xref: /openbmc/qemu/hw/net/virtio-net.c (revision e0ab7fac)
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 (!(vdev->guest_features >> VIRTIO_NET_F_CTRL_MAC_ADDR & 1) &&
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 (!!n->vhost_started ==
124         (virtio_net_started(n, status) && !nc->peer->link_down)) {
125         return;
126     }
127     if (!n->vhost_started) {
128         int r;
129         if (!vhost_net_query(get_vhost_net(nc->peer), vdev)) {
130             return;
131         }
132         n->vhost_started = 1;
133         r = vhost_net_start(vdev, n->nic->ncs, queues);
134         if (r < 0) {
135             error_report("unable to start vhost net: %d: "
136                          "falling back on userspace virtio", -r);
137             n->vhost_started = 0;
138         }
139     } else {
140         vhost_net_stop(vdev, n->nic->ncs, queues);
141         n->vhost_started = 0;
142     }
143 }
144 
145 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
146 {
147     VirtIONet *n = VIRTIO_NET(vdev);
148     VirtIONetQueue *q;
149     int i;
150     uint8_t queue_status;
151 
152     virtio_net_vhost_status(n, status);
153 
154     for (i = 0; i < n->max_queues; i++) {
155         q = &n->vqs[i];
156 
157         if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
158             queue_status = 0;
159         } else {
160             queue_status = status;
161         }
162 
163         if (!q->tx_waiting) {
164             continue;
165         }
166 
167         if (virtio_net_started(n, queue_status) && !n->vhost_started) {
168             if (q->tx_timer) {
169                 timer_mod(q->tx_timer,
170                                qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
171             } else {
172                 qemu_bh_schedule(q->tx_bh);
173             }
174         } else {
175             if (q->tx_timer) {
176                 timer_del(q->tx_timer);
177             } else {
178                 qemu_bh_cancel(q->tx_bh);
179             }
180         }
181     }
182 }
183 
184 static void virtio_net_set_link_status(NetClientState *nc)
185 {
186     VirtIONet *n = qemu_get_nic_opaque(nc);
187     VirtIODevice *vdev = VIRTIO_DEVICE(n);
188     uint16_t old_status = n->status;
189 
190     if (nc->link_down)
191         n->status &= ~VIRTIO_NET_S_LINK_UP;
192     else
193         n->status |= VIRTIO_NET_S_LINK_UP;
194 
195     if (n->status != old_status)
196         virtio_notify_config(vdev);
197 
198     virtio_net_set_status(vdev, vdev->status);
199 }
200 
201 static void rxfilter_notify(NetClientState *nc)
202 {
203     VirtIONet *n = qemu_get_nic_opaque(nc);
204 
205     if (nc->rxfilter_notify_enabled) {
206         gchar *path = object_get_canonical_path(OBJECT(n->qdev));
207         qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
208                                               n->netclient_name, path, &error_abort);
209         g_free(path);
210 
211         /* disable event notification to avoid events flooding */
212         nc->rxfilter_notify_enabled = 0;
213     }
214 }
215 
216 static char *mac_strdup_printf(const uint8_t *mac)
217 {
218     return g_strdup_printf("%.2x:%.2x:%.2x:%.2x:%.2x:%.2x", mac[0],
219                             mac[1], mac[2], mac[3], mac[4], mac[5]);
220 }
221 
222 static intList *get_vlan_table(VirtIONet *n)
223 {
224     intList *list, *entry;
225     int i, j;
226 
227     list = NULL;
228     for (i = 0; i < MAX_VLAN >> 5; i++) {
229         for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
230             if (n->vlans[i] & (1U << j)) {
231                 entry = g_malloc0(sizeof(*entry));
232                 entry->value = (i << 5) + j;
233                 entry->next = list;
234                 list = entry;
235             }
236         }
237     }
238 
239     return list;
