xref: /openbmc/qemu/hw/net/virtio-net.c (revision d6032e06)
1 /*
2  * Virtio Network Device
3  *
4  * Copyright IBM, Corp. 2007
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/iov.h"
15 #include "hw/virtio/virtio.h"
16 #include "net/net.h"
17 #include "net/checksum.h"
18 #include "net/tap.h"
19 #include "qemu/error-report.h"
20 #include "qemu/timer.h"
21 #include "hw/virtio/virtio-net.h"
22 #include "net/vhost_net.h"
23 #include "hw/virtio/virtio-bus.h"
24 #include "qapi/qmp/qjson.h"
25 #include "monitor/monitor.h"
26 
27 #define VIRTIO_NET_VM_VERSION    11
28 
29 #define MAC_TABLE_ENTRIES    64
30 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
31 
32 /*
33  * Calculate the number of bytes up to and including the given 'field' of
34  * 'container'.
35  */
36 #define endof(container, field) \
37     (offsetof(container, field) + sizeof(((container *)0)->field))
38 
39 typedef struct VirtIOFeature {
40     uint32_t flags;
41     size_t end;
42 } VirtIOFeature;
43 
44 static VirtIOFeature feature_sizes[] = {
45     {.flags = 1 << VIRTIO_NET_F_MAC,
46      .end = endof(struct virtio_net_config, mac)},
47     {.flags = 1 << VIRTIO_NET_F_STATUS,
48      .end = endof(struct virtio_net_config, status)},
49     {.flags = 1 << VIRTIO_NET_F_MQ,
50      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
51     {}
52 };
53 
54 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
55 {
56     VirtIONet *n = qemu_get_nic_opaque(nc);
57 
58     return &n->vqs[nc->queue_index];
59 }
60 
61 static int vq2q(int queue_index)
62 {
63     return queue_index / 2;
64 }
65 
66 /* TODO
67  * - we could suppress RX interrupt if we were so inclined.
68  */
69 
70 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
71 {
72     VirtIONet *n = VIRTIO_NET(vdev);
73     struct virtio_net_config netcfg;
74 
75     stw_p(&netcfg.status, n->status);
76     stw_p(&netcfg.max_virtqueue_pairs, n->max_queues);
77     memcpy(netcfg.mac, n->mac, ETH_ALEN);
78     memcpy(config, &netcfg, n->config_size);
79 }
80 
81 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
82 {
83     VirtIONet *n = VIRTIO_NET(vdev);
84     struct virtio_net_config netcfg = {};
85 
86     memcpy(&netcfg, config, n->config_size);
87 
88     if (!(vdev->guest_features >> VIRTIO_NET_F_CTRL_MAC_ADDR & 1) &&
89         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
90         memcpy(n->mac, netcfg.mac, ETH_ALEN);
91         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
92     }
93 }
94 
95 static bool virtio_net_started(VirtIONet *n, uint8_t status)
96 {
97     VirtIODevice *vdev = VIRTIO_DEVICE(n);
98     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
99         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
100 }
101 
102 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
103 {
104     VirtIODevice *vdev = VIRTIO_DEVICE(n);
105     NetClientState *nc = qemu_get_queue(n->nic);
106     int queues = n->multiqueue ? n->max_queues : 1;
107 
108     if (!nc->peer) {
109         return;
110     }
111     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
112         return;
113     }
114 
115     if (!tap_get_vhost_net(nc->peer)) {
116         return;
117     }
118 
119     if (!!n->vhost_started ==
120         (virtio_net_started(n, status) && !nc->peer->link_down)) {
121         return;
122     }
123     if (!n->vhost_started) {
124         int r;
125         if (!vhost_net_query(tap_get_vhost_net(nc->peer), vdev)) {
126             return;
127         }
128         n->vhost_started = 1;
129         r = vhost_net_start(vdev, n->nic->ncs, queues);
130         if (r < 0) {
131             error_report("unable to start vhost net: %d: "
132                          "falling back on userspace virtio", -r);
133             n->vhost_started = 0;
134         }
135     } else {
136         vhost_net_stop(vdev, n->nic->ncs, queues);
137         n->vhost_started = 0;
138     }
139 }
140 
141 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
142 {
143     VirtIONet *n = VIRTIO_NET(vdev);
144     VirtIONetQueue *q;
145     int i;
146     uint8_t queue_status;
147 
148     virtio_net_vhost_status(n, status);
149 
150     for (i = 0; i < n->max_queues; i++) {
151         q = &n->vqs[i];
152 
153         if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
154             queue_status = 0;
155         } else {
156             queue_status = status;
157         }
158 
159         if (!q->tx_waiting) {
160             continue;
161         }
162 
163         if (virtio_net_started(n, queue_status) && !n->vhost_started) {
164             if (q->tx_timer) {
165                 timer_mod(q->tx_timer,
166                                qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
167             } else {
168                 qemu_bh_schedule(q->tx_bh);
169             }
170         } else {
171             if (q->tx_timer) {
172                 timer_del(q->tx_timer);
173             } else {
174                 qemu_bh_cancel(q->tx_bh);
175             }
176         }
177     }
178 }
179 
180 static void virtio_net_set_link_status(NetClientState *nc)
181 {
182     VirtIONet *n = qemu_get_nic_opaque(nc);
183     VirtIODevice *vdev = VIRTIO_DEVICE(n);
184     uint16_t old_status = n->status;
185 
