilterInfo *info;
254     strList *str_list, *entry;
255     int i;
256 
257     info = g_malloc0(sizeof(*info));
258     info->name = g_strdup(nc->name);
259     info->promiscuous = n->promisc;
260 
261     if (n->nouni) {
262         info->unicast = RX_STATE_NONE;
263     } else if (n->alluni) {
264         info->unicast = RX_STATE_ALL;
265     } else {
266         info->unicast = RX_STATE_NORMAL;
267     }
268 
269     if (n->nomulti) {
270         info->multicast = RX_STATE_NONE;
271     } else if (n->allmulti) {
272         info->multicast = RX_STATE_ALL;
273     } else {
274         info->multicast = RX_STATE_NORMAL;
275     }
276 
277     info->broadcast_allowed = n->nobcast;
278     info->multicast_overflow = n->mac_table.multi_overflow;
279     info->unicast_overflow = n->mac_table.uni_overflow;
280 
281     info->main_mac = qemu_mac_strdup_printf(n->mac);
282 
283     str_list = NULL;
284     for (i = 0; i < n->mac_table.first_multi; i++) {
285         entry = g_malloc0(sizeof(*entry));
286         entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
287         entry->next = str_list;
288         str_list = entry;
289     }
290     info->unicast_table = str_list;
291 
292     str_list = NULL;
293     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
294         entry = g_malloc0(sizeof(*entry));
295         entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
296         entry->next = str_list;
297         str_list = entry;
298     }
299     info->multicast_table = str_list;
300     info->vlan_table = get_vlan_table(n);
301 
302     if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
303         info->vlan = RX_STATE_ALL;
304     } else if (!info->vlan_table) {
305         info->vlan = RX_STATE_NONE;
306     } else {
307         info->vlan = RX_STATE_NORMAL;
308     }
309 
310     /* enable event notification after query */
311     nc->rxfilter_notify_enabled = 1;
312 
313     return info;
314 }
315 
316 static void virtio_net_reset(VirtIODevice *vdev)
317 {
318     VirtIONet *n = VIRTIO_NET(vdev);
319 
320     /* Reset back to compatibility mode */
321     n->promisc = 1;
322     n->allmulti = 0;
323     n->alluni = 0;
324     n->nomulti = 0;
325     n->nouni = 0;
326     n->nobcast = 0;
327     /* multiqueue is disabled by default */
328     n->curr_queues = 1;
329     timer_del(n->announce_timer);
330     n->announce_counter = 0;
331     n->status &= ~VIRTIO_NET_S_ANNOUNCE;
332 
333     /* Flush any MAC and VLAN filter table state */
334     n->mac_table.in_use = 0;
335     n->mac_table.first_multi = 0;
336     n->mac_table.multi_overflow = 0;
337     n->mac_table.uni_overflow = 0;
338     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
339     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
340     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
341     memset(n->vlans, 0, MAX_VLAN >> 3);
342 }
343 
344 static void peer_test_vnet_hdr(VirtIONet *n)
345 {
346     NetClientState *nc = qemu_get_queue(n->nic);
347     if (!nc->peer) {
348         return;
349     }
350 
351     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
352 }
353 
354 static int peer_has_vnet_hdr(VirtIONet *n)
355 {
356     return n->has_vnet_hdr;
357 }
358 
359 static int peer_has_ufo(VirtIONet *n)
360 {
361     if (!peer_has_vnet_hdr(n))
362         return 0;
363 
364     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
365 
366     return n->has_ufo;
367 }
368 
369 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs)
370 {
371     int i;
372     NetClientState *nc;
373 
374     n->mergeable_rx_bufs = mergeable_rx_bufs;
375 
376     n->guest_hdr_len = n->mergeable_rx_bufs ?
377         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
378 
379     for (i = 0; i < n->max_queues; i++) {
380         nc = qemu_get_subqueue(n->nic, i);
381 
382         if (peer_has_vnet_hdr(n) &&
383             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
384             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
385             n->host_hdr_len = n->guest_hdr_len;
386         }
387     }
388 }
389 
390 static int peer_attach(VirtIONet *n, int index)
391 {
392     NetClientState *nc = qemu_get_subqueue(n->nic, index);
393 
394     if (!nc->peer) {
395         return 0;
396     }
397 
398     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
399         return 0;
400     }
401 
402     return tap_enable(nc->peer);
403 }
404 
405 static int peer_detach(VirtIONet *n, int index)
406 {
407     NetClientState *nc = qemu_get_subqueue(n->nic, index);
408 
409     if (!nc->peer) {
410         return 0;
411     }
412 
413     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
414         return 0;
415     }
416 
417     return tap_disable(nc->peer);
418 }
419 
420 static void virtio_net_set_queues(VirtIONet *n)
421 {
422     int i;
423     int r;
424 
425     for (i = 0; i < n->max_queues; i++) {
426         if (i < n->curr_queues) {
427             r = peer_attach(n, i);
428             assert(!r);
429         } else {
430             r = peer_detach(n, i);
431             assert(!r);
432         }
433     }
434 }
435 
436 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
437 
438 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
439 {
440     VirtIONet *n = VIRTIO_NET(vdev);
441     NetClientState *nc = qemu_get_queue(n->nic);
442 
443     /* Firstly sync all virtio-net possible supported features */
444     features |= n->host_features;
445 
446     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
447 
448     if (!peer_has_vnet_hdr(n)) {
449         virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
450         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
451         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
452         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
453 
454         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
455         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
456         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
457         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
458     }
459 
460     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
461         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
462         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
463     }
464 
465     if (!get_vhost_net(nc->peer)) {
466         return features;
467     }
468     return vhost_net_get_features(get_vhost_net(nc->peer), features);
469 }
470 
471 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
472 {
473     uint32_t features = 0;
474 
475     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
476      * but also these: */
477     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
478     virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
479     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