240 }
241 
242 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
243 {
244     VirtIONet *n = qemu_get_nic_opaque(nc);
245     VirtIODevice *vdev = VIRTIO_DEVICE(n);
246     RxFilterInfo *info;
247     strList *str_list, *entry;
248     int i;
249 
250     info = g_malloc0(sizeof(*info));
251     info->name = g_strdup(nc->name);
252     info->promiscuous = n->promisc;
253 
254     if (n->nouni) {
255         info->unicast = RX_STATE_NONE;
256     } else if (n->alluni) {
257         info->unicast = RX_STATE_ALL;
258     } else {
259         info->unicast = RX_STATE_NORMAL;
260     }
261 
262     if (n->nomulti) {
263         info->multicast = RX_STATE_NONE;
264     } else if (n->allmulti) {
265         info->multicast = RX_STATE_ALL;
266     } else {
267         info->multicast = RX_STATE_NORMAL;
268     }
269 
270     info->broadcast_allowed = n->nobcast;
271     info->multicast_overflow = n->mac_table.multi_overflow;
272     info->unicast_overflow = n->mac_table.uni_overflow;
273 
274     info->main_mac = mac_strdup_printf(n->mac);
275 
276     str_list = NULL;
277     for (i = 0; i < n->mac_table.first_multi; i++) {
278         entry = g_malloc0(sizeof(*entry));
279         entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
280         entry->next = str_list;
281         str_list = entry;
282     }
283     info->unicast_table = str_list;
284 
285     str_list = NULL;
286     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
287         entry = g_malloc0(sizeof(*entry));
288         entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
289         entry->next = str_list;
290         str_list = entry;
291     }
292     info->multicast_table = str_list;
293     info->vlan_table = get_vlan_table(n);
294 
295     if (!((1 << VIRTIO_NET_F_CTRL_VLAN) & vdev->guest_features)) {
296         info->vlan = RX_STATE_ALL;
297     } else if (!info->vlan_table) {
298         info->vlan = RX_STATE_NONE;
299     } else {
300         info->vlan = RX_STATE_NORMAL;
301     }
302 
303     /* enable event notification after query */
304     nc->rxfilter_notify_enabled = 1;
305 
306     return info;
307 }
308 
309 static void virtio_net_reset(VirtIODevice *vdev)
310 {
311     VirtIONet *n = VIRTIO_NET(vdev);
312 
313     /* Reset back to compatibility mode */
314     n->promisc = 1;
315     n->allmulti = 0;
316     n->alluni = 0;
317     n->nomulti = 0;
318     n->nouni = 0;
319     n->nobcast = 0;
320     /* multiqueue is disabled by default */
321     n->curr_queues = 1;
322     timer_del(n->announce_timer);
323     n->announce_counter = 0;
324     n->status &= ~VIRTIO_NET_S_ANNOUNCE;
325 
326     /* Flush any MAC and VLAN filter table state */
327     n->mac_table.in_use = 0;
328     n->mac_table.first_multi = 0;
329     n->mac_table.multi_overflow = 0;
330     n->mac_table.uni_overflow = 0;
331     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
332     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
333     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
334     memset(n->vlans, 0, MAX_VLAN >> 3);
335 }
336 
337 static void peer_test_vnet_hdr(VirtIONet *n)
338 {
339     NetClientState *nc = qemu_get_queue(n->nic);
340     if (!nc->peer) {
341         return;
342     }
343 
344     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
345 }
346 
347 static int peer_has_vnet_hdr(VirtIONet *n)
348 {
349     return n->has_vnet_hdr;
350 }
351 
352 static int peer_has_ufo(VirtIONet *n)
353 {
354     if (!peer_has_vnet_hdr(n))
355         return 0;
356 
357     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
358 
359     return n->has_ufo;
360 }
361 
362 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs)
363 {
364     int i;
365     NetClientState *nc;
366 
367     n->mergeable_rx_bufs = mergeable_rx_bufs;
368 
369     n->guest_hdr_len = n->mergeable_rx_bufs ?