186     if (nc->link_down)
187         n->status &= ~VIRTIO_NET_S_LINK_UP;
188     else
189         n->status |= VIRTIO_NET_S_LINK_UP;
190 
191     if (n->status != old_status)
192         virtio_notify_config(vdev);
193 
194     virtio_net_set_status(vdev, vdev->status);
195 }
196 
197 static void rxfilter_notify(NetClientState *nc)
198 {
199     QObject *event_data;
200     VirtIONet *n = qemu_get_nic_opaque(nc);
201 
202     if (nc->rxfilter_notify_enabled) {
203         gchar *path = object_get_canonical_path(OBJECT(n->qdev));
204         if (n->netclient_name) {
205             event_data = qobject_from_jsonf("{ 'name': %s, 'path': %s }",
206                                     n->netclient_name, path);
207         } else {
208             event_data = qobject_from_jsonf("{ 'path': %s }", path);
209         }
210         monitor_protocol_event(QEVENT_NIC_RX_FILTER_CHANGED, event_data);
211         qobject_decref(event_data);
212         g_free(path);
213 
214         /* disable event notification to avoid events flooding */
215         nc->rxfilter_notify_enabled = 0;
216     }
217 }
218 
219 static char *mac_strdup_printf(const uint8_t *mac)
220 {
221     return g_strdup_printf("%.2x:%.2x:%.2x:%.2x:%.2x:%.2x", mac[0],
222                             mac[1], mac[2], mac[3], mac[4], mac[5]);
223 }
224 
225 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
226 {
227     VirtIONet *n = qemu_get_nic_opaque(nc);
228     RxFilterInfo *info;
229     strList *str_list, *entry;
230     intList *int_list, *int_entry;
231     int i, j;
232 
233     info = g_malloc0(sizeof(*info));
234     info->name = g_strdup(nc->name);
235     info->promiscuous = n->promisc;
236 
237     if (n->nouni) {
238         info->unicast = RX_STATE_NONE;
239     } else if (n->alluni) {
240         info->unicast = RX_STATE_ALL;
241     } else {
242         info->unicast = RX_STATE_NORMAL;
243     }
244 
245     if (n->nomulti) {
246         info->multicast = RX_STATE_NONE;
247     } else if (n->allmulti) {
248         info->multicast = RX_STATE_ALL;
249     } else {
250         info->multicast = RX_STATE_NORMAL;
251     }
252 
253     info->broadcast_allowed = n->nobcast;
254     info->multicast_overflow = n->mac_table.multi_overflow;
255     info->unicast_overflow = n->mac_table.uni_overflow;
256 
257     info->main_mac = mac_strdup_printf(n->mac);
258 
259     str_list = NULL;
260     for (i = 0; i < n->mac_table.first_multi; i++) {
261         entry = g_malloc0(sizeof(*entry));
262         entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
263         entry->next = str_list;
264         str_list = entry;
265     }
266     info->unicast_table = str_list;
267 
268     str_list = NULL;
269     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
270         entry = g_malloc0(sizeof(*entry));
271         entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
272         entry->next = str_list;
273         str_list = entry;
274     }
275     info->multicast_table = str_list;
276 
277     int_list = NULL;
278     for (i = 0; i < MAX_VLAN >> 5; i++) {
279         for (j = 0; n->vlans[i] && j < 0x1f; j++) {
280             if (n->vlans[i] & (1U << j)) {
281                 int_entry = g_malloc0(sizeof(*int_entry));
282                 int_entry->value = (i << 5) + j;
283                 int_entry->next = int_list;
284                 int_list = int_entry;
285             }
286         }
287     }
288     info->vlan_table = int_list;
289 
290     /* enable event notification after query */
291     nc->rxfilter_notify_enabled = 1;
292 
293     return info;
294 }
295 
296 static void virtio_net_reset(VirtIODevice *vdev)
297 {
298     VirtIONet *n = VIRTIO_NET(vdev);
299 
300     /* Reset back to compatibility mode */
301     n->promisc = 1;
302     n->allmulti = 0;
303     n->alluni = 0;
304     n->nomulti = 0;
305     n->nouni = 0;
306     n->nobcast = 0;
307     /* multiqueue is disabled by default */
308     n->curr_queues = 1;
309 
310     /* Flush any MAC and VLAN filter table state */
311     n->mac_table.in_use = 0;
312     n->mac_table.first_multi = 0;
313     n->mac_table.multi_overflow = 0;
314     n->mac_table.uni_overflow = 0;
315     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
316     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
317     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
318     memset(n->vlans, 0, MAX_VLAN >> 3);
319 }
320 
321 static void peer_test_vnet_hdr(VirtIONet *n)
322 {
323     NetClientState *nc = qemu_get_queue(n->nic);
324     if (!nc->peer) {
325         return;
326     }
327 
328     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
329         return;
330     }
331 
332     n->has_vnet_hdr = tap_has_vnet_hdr(nc->peer);
333 }
334 
335 static int peer_has_vnet_hdr(VirtIONet *n)
336 {
337     return n->has_vnet_hdr;
338 }
339 
340 static int peer_has_ufo(VirtIONet *n)
341 {
342     if (!peer_has_vnet_hdr(n))
343         return 0;
344 
345     n->has_ufo = tap_has_ufo(qemu_get_queue(n->nic)->peer);
346 
347     return n->has_ufo;
348 }
349 
350 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs)
351 {
352     int i;
353     NetClientState *nc;
354 
355     n->mergeable_rx_bufs = mergeable_rx_bufs;
356 
357     n->guest_hdr_len = n->mergeable_rx_bufs ?