480     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
481     virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
482 
483     return features;
484 }
485 
486 static void virtio_net_apply_guest_offloads(VirtIONet *n)
487 {
488     qemu_set_offload(qemu_get_queue(n->nic)->peer,
489             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
490             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
491             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
492             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
493             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
494 }
495 
496 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
497 {
498     static const uint64_t guest_offloads_mask =
499         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
500         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
501         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
502         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
503         (1ULL << VIRTIO_NET_F_GUEST_UFO);
504 
505     return guest_offloads_mask & features;
506 }
507 
508 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
509 {
510     VirtIODevice *vdev = VIRTIO_DEVICE(n);
511     return virtio_net_guest_offloads_by_features(vdev->guest_features);
512 }
513 
514 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
515 {
516     VirtIONet *n = VIRTIO_NET(vdev);
517     int i;
518 
519     virtio_net_set_multiqueue(n,
520                               __virtio_has_feature(features, VIRTIO_NET_F_MQ));
521 
522     virtio_net_set_mrg_rx_bufs(n,
523                                __virtio_has_feature(features,
524                                                     VIRTIO_NET_F_MRG_RXBUF));
525 
526     if (n->has_vnet_hdr) {
527         n->curr_guest_offloads =
528             virtio_net_guest_offloads_by_features(features);
529         virtio_net_apply_guest_offloads(n);
530     }
531 
532     for (i = 0;  i < n->max_queues; i++) {
533         NetClientState *nc = qemu_get_subqueue(n->nic, i);
534 
535         if (!get_vhost_net(nc->peer)) {
536             continue;
537         }
538         vhost_net_ack_features(get_vhost_net(nc->peer), features);
539     }
540 
541     if (__virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
542         memset(n->vlans, 0, MAX_VLAN >> 3);
543     } else {
544         memset(n->vlans, 0xff, MAX_VLAN >> 3);
545     }
546 }
547 
548 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
549                                      struct iovec *iov, unsigned int iov_cnt)
550 {
551     uint8_t on;
552     size_t s;
553     NetClientState *nc = qemu_get_queue(n->nic);
554 
555     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
556     if (s != sizeof(on)) {
557         return VIRTIO_NET_ERR;
558     }
559 
560     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
561         n->promisc = on;
562     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
563         n->allmulti = on;
564     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
565         n->alluni = on;
566     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
567         n->nomulti = on;
568     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
569         n->nouni = on;
570     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
571         n->nobcast = on;
572     } else {
573         return VIRTIO_NET_ERR;
574     }
575 
576     rxfilter_notify(nc);
577 
578     return VIRTIO_NET_OK;
579 }
580 
581 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
582                                      struct iovec *iov, unsigned int iov_cnt)
583 {
584     VirtIODevice *vdev = VIRTIO_DEVICE(n);
585     uint64_t offloads;
586     size_t s;
587 
588     if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
589         return VIRTIO_NET_ERR;
590     }
591 
592     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
593     if (s != sizeof(offloads)) {
594         return VIRTIO_NET_ERR;
595     }
596 
597     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
598         uint64_t supported_offloads;
599 
600         if (!n->has_vnet_hdr) {
601             return VIRTIO_NET_ERR;
602         }
603 
604         supported_offloads = virtio_net_supported_guest_offloads(n);
605         if (offloads & ~supported_offloads) {
606             return VIRTIO_NET_ERR;
607         }
608 
609         n->curr_guest_offloads = offloads;
610         virtio_net_apply_guest_offloads(n);
611 
612         return VIRTIO_NET_OK;
613     } else {
614         return VIRTIO_NET_ERR;
615     }
616 }
617 
618 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
619                                  struct iovec *iov, unsigned int iov_cnt)
620 {
621     VirtIODevice *vdev = VIRTIO_DEVICE(n);
622     struct virtio_net_ctrl_mac mac_data;
623     size_t s;
624     NetClientState *nc = qemu_get_queue(n->nic);
625 
626     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
627         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
628             return VIRTIO_NET_ERR;
629         }
630         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
631         assert(s == sizeof(n->mac));
632         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
633         rxfilter_notify(nc);
634 
635         return VIRTIO_NET_OK;
636     }
637 
638     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
639         return VIRTIO_NET_ERR;
640     }
641 
642     int in_use = 0;
643     int first_multi = 0;
644     uint8_t uni_overflow = 0;
645     uint8_t multi_overflow = 0;
646     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
647 
648     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
649                    sizeof(mac_data.entries));
650     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
651     if (s != sizeof(mac_data.entries)) {
652         goto error;
653     }
654     iov_discard_front(&iov, &iov_cnt, s);
655 
656     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
657         goto error;
658     }
659 
660     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
661         s = iov_to_buf(iov, iov_cnt, 0, macs,
662                        mac_data.entries * ETH_ALEN);
663         if (s != mac_data.entries * ETH_ALEN) {
664             goto error;
665         }
666         in_use += mac_data.entries;
667     } else {
668         uni_overflow = 1;
669     }
670 
671     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
672 
673     first_multi = in_use;
674 
675     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
676                    sizeof(mac_data.entries));
677     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
678     if (s != sizeof(mac_data.entries)) {
679         goto error;
680     }
681 
682     iov_discard_front(&iov, &iov_cnt, s);
683 
684     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