370         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
371 
372     for (i = 0; i < n->max_queues; i++) {
373         nc = qemu_get_subqueue(n->nic, i);
374 
375         if (peer_has_vnet_hdr(n) &&
376             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
377             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
378             n->host_hdr_len = n->guest_hdr_len;
379         }
380     }
381 }
382 
383 static int peer_attach(VirtIONet *n, int index)
384 {
385     NetClientState *nc = qemu_get_subqueue(n->nic, index);
386 
387     if (!nc->peer) {
388         return 0;
389     }
390 
391     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
392         return 0;
393     }
394 
395     return tap_enable(nc->peer);
396 }
397 
398 static int peer_detach(VirtIONet *n, int index)
399 {
400     NetClientState *nc = qemu_get_subqueue(n->nic, index);
401 
402     if (!nc->peer) {
403         return 0;
404     }
405 
406     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
407         return 0;
408     }
409 
410     return tap_disable(nc->peer);
411 }
412 
413 static void virtio_net_set_queues(VirtIONet *n)
414 {
415     int i;
416     int r;
417 
418     for (i = 0; i < n->max_queues; i++) {
419         if (i < n->curr_queues) {
420             r = peer_attach(n, i);
421             assert(!r);
422         } else {
423             r = peer_detach(n, i);
424             assert(!r);
425         }
426     }
427 }
428 
429 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
430 
431 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
432 {
433     VirtIONet *n = VIRTIO_NET(vdev);
434     NetClientState *nc = qemu_get_queue(n->nic);
435 
436     features |= (1 << VIRTIO_NET_F_MAC);
437 
438     if (!peer_has_vnet_hdr(n)) {
439         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
440         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
441         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
442         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
443 
444         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
445         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
446         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
447         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
448     }
449 
450     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
451         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
452         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
453     }
454 
455     if (!get_vhost_net(nc->peer)) {
456         return features;
457     }
458     return vhost_net_get_features(get_vhost_net(nc->peer), features);
459 }
460 
461 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
462 {
463     uint32_t features = 0;
464 
465     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
466      * but also these: */
467     features |= (1 << VIRTIO_NET_F_MAC);
468     features |= (1 << VIRTIO_NET_F_CSUM);
469     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
470     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
471     features |= (1 << VIRTIO_NET_F_HOST_ECN);
472 
473     return features;
474 }
475 
476 static void virtio_net_apply_guest_offloads(VirtIONet *n)
477 {
478     qemu_set_offload(qemu_get_queue(n->nic)->peer,
479             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
480             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
481             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
482             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
483             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
484 }
485 
486 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
487 {
488     static const uint64_t guest_offloads_mask =
489         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
490         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
491         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
492         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
493         (1ULL << VIRTIO_NET_F_GUEST_UFO);
494 
495     return guest_offloads_mask & features;
496 }
497 
498 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
499 {
500     VirtIODevice *vdev = VIRTIO_DEVICE(n);
501     return virtio_net_guest_offloads_by_features(vdev->guest_features);
502 }
503 
504 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
505 {
506     VirtIONet *n = VIRTIO_NET(vdev);
507     int i;
508 
509     virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)));
510 
511     virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
512 
513     if (n->has_vnet_hdr) {
514         n->curr_guest_offloads =
515             virtio_net_guest_offloads_by_features(features);
516         virtio_net_apply_guest_offloads(n);
517     }
518 
519     for (i = 0;  i < n->max_queues; i++) {
520         NetClientState *nc = qemu_get_subqueue(n->nic, i);
521 
522         if (!get_vhost_net(nc->peer)) {
523             continue;
524         }
525         vhost_net_ack_features(get_vhost_net(nc->peer), features);
526     }
527 
528     if ((1 << VIRTIO_NET_F_CTRL_VLAN) & features) {
529         memset(n->vlans, 0, MAX_VLAN >> 3);
530     } else {
531         memset(n->vlans, 0xff, MAX_VLAN >> 3);
532     }
533 }
534 
535 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
536                                      struct iovec *iov, unsigned int iov_cnt)
537 {
538     uint8_t on;
539     size_t s;
540     NetClientState *nc = qemu_get_queue(n->nic);