358         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
359 
360     for (i = 0; i < n->max_queues; i++) {
361         nc = qemu_get_subqueue(n->nic, i);
362 
363         if (peer_has_vnet_hdr(n) &&
364             tap_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
365             tap_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
366             n->host_hdr_len = n->guest_hdr_len;
367         }
368     }
369 }
370 
371 static int peer_attach(VirtIONet *n, int index)
372 {
373     NetClientState *nc = qemu_get_subqueue(n->nic, index);
374 
375     if (!nc->peer) {
376         return 0;
377     }
378 
379     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
380         return 0;
381     }
382 
383     return tap_enable(nc->peer);
384 }
385 
386 static int peer_detach(VirtIONet *n, int index)
387 {
388     NetClientState *nc = qemu_get_subqueue(n->nic, index);
389 
390     if (!nc->peer) {
391         return 0;
392     }
393 
394     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
395         return 0;
396     }
397 
398     return tap_disable(nc->peer);
399 }
400 
401 static void virtio_net_set_queues(VirtIONet *n)
402 {
403     int i;
404 
405     for (i = 0; i < n->max_queues; i++) {
406         if (i < n->curr_queues) {
407             assert(!peer_attach(n, i));
408         } else {
409             assert(!peer_detach(n, i));
410         }
411     }
412 }
413 
414 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
415 
416 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
417 {
418     VirtIONet *n = VIRTIO_NET(vdev);
419     NetClientState *nc = qemu_get_queue(n->nic);
420 
421     features |= (1 << VIRTIO_NET_F_MAC);
422 
423     if (!peer_has_vnet_hdr(n)) {
424         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
425         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
426         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
427         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
428 
429         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
430         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
431         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
432         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
433     }
434 
435     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
436         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
437         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
438     }
439 
440     if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
441         return features;
442     }
443     if (!tap_get_vhost_net(nc->peer)) {
444         return features;
445     }
446     return vhost_net_get_features(tap_get_vhost_net(nc->peer), features);
447 }
448 
449 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
450 {
451     uint32_t features = 0;
452 
453     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
454      * but also these: */
455     features |= (1 << VIRTIO_NET_F_MAC);
456     features |= (1 << VIRTIO_NET_F_CSUM);
457     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
458     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
459     features |= (1 << VIRTIO_NET_F_HOST_ECN);
460 
461     return features;
462 }
463 
464 static void virtio_net_apply_guest_offloads(VirtIONet *n)
465 {
466     tap_set_offload(qemu_get_subqueue(n->nic, 0)->peer,
467             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
468             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
469             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
470             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
471             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
472 }
473 
474 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
475 {
476     static const uint64_t guest_offloads_mask =
477         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
478         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
479         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
480         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
481         (1ULL << VIRTIO_NET_F_GUEST_UFO);
482 
483     return guest_offloads_mask & features;
484 }
485 
486 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
487 {
488     VirtIODevice *vdev = VIRTIO_DEVICE(n);
489     return virtio_net_guest_offloads_by_features(vdev->guest_features);
490 }
491 
492 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
493 {
494     VirtIONet *n = VIRTIO_NET(vdev);
495     int i;
496 
497     virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)));
498 
499     virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
500 
501     if (n->has_vnet_hdr) {
502         n->curr_guest_offloads =
503             virtio_net_guest_offloads_by_features(features);
504         virtio_net_apply_guest_offloads(n);
505     }
506 
507     for (i = 0;  i < n->max_queues; i++) {
508         NetClientState *nc = qemu_get_subqueue(n->nic, i);
509 
510         if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
511             continue;
512         }
513         if (!tap_get_vhost_net(nc->peer)) {
514             continue;
515         }
516         vhost_net_ack_features(tap_get_vhost_net(nc->peer), features);
517     }
518 }
519 
520 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
521                                      struct iovec *iov, unsigned int iov_cnt)
522 {
523     uint8_t on;
524     size_t s;
525     NetClientState *nc = qemu_get_queue(n->nic);
526 
527     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
528     if (s != sizeof(on)) {
529         return VIRTIO_NET_ERR;
530     }
531 
532     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
533         n->promisc = on;
534     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
535         n->allmulti = on;
536     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
537         n->alluni = on;
538     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
539         n->nomulti = on;
540     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
541         n->nouni = on;
542     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
543         n->nobcast = on;
544     } else {
545         return VIRTIO_NET_ERR;
546     }
547 
548     rxfilter_notify(nc);
549 
550     return VIRTIO_NET_OK;
551 }
552 
553 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
554                                      struct iovec *iov, unsigned int iov_cnt)
555 {
556     VirtIODevice *vdev = VIRTIO_DEVICE(n);
557     uint64_t offloads;
558     size_t s;
559 
560     if (!((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features)) {
561         return VIRTIO_NET_ERR;
562     }
563 
564     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
565     if (s != sizeof(offloads)) {
566         return VIRTIO_NET_ERR;
567     }
568 
569     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
570         uint64_t supported_offloads;
571 
572         if (!n->has_vnet_hdr) {
573             return VIRTIO_NET_ERR;
574         }
575 
576         supported_offloads = virtio_net_supported_guest_offloads(n);
577         if (offloads & ~supported_offloads) {
578             return VIRTIO_NET_ERR;
579         }
580 
581         n->curr_guest_offloads = offloads;
582         virtio_net_apply_guest_offloads(n);
583 
584         return VIRTIO_NET_OK;
585     } else {
586         return VIRTIO_NET_ERR;
587     }
588 }
589 
590 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
591                                  struct iovec *iov, unsigned int iov_cnt)
592 {
593     struct virtio_net_ctrl_mac mac_data;
594     size_t s;
595     NetClientState *nc = qemu_get_queue(n->nic);
596 
597     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
598         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
599             return VIRTIO_NET_ERR;
600         }
601         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
602         assert(s == sizeof(n->mac));
603         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
604         rxfilter_notify(nc);
605 
606         return VIRTIO_NET_OK;
607     }
608 
609     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
610         return VIRTIO_NET_ERR;
611     }
612 
613     int in_use = 0;