685         goto error;
686     }
687 
688     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
689         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
690                        mac_data.entries * ETH_ALEN);
691         if (s != mac_data.entries * ETH_ALEN) {
692             goto error;
693         }
694         in_use += mac_data.entries;
695     } else {
696         multi_overflow = 1;
697     }
698 
699     n->mac_table.in_use = in_use;
700     n->mac_table.first_multi = first_multi;
701     n->mac_table.uni_overflow = uni_overflow;
702     n->mac_table.multi_overflow = multi_overflow;
703     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
704     g_free(macs);
705     rxfilter_notify(nc);
706 
707     return VIRTIO_NET_OK;
708 
709 error:
710     g_free(macs);
711     return VIRTIO_NET_ERR;
712 }
713 
714 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
715                                         struct iovec *iov, unsigned int iov_cnt)
716 {
717     VirtIODevice *vdev = VIRTIO_DEVICE(n);
718     uint16_t vid;
719     size_t s;
720     NetClientState *nc = qemu_get_queue(n->nic);
721 
722     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
723     vid = virtio_lduw_p(vdev, &vid);
724     if (s != sizeof(vid)) {
725         return VIRTIO_NET_ERR;
726     }
727 
728     if (vid >= MAX_VLAN)
729         return VIRTIO_NET_ERR;
730 
731     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
732         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
733     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
734         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
735     else
736         return VIRTIO_NET_ERR;
737 
738     rxfilter_notify(nc);
739 
740     return VIRTIO_NET_OK;
741 }
742 
743 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
744                                       struct iovec *iov, unsigned int iov_cnt)
745 {
746     if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
747         n->status & VIRTIO_NET_S_ANNOUNCE) {
748         n->status &= ~VIRTIO_NET_S_ANNOUNCE;
749         if (n->announce_counter) {
750             timer_mod(n->announce_timer,
751                       qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
752                       self_announce_delay(n->announce_counter));
753         }
754         return VIRTIO_NET_OK;
755     } else {
756         return VIRTIO_NET_ERR;
757     }
758 }
759 
760 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
761                                 struct iovec *iov, unsigned int iov_cnt)
762 {
763     VirtIODevice *vdev = VIRTIO_DEVICE(n);
764     struct virtio_net_ctrl_mq mq;
765     size_t s;
766     uint16_t queues;
767 
768     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
769     if (s != sizeof(mq)) {
770         return VIRTIO_NET_ERR;
771     }
772 
773     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
774         return VIRTIO_NET_ERR;
775     }
776 
777     queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
778 
779     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
780         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
781         queues > n->max_queues ||
782         !n->multiqueue) {
783         return VIRTIO_NET_ERR;
784     }
785 
786     n->curr_queues = queues;
787     /* stop the backend before changing the number of queues to avoid handling a
788      * disabled queue */
789     virtio_net_set_status(vdev, vdev->status);
790     virtio_net_set_queues(n);
791 
792     return VIRTIO_NET_OK;
793 }
794 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
795 {
796     VirtIONet *n = VIRTIO_NET(vdev);
797     struct virtio_net_ctrl_hdr ctrl;
798     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
799     VirtQueueElement elem;
800     size_t s;
801     struct iovec *iov, *iov2;
802     unsigned int iov_cnt;
803 
804     while (virtqueue_pop(vq, &elem)) {
805         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
806             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
807             error_report("virtio-net ctrl missing headers");
808             exit(1);
809         }
810 
811         iov_cnt = elem.out_num;
812         iov2 = iov = g_memdup(elem.out_sg, sizeof(struct iovec) * elem.out_num);
813         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
814         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
815         if (s != sizeof(ctrl)) {
816             status = VIRTIO_NET_ERR;
817         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
818             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
819         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
820             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
821         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
822             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
823         } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
824             status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
825         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
826             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
827         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
828             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
829         }
830 
831         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
832         assert(s == sizeof(status));
833 
834         virtqueue_push(vq, &elem, sizeof(status));
835         virtio_notify(vdev, vq);
836         g_free(iov2);
837     }
838 }
839 
840 /* RX */
841 
842 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
843 {
844     VirtIONet *n = VIRTIO_NET(vdev);
845     int queue_index = vq2q(virtio_get_queue_index(vq));
846 
847     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
848 }
849 
850 static int virtio_net_can_receive(NetClientState *nc)
851 {
852     VirtIONet *n = qemu_get_nic_opaque(nc);
853     VirtIODevice *vdev = VIRTIO_DEVICE(n);
854     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
855 
856     if (!vdev->vm_running) {
857         return 0;
858     }
859 
860     if (nc->queue_index >= n->curr_queues) {
861         return 0;
862     }
863 
864     if (!virtio_queue_ready(q->rx_vq) ||
865         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
866         return 0;
867     }
868 
869     return 1;
870 }
871 
872 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
873 {
874     VirtIONet *n = q->n;
875     if (virtio_queue_empty(q->rx_vq) ||
876         (n->mergeable_rx_bufs &&
877          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
878         virtio_queue_set_notification(q->rx_vq, 1);
879 
880         /* To avoid a race condition where the guest has made some buffers
881          * available after the above check but before notification was
882          * enabled, check for available buffers again.