541 
542     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
543     if (s != sizeof(on)) {
544         return VIRTIO_NET_ERR;
545     }
546 
547     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
548         n->promisc = on;
549     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
550         n->allmulti = on;
551     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
552         n->alluni = on;
553     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
554         n->nomulti = on;
555     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
556         n->nouni = on;
557     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
558         n->nobcast = on;
559     } else {
560         return VIRTIO_NET_ERR;
561     }
562 
563     rxfilter_notify(nc);
564 
565     return VIRTIO_NET_OK;
566 }
567 
568 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
569                                      struct iovec *iov, unsigned int iov_cnt)
570 {
571     VirtIODevice *vdev = VIRTIO_DEVICE(n);
572     uint64_t offloads;
573     size_t s;
574 
575     if (!((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features)) {
576         return VIRTIO_NET_ERR;
577     }
578 
579     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
580     if (s != sizeof(offloads)) {
581         return VIRTIO_NET_ERR;
582     }
583 
584     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
585         uint64_t supported_offloads;
586 
587         if (!n->has_vnet_hdr) {
588             return VIRTIO_NET_ERR;
589         }
590 
591         supported_offloads = virtio_net_supported_guest_offloads(n);
592         if (offloads & ~supported_offloads) {
593             return VIRTIO_NET_ERR;
594         }
595 
596         n->curr_guest_offloads = offloads;
597         virtio_net_apply_guest_offloads(n);
598 
599         return VIRTIO_NET_OK;
600     } else {
601         return VIRTIO_NET_ERR;
602     }
603 }
604 
605 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
606                                  struct iovec *iov, unsigned int iov_cnt)
607 {
608     VirtIODevice *vdev = VIRTIO_DEVICE(n);
609     struct virtio_net_ctrl_mac mac_data;
610     size_t s;
611     NetClientState *nc = qemu_get_queue(n->nic);
612 
613     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
614         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
615             return VIRTIO_NET_ERR;
616         }
617         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
618         assert(s == sizeof(n->mac));
619         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
620         rxfilter_notify(nc);
621 
622         return VIRTIO_NET_OK;
623     }
624 
625     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
626         return VIRTIO_NET_ERR;
627     }
628 
629     int in_use = 0;
630     int first_multi = 0;
631     uint8_t uni_overflow = 0;
632     uint8_t multi_overflow = 0;
633     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
634 
635     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
636                    sizeof(mac_data.entries));
637     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
638     if (s != sizeof(mac_data.entries)) {
639         goto error;
640     }
641     iov_discard_front(&iov, &iov_cnt, s);
642 
643     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
644         goto error;
645     }
646 
647     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
648         s = iov_to_buf(iov, iov_cnt, 0, macs,
649                        mac_data.entries * ETH_ALEN);
650         if (s != mac_data.entries * ETH_ALEN) {
651             goto error;
652         }
653         in_use += mac_data.entries;
654     } else {
655         uni_overflow = 1;
656     }
657 
658     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
659 
660     first_multi = in_use;
661 
662     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
663                    sizeof(mac_data.entries));
664     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
665     if (s != sizeof(mac_data.entries)) {
666         goto error;
667     }
668 
669     iov_discard_front(&iov, &iov_cnt, s);
670 
671     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
672         goto error;
673     }
674 
675     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
676         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
677                        mac_data.entries * ETH_ALEN);
678         if (s != mac_data.entries * ETH_ALEN) {
679             goto error;
680         }
681         in_use += mac_data.entries;
682     } else {
683         multi_overflow = 1;
684     }
685 
686     n->mac_table.in_use = in_use;
687     n->mac_table.first_multi = first_multi;
688     n->mac_table.uni_overflow = uni_overflow;
689     n->mac_table.multi_overflow = multi_overflow;
690     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
691     g_free(macs);
692     rxfilter_notify(nc);
693 
694     return VIRTIO_NET_OK;
695 
696 error:
697     g_free(macs);
698     return VIRTIO_NET_ERR;
699 }
700 
701 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
702                                         struct iovec *iov, unsigned int iov_cnt)
703 {
704     VirtIODevice *vdev = VIRTIO_DEVICE(n);
705     uint16_t vid;
706     size_t s;
707     NetClientState *nc = qemu_get_queue(n->nic);
708 
709     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
710     vid = virtio_lduw_p(vdev, &vid);
711     if (s != sizeof(vid)) {
712         return VIRTIO_NET_ERR;