614     int first_multi = 0;
615     uint8_t uni_overflow = 0;
616     uint8_t multi_overflow = 0;
617     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
618 
619     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
620                    sizeof(mac_data.entries));
621     mac_data.entries = ldl_p(&mac_data.entries);
622     if (s != sizeof(mac_data.entries)) {
623         goto error;
624     }
625     iov_discard_front(&iov, &iov_cnt, s);
626 
627     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
628         goto error;
629     }
630 
631     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
632         s = iov_to_buf(iov, iov_cnt, 0, macs,
633                        mac_data.entries * ETH_ALEN);
634         if (s != mac_data.entries * ETH_ALEN) {
635             goto error;
636         }
637         in_use += mac_data.entries;
638     } else {
639         uni_overflow = 1;
640     }
641 
642     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
643 
644     first_multi = in_use;
645 
646     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
647                    sizeof(mac_data.entries));
648     mac_data.entries = ldl_p(&mac_data.entries);
649     if (s != sizeof(mac_data.entries)) {
650         goto error;
651     }
652 
653     iov_discard_front(&iov, &iov_cnt, s);
654 
655     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
656         goto error;
657     }
658 
659     if (in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
660         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
661                        mac_data.entries * ETH_ALEN);
662         if (s != mac_data.entries * ETH_ALEN) {
663             goto error;
664         }
665         in_use += mac_data.entries;
666     } else {
667         multi_overflow = 1;
668     }
669 
670     n->mac_table.in_use = in_use;
671     n->mac_table.first_multi = first_multi;
672     n->mac_table.uni_overflow = uni_overflow;
673     n->mac_table.multi_overflow = multi_overflow;
674     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
675     g_free(macs);
676     rxfilter_notify(nc);
677 
678     return VIRTIO_NET_OK;
679 
680 error:
681     g_free(macs);
682     return VIRTIO_NET_ERR;
683 }
684 
685 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
686                                         struct iovec *iov, unsigned int iov_cnt)
687 {
688     uint16_t vid;
689     size_t s;
690     NetClientState *nc = qemu_get_queue(n->nic);
691 
692     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
693     vid = lduw_p(&vid);
694     if (s != sizeof(vid)) {
695         return VIRTIO_NET_ERR;
696     }
697 
698     if (vid >= MAX_VLAN)
699         return VIRTIO_NET_ERR;
700 
701     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
702         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
703     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
704         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
705     else
706         return VIRTIO_NET_ERR;
707 
708     rxfilter_notify(nc);
709 
710     return VIRTIO_NET_OK;
711 }
712 
713 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
714                                 struct iovec *iov, unsigned int iov_cnt)
715 {
716     VirtIODevice *vdev = VIRTIO_DEVICE(n);
717     struct virtio_net_ctrl_mq mq;
718     size_t s;
719     uint16_t queues;
720 
721     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
722     if (s != sizeof(mq)) {
723         return VIRTIO_NET_ERR;
724     }
725 
726     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
727         return VIRTIO_NET_ERR;
728     }
729 
730     queues = lduw_p(&mq.virtqueue_pairs);
731 
732     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
733         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
734         queues > n->max_queues ||
735         !n->multiqueue) {
736         return VIRTIO_NET_ERR;
737     }
738 
739     n->curr_queues = queues;
740     /* stop the backend before changing the number of queues to avoid handling a
741      * disabled queue */
742     virtio_net_set_status(vdev, vdev->status);
743     virtio_net_set_queues(n);
744 
745     return VIRTIO_NET_OK;
746 }
747 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
748 {
749     VirtIONet *n = VIRTIO_NET(vdev);
750     struct virtio_net_ctrl_hdr ctrl;
751     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
752     VirtQueueElement elem;
753     size_t s;
754     struct iovec *iov;
755     unsigned int iov_cnt;
756 
757     while (virtqueue_pop(vq, &elem)) {
758         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
759             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
760             error_report("virtio-net ctrl missing headers");
761             exit(1);
762         }
763 
764         iov = elem.out_sg;
765         iov_cnt = elem.out_num;
766         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
767         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
768         if (s != sizeof(ctrl)) {
769             status = VIRTIO_NET_ERR;
770         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
771             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
772         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
773             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
774         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
775             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
776         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
777             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
778         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
779             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
780         }
781 
782         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
783         assert(s == sizeof(status));
784 
785         virtqueue_push(vq, &elem, sizeof(status));
786         virtio_notify(vdev, vq);
787     }
788 }
789 
790 /* RX */
791 
792 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
793 {
794     VirtIONet *n = VIRTIO_NET(vdev);
795     int queue_index = vq2q(virtio_get_queue_index(vq));
796 
797     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
798 }
799 
800 static int virtio_net_can_receive(NetClientState *nc)
801 {
802     VirtIONet *n = qemu_get_nic_opaque(nc);
803     VirtIODevice *vdev = VIRTIO_DEVICE(n);
804     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
805 
806     if (!vdev->vm_running) {
807         return 0;
808     }
809 
810     if (nc->queue_index >= n->curr_queues) {
811         return 0;
812     }
813 
814     if (!virtio_queue_ready(q->rx_vq) ||
815         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
816         return 0;
817     }
818 
819     return 1;
820 }
821 
822 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
823 {
824     VirtIONet *n = q->n;
825     if (virtio_queue_empty(q->rx_vq) ||
826         (n->mergeable_rx_bufs &&
827          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
828         virtio_queue_set_notification(q->rx_vq, 1);
829 
830         /* To avoid a race condition where the guest has made some buffers
831          * available after the above check but before notification was
832          * enabled, check for available buffers again.
833          */
834         if (virtio_queue_empty(q->rx_vq) ||
835             (n->mergeable_rx_bufs &&
836              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
837             return 0;
838         }
839     }
840 
841     virtio_queue_set_notification(q->rx_vq, 0);
842     return 1;
843 }
844 
845 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
846  * it never finds out that the packets don't have valid checksums.  This
847  * causes dhclient to get upset.  Fedora's carried a patch for ages to
848  * fix this with Xen but it hasn't appeared in an upstream release of
849  * dhclient yet.
850  *
851  * To avoid breaking existing guests, we catch udp packets and add
852  * checksums.  This is terrible but it's better than hacking the guest
853  * kernels.
854  *
855  * N.B. if we introduce a zero-copy API, this operation is no longer free so
856  * we should provide a mechanism to disable it to avoid polluting the host
857  * cache.