883          */
884         if (virtio_queue_empty(q->rx_vq) ||
885             (n->mergeable_rx_bufs &&
886              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
887             return 0;
888         }
889     }
890 
891     virtio_queue_set_notification(q->rx_vq, 0);
892     return 1;
893 }
894 
895 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
896 {
897     virtio_tswap16s(vdev, &hdr->hdr_len);
898     virtio_tswap16s(vdev, &hdr->gso_size);
899     virtio_tswap16s(vdev, &hdr->csum_start);
900     virtio_tswap16s(vdev, &hdr->csum_offset);
901 }
902 
903 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
904  * it never finds out that the packets don't have valid checksums.  This
905  * causes dhclient to get upset.  Fedora's carried a patch for ages to
906  * fix this with Xen but it hasn't appeared in an upstream release of
907  * dhclient yet.
908  *
909  * To avoid breaking existing guests, we catch udp packets and add
910  * checksums.  This is terrible but it's better than hacking the guest
911  * kernels.
912  *
913  * N.B. if we introduce a zero-copy API, this operation is no longer free so
914  * we should provide a mechanism to disable it to avoid polluting the host
915  * cache.
916  */
917 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
918                                         uint8_t *buf, size_t size)
919 {
920     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
921         (size > 27 && size < 1500) && /* normal sized MTU */
922         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
923         (buf[23] == 17) && /* ip.protocol == UDP */
924         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
925         net_checksum_calculate(buf, size);
926         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
927     }
928 }
929 
930 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
931                            const void *buf, size_t size)
932 {
933     if (n->has_vnet_hdr) {
934         /* FIXME this cast is evil */
935         void *wbuf = (void *)buf;
936         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
937                                     size - n->host_hdr_len);
938         virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
939         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
940     } else {
941         struct virtio_net_hdr hdr = {
942             .flags = 0,
943             .gso_type = VIRTIO_NET_HDR_GSO_NONE
944         };
945         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
946     }
947 }
948 
949 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
950 {
951     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
952     static const uint8_t vlan[] = {0x81, 0x00};
953     uint8_t *ptr = (uint8_t *)buf;
954     int i;
955 
956     if (n->promisc)
957         return 1;
958 
959     ptr += n->host_hdr_len;
960 
961     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
962         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
963         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
964             return 0;
965     }
966 
967     if (ptr[0] & 1) { // multicast
968         if (!memcmp(ptr, bcast, sizeof(bcast))) {
969             return !n->nobcast;
970         } else if (n->nomulti) {
971             return 0;
972         } else if (n->allmulti || n->mac_table.multi_overflow) {
973             return 1;
974         }
975 
976         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
977             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
978                 return 1;
979             }
980         }
981     } else { // unicast
982         if (n->nouni) {
983             return 0;
984         } else if (n->alluni || n->mac_table.uni_overflow) {
985             return 1;
986         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
987             return 1;
988         }
989 
990         for (i = 0; i < n->mac_table.first_multi; i++) {
991             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
992                 return 1;
993             }
994         }
995     }
996 
997     return 0;
998 }
999 
1000 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
1001 {
1002     VirtIONet *n = qemu_get_nic_opaque(nc);
1003     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1004     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1005     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1006     struct virtio_net_hdr_mrg_rxbuf mhdr;
1007     unsigned mhdr_cnt = 0;
1008     size_t offset, i, guest_offset;
1009 
1010     if (!virtio_net_can_receive(nc)) {
1011         return -1;
1012     }
1013 
1014     /* hdr_len refers to the header we supply to the guest */
1015     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1016         return 0;
1017     }
1018 
1019     if (!receive_filter(n, buf, size))
1020         return size;
1021 
1022     offset = i = 0;
1023 
1024     while (offset < size) {
1025         VirtQueueElement elem;
1026         int len, total;
1027         const struct iovec *sg = elem.in_sg;
1028 
1029         total = 0;
1030 
1031         if (virtqueue_pop(q->rx_vq, &elem) == 0) {
1032             if (i == 0)
1033                 return -1;
1034             error_report("virtio-net unexpected empty queue: "
1035                     "i %zd mergeable %d offset %zd, size %zd, "
1036                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
1037                     i, n->mergeable_rx_bufs, offset, size,
1038                     n->guest_hdr_len, n->host_hdr_len, vdev->guest_features);
1039             exit(1);
1040         }
1041 
1042         if (elem.in_num < 1) {
1043             error_report("virtio-net receive queue contains no in buffers");
1044             exit(1);
1045         }
1046 
1047         if (i == 0) {
1048             assert(offset == 0);
1049             if (n->mergeable_rx_bufs) {
1050                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1051                                     sg, elem.in_num,
1052                                     offsetof(typeof(mhdr), num_buffers),
1053                                     sizeof(mhdr.num_buffers));
1054             }
1055 
1056             receive_header(n, sg, elem.in_num, buf, size);
1057             offset = n->host_hdr_len;
1058             total += n->guest_hdr_len;
1059             guest_offset = n->guest_hdr_len;
1060         } else {
1061             guest_offset = 0;
1062         }
1063 
1064         /* copy in packet.  ugh */
1065         len = iov_from_buf(sg, elem.in_num, guest_offset,
1066                            buf + offset, size - offset);
1067         total += len;
1068         offset += len;
1069         /* If buffers can't be merged, at this point we
1070          * must have consumed the complete packet.