713     }
714 
715     if (vid >= MAX_VLAN)
716         return VIRTIO_NET_ERR;
717 
718     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
719         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
720     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
721         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
722     else
723         return VIRTIO_NET_ERR;
724 
725     rxfilter_notify(nc);
726 
727     return VIRTIO_NET_OK;
728 }
729 
730 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
731                                       struct iovec *iov, unsigned int iov_cnt)
732 {
733     if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
734         n->status & VIRTIO_NET_S_ANNOUNCE) {
735         n->status &= ~VIRTIO_NET_S_ANNOUNCE;
736         if (n->announce_counter) {
737             timer_mod(n->announce_timer,
738                       qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
739                       self_announce_delay(n->announce_counter));
740         }
741         return VIRTIO_NET_OK;
742     } else {
743         return VIRTIO_NET_ERR;
744     }
745 }
746 
747 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
748                                 struct iovec *iov, unsigned int iov_cnt)
749 {
750     VirtIODevice *vdev = VIRTIO_DEVICE(n);
751     struct virtio_net_ctrl_mq mq;
752     size_t s;
753     uint16_t queues;
754 
755     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
756     if (s != sizeof(mq)) {
757         return VIRTIO_NET_ERR;
758     }
759 
760     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
761         return VIRTIO_NET_ERR;
762     }
763 
764     queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
765 
766     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
767         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
768         queues > n->max_queues ||
769         !n->multiqueue) {
770         return VIRTIO_NET_ERR;
771     }
772 
773     n->curr_queues = queues;
774     /* stop the backend before changing the number of queues to avoid handling a
775      * disabled queue */
776     virtio_net_set_status(vdev, vdev->status);
777     virtio_net_set_queues(n);
778 
779     return VIRTIO_NET_OK;
780 }
781 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
782 {
783     VirtIONet *n = VIRTIO_NET(vdev);
784     struct virtio_net_ctrl_hdr ctrl;
785     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
786     VirtQueueElement elem;
787     size_t s;
788     struct iovec *iov;
789     unsigned int iov_cnt;
790 
791     while (virtqueue_pop(vq, &elem)) {
792         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
793             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
794             error_report("virtio-net ctrl missing headers");
795             exit(1);
796         }
797 
798         iov = elem.out_sg;
799         iov_cnt = elem.out_num;
800         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
801         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
802         if (s != sizeof(ctrl)) {
803             status = VIRTIO_NET_ERR;
804         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
805             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
806         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
807             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
808         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
809             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
810         } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
811             status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
812         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
813             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
814         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
815             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
816         }
817 
818         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
819         assert(s == sizeof(status));
820 
821         virtqueue_push(vq, &elem, sizeof(status));
822         virtio_notify(vdev, vq);
823     }
824 }
825 
826 /* RX */
827 
828 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
829 {
830     VirtIONet *n = VIRTIO_NET(vdev);
831     int queue_index = vq2q(virtio_get_queue_index(vq));
832 
833     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
834 }
835 
836 static int virtio_net_can_receive(NetClientState *nc)
837 {
838     VirtIONet *n = qemu_get_nic_opaque(nc);
839     VirtIODevice *vdev = VIRTIO_DEVICE(n);
840     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
841 
842     if (!vdev->vm_running) {
843         return 0;
844     }
845 
846     if (nc->queue_index >= n->curr_queues) {
847         return 0;
848     }
849 
850     if (!virtio_queue_ready(q->rx_vq) ||
851         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
852         return 0;
853     }
854 
855     return 1;
856 }
857 
858 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
859 {
860     VirtIONet *n = q->n;
861     if (virtio_queue_empty(q->rx_vq) ||
862         (n->mergeable_rx_bufs &&
863          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
864         virtio_queue_set_notification(q->rx_vq, 1);
865 
866         /* To avoid a race condition where the guest has made some buffers
867          * available after the above check but before notification was
868          * enabled, check for available buffers again.