858  */
859 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
860                                         uint8_t *buf, size_t size)
861 {
862     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
863         (size > 27 && size < 1500) && /* normal sized MTU */
864         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
865         (buf[23] == 17) && /* ip.protocol == UDP */
866         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
867         net_checksum_calculate(buf, size);
868         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
869     }
870 }
871 
872 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
873                            const void *buf, size_t size)
874 {
875     if (n->has_vnet_hdr) {
876         /* FIXME this cast is evil */
877         void *wbuf = (void *)buf;
878         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
879                                     size - n->host_hdr_len);
880         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
881     } else {
882         struct virtio_net_hdr hdr = {
883             .flags = 0,
884             .gso_type = VIRTIO_NET_HDR_GSO_NONE
885         };
886         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
887     }
888 }
889 
890 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
891 {
892     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
893     static const uint8_t vlan[] = {0x81, 0x00};
894     uint8_t *ptr = (uint8_t *)buf;
895     int i;
896 
897     if (n->promisc)
898         return 1;
899 
900     ptr += n->host_hdr_len;
901 
902     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
903         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
904         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
905             return 0;
906     }
907 
908     if (ptr[0] & 1) { // multicast
909         if (!memcmp(ptr, bcast, sizeof(bcast))) {
910             return !n->nobcast;
911         } else if (n->nomulti) {
912             return 0;
913         } else if (n->allmulti || n->mac_table.multi_overflow) {
914             return 1;
915         }
916 
917         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
918             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
919                 return 1;
920             }
921         }
922     } else { // unicast
923         if (n->nouni) {
924             return 0;
925         } else if (n->alluni || n->mac_table.uni_overflow) {
926             return 1;
927         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
928             return 1;
929         }
930 
931         for (i = 0; i < n->mac_table.first_multi; i++) {
932             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
933                 return 1;
934             }
935         }
936     }
937 
938     return 0;
939 }
940 
941 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
942 {
943     VirtIONet *n = qemu_get_nic_opaque(nc);
944     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
945     VirtIODevice *vdev = VIRTIO_DEVICE(n);
946     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
947     struct virtio_net_hdr_mrg_rxbuf mhdr;
948     unsigned mhdr_cnt = 0;
949     size_t offset, i, guest_offset;
950 
951     if (!virtio_net_can_receive(nc)) {
952         return -1;
953     }
954 
955     /* hdr_len refers to the header we supply to the guest */
956     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
957         return 0;
958     }
959 
960     if (!receive_filter(n, buf, size))
961         return size;
962 
963     offset = i = 0;
964 
965     while (offset < size) {
966         VirtQueueElement elem;
967         int len, total;
968         const struct iovec *sg = elem.in_sg;
969 
970         total = 0;
971 
972         if (virtqueue_pop(q->rx_vq, &elem) == 0) {
973             if (i == 0)
974                 return -1;
975             error_report("virtio-net unexpected empty queue: "
976                     "i %zd mergeable %d offset %zd, size %zd, "
977                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
978                     i, n->mergeable_rx_bufs, offset, size,
979                     n->guest_hdr_len, n->host_hdr_len, vdev->guest_features);
980             exit(1);
981         }
982 
983         if (elem.in_num < 1) {
984             error_report("virtio-net receive queue contains no in buffers");
985             exit(1);
986         }
987 
988         if (i == 0) {
989             assert(offset == 0);
990             if (n->mergeable_rx_bufs) {
991                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
992                                     sg, elem.in_num,
993                                     offsetof(typeof(mhdr), num_buffers),
994                                     sizeof(mhdr.num_buffers));
995             }
996 
997             receive_header(n, sg, elem.in_num, buf, size);
998             offset = n->host_hdr_len;
999             total += n->guest_hdr_len;
1000             guest_offset = n->guest_hdr_len;
1001         } else {
1002             guest_offset = 0;
1003         }
1004 
1005         /* copy in packet.  ugh */
1006         len = iov_from_buf(sg, elem.in_num, guest_offset,
1007                            buf + offset, size - offset);
1008         total += len;
1009         offset += len;
1010         /* If buffers can't be merged, at this point we
1011          * must have consumed the complete packet.
1012          * Otherwise, drop it. */
1013         if (!n->mergeable_rx_bufs && offset < size) {
1014 #if 0
1015             error_report("virtio-net truncated non-mergeable packet: "
1016                          "i %zd mergeable %d offset %zd, size %zd, "
1017                          "guest hdr len %zd, host hdr len %zd",
1018                          i, n->mergeable_rx_bufs,
1019                          offset, size, n->guest_hdr_len, n->host_hdr_len);
1020 #endif
1021             return size;
1022         }
1023 
1024         /* signal other side */
1025         virtqueue_fill(q->rx_vq, &elem, total, i++);
1026     }
1027 
1028     if (mhdr_cnt) {
1029         stw_p(&mhdr.num_buffers, i);
1030         iov_from_buf(mhdr_sg, mhdr_cnt,
1031                      0,
1032                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1033     }
1034 
1035     virtqueue_flush(q->rx_vq, i);
1036     virtio_notify(vdev, q->rx_vq);
1037 
1038     return size;
1039 }
1040 
1041 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1042 
1043 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1044 {
1045     VirtIONet *n = qemu_get_nic_opaque(nc);
1046     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1047     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1048 
1049     virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
1050     virtio_notify(vdev, q->tx_vq);
1051 
1052     q->async_tx.elem.out_num = q->async_tx.len = 0;
1053 
1054     virtio_queue_set_notification(q->tx_vq, 1);
1055     virtio_net_flush_tx(q);
1056 }
1057 
1058 /* TX */
1059 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
1060 {
1061     VirtIONet *n = q->n;
1062     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1063     VirtQueueElement elem;
1064     int32_t num_packets = 0;
1065     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
1066     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1067         return num_packets;
1068     }
1069 
1070     assert(vdev->vm_running);
1071 
1072     if (q->async_tx.elem.out_num) {
1073         virtio_queue_set_notification(q->tx_vq, 0);
1074         return num_packets;
1075     }
1076 
1077     while (virtqueue_pop(q->tx_vq, &elem)) {
1078         ssize_t ret, len;
1079         unsigned int out_num = elem.out_num;
1080         struct iovec *out_sg = &elem.out_sg[0];
1081         struct iovec sg[VIRTQUEUE_MAX_SIZE];
1082 
1083         if (out_num < 1) {
1084             error_report("virtio-net header not in first element");
1085             exit(1);
1086         }
1087 
1088         /*
1089          * If host wants to see the guest header as is, we can
1090          * pass it on unchanged. Otherwise, copy just the parts
1091          * that host is interested in.