1071          * Otherwise, drop it. */
1072         if (!n->mergeable_rx_bufs && offset < size) {
1073 #if 0
1074             error_report("virtio-net truncated non-mergeable packet: "
1075                          "i %zd mergeable %d offset %zd, size %zd, "
1076                          "guest hdr len %zd, host hdr len %zd",
1077                          i, n->mergeable_rx_bufs,
1078                          offset, size, n->guest_hdr_len, n->host_hdr_len);
1079 #endif
1080             return size;
1081         }
1082 
1083         /* signal other side */
1084         virtqueue_fill(q->rx_vq, &elem, total, i++);
1085     }
1086 
1087     if (mhdr_cnt) {
1088         virtio_stw_p(vdev, &mhdr.num_buffers, i);
1089         iov_from_buf(mhdr_sg, mhdr_cnt,
1090                      0,
1091                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1092     }
1093 
1094     virtqueue_flush(q->rx_vq, i);
1095     virtio_notify(vdev, q->rx_vq);
1096 
1097     return size;
1098 }
1099 
1100 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1101 
1102 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1103 {
1104     VirtIONet *n = qemu_get_nic_opaque(nc);
1105     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1106     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1107 
1108     virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
1109     virtio_notify(vdev, q->tx_vq);
1110 
1111     q->async_tx.elem.out_num = q->async_tx.len = 0;
1112 
1113     virtio_queue_set_notification(q->tx_vq, 1);
1114     virtio_net_flush_tx(q);
1115 }
1116 
1117 /* TX */
1118 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
1119 {
1120     VirtIONet *n = q->n;
1121     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1122     VirtQueueElement elem;
1123     int32_t num_packets = 0;
1124     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
1125     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1126         return num_packets;
1127     }
1128 
1129     if (q->async_tx.elem.out_num) {
1130         virtio_queue_set_notification(q->tx_vq, 0);
1131         return num_packets;
1132     }
1133 
1134     while (virtqueue_pop(q->tx_vq, &elem)) {
1135         ssize_t ret, len;
1136         unsigned int out_num = elem.out_num;
1137         struct iovec *out_sg = &elem.out_sg[0];
1138         struct iovec sg[VIRTQUEUE_MAX_SIZE];
1139 
1140         if (out_num < 1) {
1141             error_report("virtio-net header not in first element");
1142             exit(1);
1143         }
1144 
1145         if (n->has_vnet_hdr) {
1146             if (out_sg[0].iov_len < n->guest_hdr_len) {
1147                 error_report("virtio-net header incorrect");
1148                 exit(1);
1149             }
1150             virtio_net_hdr_swap(vdev, (void *) out_sg[0].iov_base);
1151         }
1152 
1153         /*
1154          * If host wants to see the guest header as is, we can
1155          * pass it on unchanged. Otherwise, copy just the parts
1156          * that host is interested in.
1157          */
1158         assert(n->host_hdr_len <= n->guest_hdr_len);
1159         if (n->host_hdr_len != n->guest_hdr_len) {
1160             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
1161                                        out_sg, out_num,
1162                                        0, n->host_hdr_len);
1163             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
1164                              out_sg, out_num,
1165                              n->guest_hdr_len, -1);
1166             out_num = sg_num;
1167             out_sg = sg;
1168         }
1169 
1170         len = n->guest_hdr_len;
1171 
1172         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
1173                                       out_sg, out_num, virtio_net_tx_complete);
1174         if (ret == 0) {
1175             virtio_queue_set_notification(q->tx_vq, 0);
1176             q->async_tx.elem = elem;
1177             q->async_tx.len  = len;
1178             return -EBUSY;
1179         }
1180 
1181         len += ret;
1182 
1183         virtqueue_push(q->tx_vq, &elem, 0);
1184         virtio_notify(vdev, q->tx_vq);
1185 
1186         if (++num_packets >= n->tx_burst) {
1187             break;
1188         }
1189     }
1190     return num_packets;
1191 }
1192 
1193 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
1194 {
1195     VirtIONet *n = VIRTIO_NET(vdev);
1196     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1197 
1198     /* This happens when device was stopped but VCPU wasn't. */
1199     if (!vdev->vm_running) {
1200         q->tx_waiting = 1;
1201         return;
1202     }
1203 
1204     if (q->tx_waiting) {
1205         virtio_queue_set_notification(vq, 1);
1206         timer_del(q->tx_timer);
1207         q->tx_waiting = 0;
1208         virtio_net_flush_tx(q);
1209     } else {
1210         timer_mod(q->tx_timer,
1211                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
1212         q->tx_waiting = 1;
1213         virtio_queue_set_notification(vq, 0);
1214     }
1215 }
1216 
1217 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1218 {
1219     VirtIONet *n = VIRTIO_NET(vdev);
1220     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1221 
1222     if (unlikely(q->tx_waiting)) {
1223         return;
1224     }
1225     q->tx_waiting = 1;
1226     /* This happens when device was stopped but VCPU wasn't. */
1227     if (!vdev->vm_running) {
1228         return;
1229     }
1230     virtio_queue_set_notification(vq, 0);
1231     qemu_bh_schedule(q->tx_bh);
1232 }
1233 
1234 static void virtio_net_tx_timer(void *opaque)
1235 {
1236     VirtIONetQueue *q = opaque;
1237     VirtIONet *n = q->n;