1092          */
1093         assert(n->host_hdr_len <= n->guest_hdr_len);
1094         if (n->host_hdr_len != n->guest_hdr_len) {
1095             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
1096                                        out_sg, out_num,
1097                                        0, n->host_hdr_len);
1098             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
1099                              out_sg, out_num,
1100                              n->guest_hdr_len, -1);
1101             out_num = sg_num;
1102             out_sg = sg;
1103         }
1104 
1105         len = n->guest_hdr_len;
1106 
1107         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
1108                                       out_sg, out_num, virtio_net_tx_complete);
1109         if (ret == 0) {
1110             virtio_queue_set_notification(q->tx_vq, 0);
1111             q->async_tx.elem = elem;
1112             q->async_tx.len  = len;
1113             return -EBUSY;
1114         }
1115 
1116         len += ret;
1117 
1118         virtqueue_push(q->tx_vq, &elem, 0);
1119         virtio_notify(vdev, q->tx_vq);
1120 
1121         if (++num_packets >= n->tx_burst) {
1122             break;
1123         }
1124     }
1125     return num_packets;
1126 }
1127 
1128 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
1129 {
1130     VirtIONet *n = VIRTIO_NET(vdev);
1131     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1132 
1133     /* This happens when device was stopped but VCPU wasn't. */
1134     if (!vdev->vm_running) {
1135         q->tx_waiting = 1;
1136         return;
1137     }
1138 
1139     if (q->tx_waiting) {
1140         virtio_queue_set_notification(vq, 1);
1141         timer_del(q->tx_timer);
1142         q->tx_waiting = 0;
1143         virtio_net_flush_tx(q);
1144     } else {
1145         timer_mod(q->tx_timer,
1146                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
1147         q->tx_waiting = 1;
1148         virtio_queue_set_notification(vq, 0);
1149     }
1150 }
1151 
1152 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1153 {
1154     VirtIONet *n = VIRTIO_NET(vdev);
1155     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1156 
1157     if (unlikely(q->tx_waiting)) {
1158         return;
1159     }
1160     q->tx_waiting = 1;
1161     /* This happens when device was stopped but VCPU wasn't. */
1162     if (!vdev->vm_running) {
1163         return;
1164     }
1165     virtio_queue_set_notification(vq, 0);
1166     qemu_bh_schedule(q->tx_bh);
1167 }
1168 
1169 static void virtio_net_tx_timer(void *opaque)
1170 {
1171     VirtIONetQueue *q = opaque;
1172     VirtIONet *n = q->n;
1173     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1174     assert(vdev->vm_running);
1175 
1176     q->tx_waiting = 0;
1177 
1178     /* Just in case the driver is not ready on more */
1179     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1180         return;
1181     }
1182 
1183     virtio_queue_set_notification(q->tx_vq, 1);
1184     virtio_net_flush_tx(q);
1185 }
1186 
1187 static void virtio_net_tx_bh(void *opaque)
1188 {
1189     VirtIONetQueue *q = opaque;
1190     VirtIONet *n = q->n;
1191     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1192     int32_t ret;
1193 
1194     assert(vdev->vm_running);
1195 
1196     q->tx_waiting = 0;
1197 
1198     /* Just in case the driver is not ready on more */
1199     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1200         return;
1201     }
1202 
1203     ret = virtio_net_flush_tx(q);
1204     if (ret == -EBUSY) {
1205         return; /* Notification re-enable handled by tx_complete */
1206     }
1207 
1208     /* If we flush a full burst of packets, assume there are
1209      * more coming and immediately reschedule */
1210     if (ret >= n->tx_burst) {
1211         qemu_bh_schedule(q->tx_bh);
1212         q->tx_waiting = 1;
1213         return;
1214     }
1215 
1216     /* If less than a full burst, re-enable notification and flush
1217      * anything that may have come in while we weren't looking.  If
1218      * we find something, assume the guest is still active and reschedule */
1219     virtio_queue_set_notification(q->tx_vq, 1);
1220     if (virtio_net_flush_tx(q) > 0) {
1221         virtio_queue_set_notification(q->tx_vq, 0);
1222         qemu_bh_schedule(q->tx_bh);
1223         q->tx_waiting = 1;
1224     }
1225 }
1226 
1227 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1228 {
1229     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1230     int i, max = multiqueue ? n->max_queues : 1;
1231 
1232     n->multiqueue = multiqueue;
1233 
1234     for (i = 2; i <= n->max_queues * 2 + 1; i++) {
1235         virtio_del_queue(vdev, i);
1236     }
1237 
1238     for (i = 1; i < max; i++) {
1239         n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1240         if (n->vqs[i].tx_timer) {
1241             n->vqs[i].tx_vq =
1242                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1243             n->vqs[i].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
1244                                                    virtio_net_tx_timer,
1245                                                    &n->vqs[i]);
1246         } else {
1247             n->vqs[i].tx_vq =
1248                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1249             n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1250         }
1251 
1252         n->vqs[i].tx_waiting = 0;
1253         n->vqs[i].n = n;
1254     }
1255 
1256     /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack
1257      * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid
1258      * breaking them.
1259      */
1260     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1261 
1262     virtio_net_set_queues(n);
1263 }
1264 
1265 static void virtio_net_save(QEMUFile *f, void *opaque)
1266 {
1267     int i;
1268     VirtIONet *n = opaque;
1269     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1270 
1271     /* At this point, backend must be stopped, otherwise
1272      * it might keep writing to memory. */
1273     assert(!n->vhost_started);
1274     virtio_save(vdev, f);
1275 
1276     qemu_put_buffer(f, n->mac, ETH_ALEN);
1277     qemu_put_be32(f, n->vqs[0].tx_waiting);
1278     qemu_put_be32(f, n->mergeable_rx_bufs);
1279     qemu_put_be16(f, n->status);
1280     qemu_put_byte(f, n->promisc);
1281     qemu_put_byte(f, n->allmulti);
1282     qemu_put_be32(f, n->mac_table.in_use);
1283     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1284     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1285     qemu_put_be32(f, n->has_vnet_hdr);
1286     qemu_put_byte(f, n->mac_table.multi_overflow);
1287     qemu_put_byte(f, n->mac_table.uni_overflow);
1288     qemu_put_byte(f, n->alluni);
1289     qemu_put_byte(f, n->nomulti);
1290     qemu_put_byte(f, n->nouni);
1291     qemu_put_byte(f, n->nobcast);
1292     qemu_put_byte(f, n->has_ufo);
1293     if (n->max_queues > 1) {
1294         qemu_put_be16(f, n->max_queues);
1295         qemu_put_be16(f, n->curr_queues);
1296         for (i = 1; i < n->curr_queues; i++) {
1297             qemu_put_be32(f, n->vqs[i].tx_waiting);
1298         }
1299     }
1300 
1301     if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) {
1302         qemu_put_be64(f, n->curr_guest_offloads);
1303     }
1304 }
1305 
1306 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1307 {
1308     VirtIONet *n = opaque;
1309     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1310     int ret, i, link_down;
1311 
1312     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1313         return -EINVAL;
1314 
1315     ret = virtio_load(vdev, f);
1316     if (ret) {
1317         return ret;
1318     }
1319 
1320     qemu_get_buffer(f, n->mac, ETH_ALEN);
1321     n->vqs[0].tx_waiting = qemu_get_be32(f);
1322 
1323     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1324 
1325     if (version_id >= 3)
1326         n->status = qemu_get_be16(f);
1327 
1328     if (version_id >= 4) {
1329         if (version_id < 8) {
1330             n->promisc = qemu_get_be32(f);
1331             n->allmulti = qemu_get_be32(f);
1332         } else {
1333             n->promisc = qemu_get_byte(f);
1334             n->allmulti = qemu_get_byte(f);
1335         }
1336     }
1337 
1338     if (version_id >= 5) {
1339         n->mac_table.in_use = qemu_get_be32(f);
1340         /* MAC_TABLE_ENTRIES may be different from the saved image */
1341         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1342             qemu_get_buffer(f, n->mac_table.macs,
1343                             n->mac_table.in_use * ETH_ALEN);
1344         } else if (n->mac_table.in_use) {
1345             uint8_t *buf = g_malloc0(n->mac_table.in_use);
1346             qemu_get_buffer(f, buf, n->mac_table.in_use * ETH_ALEN);
1347             g_free(buf);
1348             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1349             n->mac_table.in_use = 0;
1350         }
1351     }
1352 
1353     if (version_id >= 6)
1354         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1355 
1356     if (version_id >= 7) {
1357         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1358             error_report("virtio-net: saved image requires vnet_hdr=on");
1359             return -1;
1360         }
1361     }
1362 
1363     if (version_id >= 9) {
1364         n->mac_table.multi_overflow = qemu_get_byte(f);
1365         n->mac_table.uni_overflow = qemu_get_byte(f);
1366     }
1367 
1368     if (version_id >= 10) {
1369         n->alluni = qemu_get_byte(f);
1370         n->nomulti = qemu_get_byte(f);
1371         n->nouni = qemu_get_byte(f);
1372         n->nobcast = qemu_get_byte(f);
1373     }
1374 
1375     if (version_id >= 11) {
1376         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1377             error_report("virtio-net: saved image requires TUN_F_UFO support");
1378             return -1;
1379         }
1380     }
1381 
1382     if (n->max_queues > 1) {
1383         if (n->max_queues != qemu_get_be16(f)) {
1384             error_report("virtio-net: different max_queues ");
1385             return -1;
1386         }
1387 
1388         n->curr_queues = qemu_get_be16(f);
1389         for (i = 1; i < n->curr_queues; i++) {
1390             n->vqs[i].tx_waiting = qemu_get_be32(f);
1391         }
1392     }
1393 
1394     if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) {
1395         n->curr_guest_offloads = qemu_get_be64(f);
1396     } else {
1397         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
1398     }
1399 
1400     if (peer_has_vnet_hdr(n)) {
1401         virtio_net_apply_guest_offloads(n);
1402     }
1403 
1404     virtio_net_set_queues(n);
1405 
1406     /* Find the first multicast entry in the saved MAC filter */
1407     for (i = 0; i < n->mac_table.in_use; i++) {
1408         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1409             break;
1410         }
1411     }
1412     n->mac_table.first_multi = i;
1413 
1414     /* nc.link_down can't be migrated, so infer link_down according
1415      * to link status bit in n->status */
1416     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1417     for (i = 0; i < n->max_queues; i++) {
1418         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1419     }
1420 
1421     return 0;
1422 }
1423 
1424 static void virtio_net_cleanup(NetClientState *nc)
1425 {
1426     VirtIONet *n = qemu_get_nic_opaque(nc);
1427 
1428     n->nic = NULL;
1429 }
1430 
1431 static NetClientInfo net_virtio_info = {
1432     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1433     .size = sizeof(NICState),
1434     .can_receive = virtio_net_can_receive,
1435     .receive = virtio_net_receive,
1436     .cleanup = virtio_net_cleanup,
1437     .link_status_changed = virtio_net_set_link_status,
1438     .query_rx_filter = virtio_net_query_rxfilter,
1439 };
1440 
1441 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1442 {
1443     VirtIONet *n = VIRTIO_NET(vdev);
1444     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1445     assert(n->vhost_started);
1446     return vhost_net_virtqueue_pending(tap_get_vhost_net(nc->peer), idx);
1447 }
1448 
1449 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1450                                            bool mask)
1451 {
1452     VirtIONet *n = VIRTIO_NET(vdev);
1453     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1454     assert(n->vhost_started);
1455     vhost_net_virtqueue_mask(tap_get_vhost_net(nc->peer),
1456                              vdev, idx, mask);
1457 }
1458 
1459 void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features)
1460 {
1461     int i, config_size = 0;
1462     host_features |= (1 << VIRTIO_NET_F_MAC);
1463     for (i = 0; feature_sizes[i].flags != 0; i++) {
1464         if (host_features & feature_sizes[i].flags) {
1465             config_size = MAX(feature_sizes[i].end, config_size);
1466         }
1467     }
1468     n->config_size = config_size;
1469 }
1470 
1471 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
1472                                    const char *type)
1473 {
1474     /*
1475      * The name can be NULL, the netclient name will be type.x.