1238     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1239     /* This happens when device was stopped but BH wasn't. */
1240     if (!vdev->vm_running) {
1241         /* Make sure tx waiting is set, so we'll run when restarted. */
1242         assert(q->tx_waiting);
1243         return;
1244     }
1245 
1246     q->tx_waiting = 0;
1247 
1248     /* Just in case the driver is not ready on more */
1249     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1250         return;
1251     }
1252 
1253     virtio_queue_set_notification(q->tx_vq, 1);
1254     virtio_net_flush_tx(q);
1255 }
1256 
1257 static void virtio_net_tx_bh(void *opaque)
1258 {
1259     VirtIONetQueue *q = opaque;
1260     VirtIONet *n = q->n;
1261     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1262     int32_t ret;
1263 
1264     /* This happens when device was stopped but BH wasn't. */
1265     if (!vdev->vm_running) {
1266         /* Make sure tx waiting is set, so we'll run when restarted. */
1267         assert(q->tx_waiting);
1268         return;
1269     }
1270 
1271     q->tx_waiting = 0;
1272 
1273     /* Just in case the driver is not ready on more */
1274     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1275         return;
1276     }
1277 
1278     ret = virtio_net_flush_tx(q);
1279     if (ret == -EBUSY) {
1280         return; /* Notification re-enable handled by tx_complete */
1281     }
1282 
1283     /* If we flush a full burst of packets, assume there are
1284      * more coming and immediately reschedule */
1285     if (ret >= n->tx_burst) {
1286         qemu_bh_schedule(q->tx_bh);
1287         q->tx_waiting = 1;
1288         return;
1289     }
1290 
1291     /* If less than a full burst, re-enable notification and flush
1292      * anything that may have come in while we weren't looking.  If
1293      * we find something, assume the guest is still active and reschedule */
1294     virtio_queue_set_notification(q->tx_vq, 1);
1295     if (virtio_net_flush_tx(q) > 0) {
1296         virtio_queue_set_notification(q->tx_vq, 0);
1297         qemu_bh_schedule(q->tx_bh);
1298         q->tx_waiting = 1;
1299     }
1300 }
1301 
1302 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1303 {
1304     n->multiqueue = multiqueue;
1305 
1306     virtio_net_set_queues(n);
1307 }
1308 
1309 static void virtio_net_save(QEMUFile *f, void *opaque)
1310 {
1311     VirtIONet *n = opaque;
1312     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1313 
1314     /* At this point, backend must be stopped, otherwise
1315      * it might keep writing to memory. */
1316     assert(!n->vhost_started);
1317     virtio_save(vdev, f);
1318 }
1319 
1320 static void virtio_net_save_device(VirtIODevice *vdev, QEMUFile *f)
1321 {
1322     VirtIONet *n = VIRTIO_NET(vdev);
1323     int i;
1324 
1325     qemu_put_buffer(f, n->mac, ETH_ALEN);
1326     qemu_put_be32(f, n->vqs[0].tx_waiting);
1327     qemu_put_be32(f, n->mergeable_rx_bufs);
1328     qemu_put_be16(f, n->status);
1329     qemu_put_byte(f, n->promisc);
1330     qemu_put_byte(f, n->allmulti);
1331     qemu_put_be32(f, n->mac_table.in_use);
1332     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1333     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1334     qemu_put_be32(f, n->has_vnet_hdr);
1335     qemu_put_byte(f, n->mac_table.multi_overflow);
1336     qemu_put_byte(f, n->mac_table.uni_overflow);
1337     qemu_put_byte(f, n->alluni);
1338     qemu_put_byte(f, n->nomulti);
1339     qemu_put_byte(f, n->nouni);
1340     qemu_put_byte(f, n->nobcast);
1341     qemu_put_byte(f, n->has_ufo);
1342     if (n->max_queues > 1) {
1343         qemu_put_be16(f, n->max_queues);
1344         qemu_put_be16(f, n->curr_queues);
1345         for (i = 1; i < n->curr_queues; i++) {
1346             qemu_put_be32(f, n->vqs[i].tx_waiting);
1347         }
1348     }
1349 
1350     if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1351         qemu_put_be64(f, n->curr_guest_offloads);
1352     }
1353 }
1354 
1355 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1356 {
1357     VirtIONet *n = opaque;
1358     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1359 
1360     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1361         return -EINVAL;
1362 
1363     return virtio_load(vdev, f, version_id);
1364 }
1365 
1366 static int virtio_net_load_device(VirtIODevice *vdev, QEMUFile *f,
1367                                   int version_id)
1368 {
1369     VirtIONet *n = VIRTIO_NET(vdev);
1370     int i, link_down;
1371 
1372     qemu_get_buffer(f, n->mac, ETH_ALEN);
1373     n->vqs[0].tx_waiting = qemu_get_be32(f);
1374 
1375     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1376 
1377     if (version_id >= 3)
1378         n->status = qemu_get_be16(f);
1379 
1380     if (version_id >= 4) {
1381         if (version_id < 8) {
1382             n->promisc = qemu_get_be32(f);
1383             n->allmulti = qemu_get_be32(f);
1384         } else {
1385             n->promisc = qemu_get_byte(f);
1386             n->allmulti = qemu_get_byte(f);
1387         }
1388     }
1389 
1390     if (version_id >= 5) {
1391         n->mac_table.in_use = qemu_get_be32(f);
1392         /* MAC_TABLE_ENTRIES may be different from the saved image */
1393         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1394             qemu_get_buffer(f, n->mac_table.macs,
1395                             n->mac_table.in_use * ETH_ALEN);
1396         } else {
1397             int64_t i;
1398 
1399             /* Overflow detected - can happen if source has a larger MAC table.
1400              * We simply set overflow flag so there's no need to maintain the
1401              * table of addresses, discard them all.