1476      */
1477     assert(type != NULL);
1478 
1479     if (n->netclient_name) {
1480         g_free(n->netclient_name);
1481         n->netclient_name = NULL;
1482     }
1483     if (n->netclient_type) {
1484         g_free(n->netclient_type);
1485         n->netclient_type = NULL;
1486     }
1487 
1488     if (name != NULL) {
1489         n->netclient_name = g_strdup(name);
1490     }
1491     n->netclient_type = g_strdup(type);
1492 }
1493 
1494 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
1495 {
1496     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1497     VirtIONet *n = VIRTIO_NET(dev);
1498     NetClientState *nc;
1499     int i;
1500 
1501     virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
1502 
1503     n->max_queues = MAX(n->nic_conf.queues, 1);
1504     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1505     n->vqs[0].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1506     n->curr_queues = 1;
1507     n->vqs[0].n = n;
1508     n->tx_timeout = n->net_conf.txtimer;
1509 
1510     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1511                        && strcmp(n->net_conf.tx, "bh")) {
1512         error_report("virtio-net: "
1513                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1514                      n->net_conf.tx);
1515         error_report("Defaulting to \"bh\"");
1516     }
1517 
1518     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1519         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1520                                            virtio_net_handle_tx_timer);
1521         n->vqs[0].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, virtio_net_tx_timer,
1522                                                &n->vqs[0]);
1523     } else {
1524         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1525                                            virtio_net_handle_tx_bh);
1526         n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]);
1527     }
1528     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1529     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1530     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1531     n->status = VIRTIO_NET_S_LINK_UP;
1532 
1533     if (n->netclient_type) {
1534         /*
1535          * Happen when virtio_net_set_netclient_name has been called.
1536          */
1537         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1538                               n->netclient_type, n->netclient_name, n);
1539     } else {
1540         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1541                               object_get_typename(OBJECT(dev)), dev->id, n);
1542     }
1543 
1544     peer_test_vnet_hdr(n);
1545     if (peer_has_vnet_hdr(n)) {
1546         for (i = 0; i < n->max_queues; i++) {
1547             tap_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1548         }
1549         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1550     } else {
1551         n->host_hdr_len = 0;
1552     }
1553 
1554     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1555 
1556     n->vqs[0].tx_waiting = 0;
1557     n->tx_burst = n->net_conf.txburst;
1558     virtio_net_set_mrg_rx_bufs(n, 0);
1559     n->promisc = 1; /* for compatibility */
1560 
1561     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1562 
1563     n->vlans = g_malloc0(MAX_VLAN >> 3);
1564 
1565     nc = qemu_get_queue(n->nic);
1566     nc->rxfilter_notify_enabled = 1;
1567 
1568     n->qdev = dev;
1569     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1570                     virtio_net_save, virtio_net_load, n);
1571 
1572     add_boot_device_path(n->nic_conf.bootindex, dev, "/ethernet-phy@0");
1573 }
1574 
1575 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
1576 {
1577     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1578     VirtIONet *n = VIRTIO_NET(dev);
1579     int i;
1580 
1581     /* This will stop vhost backend if appropriate. */
1582     virtio_net_set_status(vdev, 0);
1583 
1584     unregister_savevm(dev, "virtio-net", n);
1585 
1586     if (n->netclient_name) {
1587         g_free(n->netclient_name);
1588         n->netclient_name = NULL;
1589     }
1590     if (n->netclient_type) {
1591         g_free(n->netclient_type);
1592         n->netclient_type = NULL;
1593     }
1594 
1595     g_free(n->mac_table.macs);
1596     g_free(n->vlans);
1597 
1598     for (i = 0; i < n->max_queues; i++) {
1599         VirtIONetQueue *q = &n->vqs[i];
1600         NetClientState *nc = qemu_get_subqueue(n->nic, i);
1601 
1602         qemu_purge_queued_packets(nc);
1603 
1604         if (q->tx_timer) {
1605             timer_del(q->tx_timer);
1606             timer_free(q->tx_timer);
1607         } else if (q->tx_bh) {
1608             qemu_bh_delete(q->tx_bh);
1609         }
1610     }
1611 
1612     g_free(n->vqs);
1613     qemu_del_nic(n->nic);
1614     virtio_cleanup(vdev);
1615 }
1616 
1617 static void virtio_net_instance_init(Object *obj)
1618 {
1619     VirtIONet *n = VIRTIO_NET(obj);
1620 
1621     /*
1622      * The default config_size is sizeof(struct virtio_net_config).
1623      * Can be overriden with virtio_net_set_config_size.
1624      */
1625     n->config_size = sizeof(struct virtio_net_config);
1626 }
1627 
1628 static Property virtio_net_properties[] = {
1629     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
1630     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
1631                                                TX_TIMER_INTERVAL),
1632     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
1633     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
1634     DEFINE_PROP_END_OF_LIST(),
1635 };
1636 
1637 static void virtio_net_class_init(ObjectClass *klass, void *data)
1638 {
1639     DeviceClass *dc = DEVICE_CLASS(klass);
1640     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1641 
1642     dc->props = virtio_net_properties;
1643     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
1644     vdc->realize = virtio_net_device_realize;
1645     vdc->unrealize = virtio_net_device_unrealize;
1646     vdc->get_config = virtio_net_get_config;
1647     vdc->set_config = virtio_net_set_config;
1648     vdc->get_features = virtio_net_get_features;
1649     vdc->set_features = virtio_net_set_features;
1650     vdc->bad_features = virtio_net_bad_features;
1651     vdc->reset = virtio_net_reset;
1652     vdc->set_status = virtio_net_set_status;
1653     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
1654     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
1655 }
1656 
1657 static const TypeInfo virtio_net_info = {
1658     .name = TYPE_VIRTIO_NET,
1659     .parent = TYPE_VIRTIO_DEVICE,
1660     .instance_size = sizeof(VirtIONet),
1661     .instance_init = virtio_net_instance_init,
1662     .class_init = virtio_net_class_init,
1663 };
1664 
1665 static void virtio_register_types(void)
1666 {
1667     type_register_static(&virtio_net_info);
1668 }
1669 
1670 type_init(virtio_register_types)
1671