1402              * Note: 64 bit math to avoid integer overflow.
1403              */
1404             for (i = 0; i < (int64_t)n->mac_table.in_use * ETH_ALEN; ++i) {
1405                 qemu_get_byte(f);
1406             }
1407             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1408             n->mac_table.in_use = 0;
1409         }
1410     }
1411 
1412     if (version_id >= 6)
1413         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1414 
1415     if (version_id >= 7) {
1416         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1417             error_report("virtio-net: saved image requires vnet_hdr=on");
1418             return -1;
1419         }
1420     }
1421 
1422     if (version_id >= 9) {
1423         n->mac_table.multi_overflow = qemu_get_byte(f);
1424         n->mac_table.uni_overflow = qemu_get_byte(f);
1425     }
1426 
1427     if (version_id >= 10) {
1428         n->alluni = qemu_get_byte(f);
1429         n->nomulti = qemu_get_byte(f);
1430         n->nouni = qemu_get_byte(f);
1431         n->nobcast = qemu_get_byte(f);
1432     }
1433 
1434     if (version_id >= 11) {
1435         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1436             error_report("virtio-net: saved image requires TUN_F_UFO support");
1437             return -1;
1438         }
1439     }
1440 
1441     if (n->max_queues > 1) {
1442         if (n->max_queues != qemu_get_be16(f)) {
1443             error_report("virtio-net: different max_queues ");
1444             return -1;
1445         }
1446 
1447         n->curr_queues = qemu_get_be16(f);
1448         if (n->curr_queues > n->max_queues) {
1449             error_report("virtio-net: curr_queues %x > max_queues %x",
1450                          n->curr_queues, n->max_queues);
1451             return -1;
1452         }
1453         for (i = 1; i < n->curr_queues; i++) {
1454             n->vqs[i].tx_waiting = qemu_get_be32(f);
1455         }
1456     }
1457 
1458     if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1459         n->curr_guest_offloads = qemu_get_be64(f);
1460     } else {
1461         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
1462     }
1463 
1464     if (peer_has_vnet_hdr(n)) {
1465         virtio_net_apply_guest_offloads(n);
1466     }
1467 
1468     virtio_net_set_queues(n);
1469 
1470     /* Find the first multicast entry in the saved MAC filter */
1471     for (i = 0; i < n->mac_table.in_use; i++) {
1472         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1473             break;
1474         }
1475     }
1476     n->mac_table.first_multi = i;
1477 
1478     /* nc.link_down can't be migrated, so infer link_down according
1479      * to link status bit in n->status */
1480     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1481     for (i = 0; i < n->max_queues; i++) {
1482         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1483     }
1484 
1485     if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
1486         virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
1487         n->announce_counter = SELF_ANNOUNCE_ROUNDS;
1488         timer_mod(n->announce_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL));
1489     }
1490 
1491     return 0;
1492 }
1493 
1494 static NetClientInfo net_virtio_info = {
1495     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1496     .size = sizeof(NICState),
1497     .can_receive = virtio_net_can_receive,
1498     .receive = virtio_net_receive,
1499     .link_status_changed = virtio_net_set_link_status,
1500     .query_rx_filter = virtio_net_query_rxfilter,
1501 };
1502 
1503 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1504 {
1505     VirtIONet *n = VIRTIO_NET(vdev);
1506     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1507     assert(n->vhost_started);
1508     return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
1509 }
1510 
1511 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1512                                            bool mask)
1513 {
1514     VirtIONet *n = VIRTIO_NET(vdev);
1515     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1516     assert(n->vhost_started);
1517     vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
1518                              vdev, idx, mask);
1519 }
1520 
1521 static void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features)
1522 {
1523     int i, config_size = 0;
1524     virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
1525     for (i = 0; feature_sizes[i].flags != 0; i++) {
1526         if (host_features & feature_sizes[i].flags) {
1527             config_size = MAX(feature_sizes[i].end, config_size);
1528         }
1529     }
1530     n->config_size = config_size;
1531 }
1532 
1533 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
1534                                    const char *type)
1535 {
1536     /*
1537      * The name can be NULL, the netclient name will be type.x.
1538      */
1539     assert(type != NULL);
1540 
1541     g_free(n->netclient_name);
1542     g_free(n->netclient_type);
1543     n->netclient_name = g_strdup(name);
1544     n->netclient_type = g_strdup(type);
1545 }
1546 
1547 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
1548 {
1549     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1550     VirtIONet *n = VIRTIO_NET(dev);
1551     NetClientState *nc;
1552     int i;
1553 
1554     virtio_net_set_config_size(n, n->host_features);
1555     virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
1556 
1557     n->max_queues = MAX(n->nic_conf.peers.queues, 1);
1558     if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) {
1559         error_setg(errp, "Invalid number of queues (= %" PRIu32 "), "
1560                    "must be a positive integer less than %d.",
1561                    n->max_queues, (VIRTIO_QUEUE_MAX - 1) / 2);
1562         virtio_cleanup(vdev);
1563         return;
1564     }
1565     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1566     n->curr_queues = 1;
1567     n->tx_timeout = n->net_conf.txtimer;
1568 
1569     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1570                        && strcmp(n->net_conf.tx, "bh")) {
1571         error_report("virtio-net: "
1572                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1573                      n->net_conf.tx);
1574         error_report("Defaulting to \"bh\"");
1575     }
1576 
1577     for (i = 0; i < n->max_queues; i++) {
1578         n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1579         if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1580             n->vqs[i].tx_vq =
1581                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1582             n->vqs[i].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
1583                                               virtio_net_tx_timer,
1584                                               &n->vqs[i]);
1585         } else {
1586             n->vqs[i].tx_vq =
1587                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1588             n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1589         }
1590 
1591         n->vqs[i].tx_waiting = 0;
1592         n->vqs[i].n = n;
1593     }
1594 
1595     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1596     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1597     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1598     n->status = VIRTIO_NET_S_LINK_UP;
1599     n->announce_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
1600                                      virtio_net_announce_timer, n);
1601 
1602     if (n->netclient_type) {
1603         /*
1604          * Happen when virtio_net_set_netclient_name has been called.
1605          */
1606         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1607                               n->netclient_type, n->netclient_name, n);
1608     } else {
1609         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1610                               object_get_typename(OBJECT(dev)), dev->id, n);
1611     }
1612 
1613     peer_test_vnet_hdr(n);
1614     if (peer_has_vnet_hdr(n)) {
1615         for (i = 0; i < n->max_queues; i++) {
1616             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1617         }
1618         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1619     } else {
1620         n->host_hdr_len = 0;
1621     }
1622 
1623     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1624 
1625     n->vqs[0].tx_waiting = 0;
1626     n->tx_burst = n->net_conf.txburst;
1627     virtio_net_set_mrg_rx_bufs(n, 0);
1628     n->promisc = 1; /* for compatibility */
1629 
1630     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1631 
1632     n->vlans = g_malloc0(MAX_VLAN >> 3);
1633 
1634     nc = qemu_get_queue(n->nic);
1635     nc->rxfilter_notify_enabled = 1;
1636 
1637     n->qdev = dev;
1638     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1639                     virtio_net_save, virtio_net_load, n);
1640 }
1641 
1642 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
1643 {
1644     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1645     VirtIONet *n = VIRTIO_NET(dev);
1646     int i;
1647 
1648     /* This will stop vhost backend if appropriate. */
1649     virtio_net_set_status(vdev, 0);
1650 
1651     unregister_savevm(dev, "virtio-net", n);
1652 
1653     g_free(n->netclient_name);
1654     n->netclient_name = NULL;
1655     g_free(n->netclient_type);
1656     n->netclient_type = NULL;
1657 
1658     g_free(n->mac_table.macs);
1659     g_free(n->vlans);
1660 
1661     for (i = 0; i < n->max_queues; i++) {
1662         VirtIONetQueue *q = &n->vqs[i];
1663         NetClientState *nc = qemu_get_subqueue(n->nic, i);
1664 
1665         qemu_purge_queued_packets(nc);
1666 
1667         if (q->tx_timer) {
1668             timer_del(q->tx_timer);
1669             timer_free(q->tx_timer);
1670         } else if (q->tx_bh) {
1671             qemu_bh_delete(q->tx_bh);
1672         }
1673     }
1674 
1675     timer_del(n->announce_timer);
1676     timer_free(n->announce_timer);
1677     g_free(n->vqs);
1678     qemu_del_nic(n->nic);
1679     virtio_cleanup(vdev);
1680 }
1681 
1682 static void virtio_net_instance_init(Object *obj)
1683 {
1684     VirtIONet *n = VIRTIO_NET(obj);
1685 
1686     /*
1687      * The default config_size is sizeof(struct virtio_net_config).
1688      * Can be overriden with virtio_net_set_config_size.
1689      */
1690     n->config_size = sizeof(struct virtio_net_config);
1691     device_add_bootindex_property(obj, &n->nic_conf.bootindex,
1692                                   "bootindex", "/ethernet-phy@0",
1693                                   DEVICE(n), NULL);
1694 }
1695 
1696 static Property virtio_net_properties[] = {
1697     DEFINE_VIRTIO_NET_FEATURES(VirtIONet, host_features),
1698     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
1699     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
1700                                                TX_TIMER_INTERVAL),
1701     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
1702     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
1703     DEFINE_PROP_END_OF_LIST(),
1704 };
1705 
1706 static void virtio_net_class_init(ObjectClass *klass, void *data)
1707 {
1708     DeviceClass *dc = DEVICE_CLASS(klass);
1709     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1710 
1711     dc->props = virtio_net_properties;
1712     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
1713     vdc->realize = virtio_net_device_realize;
1714     vdc->unrealize = virtio_net_device_unrealize;
1715     vdc->get_config = virtio_net_get_config;
1716     vdc->set_config = virtio_net_set_config;
1717     vdc->get_features = virtio_net_get_features;
1718     vdc->set_features = virtio_net_set_features;
1719     vdc->bad_features = virtio_net_bad_features;
1720     vdc->reset = virtio_net_reset;
1721     vdc->set_status = virtio_net_set_status;
1722     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
1723     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
1724     vdc->load = virtio_net_load_device;
1725     vdc->save = virtio_net_save_device;
1726 }
1727 
1728 static const TypeInfo virtio_net_info = {
1729     .name = TYPE_VIRTIO_NET,
1730     .parent = TYPE_VIRTIO_DEVICE,
1731     .instance_size = sizeof(VirtIONet),
1732     .instance_init = virtio_net_instance_init,
1733     .class_init = virtio_net_class_init,
1734 };
1735 
1736 static void virtio_register_types(void)
1737 {
1738     type_register_static(&virtio_net_info);
1739 }
1740 
1741 type_init(virtio_register_types)
1742