869          */
870         if (virtio_queue_empty(q->rx_vq) ||
871             (n->mergeable_rx_bufs &&
872              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
873             return 0;
874         }
875     }
876 
877     virtio_queue_set_notification(q->rx_vq, 0);
878     return 1;
879 }
880 
881 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
882 {
883     virtio_tswap16s(vdev, &hdr->hdr_len);
884     virtio_tswap16s(vdev, &hdr->gso_size);
885     virtio_tswap16s(vdev, &hdr->csum_start);
886     virtio_tswap16s(vdev, &hdr->csum_offset);
887 }
888 
889 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
890  * it never finds out that the packets don't have valid checksums.  This
891  * causes dhclient to get upset.  Fedora's carried a patch for ages to
892  * fix this with Xen but it hasn't appeared in an upstream release of
893  * dhclient yet.
894  *
895  * To avoid breaking existing guests, we catch udp packets and add
896  * checksums.  This is terrible but it's better than hacking the guest
897  * kernels.
898  *
899  * N.B. if we introduce a zero-copy API, this operation is no longer free so
900  * we should provide a mechanism to disable it to avoid polluting the host
901  * cache.
902  */
903 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
904                                         uint8_t *buf, size_t size)
905 {
906     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
907         (size > 27 && size < 1500) && /* normal sized MTU */
908         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
909         (buf[23] == 17) && /* ip.protocol == UDP */
910         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
911         net_checksum_calculate(buf, size);
912         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
913     }
914 }
915 
916 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
917                            const void *buf, size_t size)
918 {
919     if (n->has_vnet_hdr) {
920         /* FIXME this cast is evil */
921         void *wbuf = (void *)buf;
922         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
923                                     size - n->host_hdr_len);
924         virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
925         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
926     } else {
927         struct virtio_net_hdr hdr = {
928             .flags = 0,
929             .gso_type = VIRTIO_NET_HDR_GSO_NONE
930         };
931         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
932     }
933 }
934 
935 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
936 {
937     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
938     static const uint8_t vlan[] = {0x81, 0x00};
939     uint8_t *ptr = (uint8_t *)buf;
940     int i;
941 
942     if (n->promisc)
943         return 1;
944 
945     ptr += n->host_hdr_len;
946 
947     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
948         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
949         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
950             return 0;
951     }
952 
953     if (ptr[0] & 1) { // multicast
954         if (!memcmp(ptr, bcast, sizeof(bcast))) {
955             return !n->nobcast;
956         } else if (n->nomulti) {
957             return 0;
958         } else if (n->allmulti || n->mac_table.multi_overflow) {
959             return 1;
960         }
961 
962         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
963             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
964                 return 1;
965             }
966         }
967     } else { // unicast
968         if (n->nouni) {
969             return 0;
970         } else if (n->alluni || n->mac_table.uni_overflow) {
971             return 1;
972         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
973             return 1;
974         }
975 
976         for (i = 0; i < n->mac_table.first_multi; i++) {
977             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
978                 return 1;
979             }
980         }
981     }
982 
983     return 0;
984 }
985 
986 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
987 {
988     VirtIONet *n = qemu_get_nic_opaque(nc);
989     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
990     VirtIODevice *vdev = VIRTIO_DEVICE(n);
991     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
992     struct virtio_net_hdr_mrg_rxbuf mhdr;
993     unsigned mhdr_cnt = 0;
994     size_t offset, i, guest_offset;
995 
996     if (!virtio_net_can_receive(nc)) {
997         return -1;
998     }
999 
1000     /* hdr_len refers to the header we supply to the guest */
1001     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1002         return 0;
1003     }
1004 
1005     if (!receive_filter(n, buf, size))
1006         return size;
1007 
1008     offset = i = 0;
1009 
1010     while (offset < size) {
1011         VirtQueueElement elem;
1012         int len, total;
1013         const struct iovec *sg = elem.in_sg;
1014 
1015         total = 0;
1016 
1017         if (virtqueue_pop(q->rx_vq, &elem) == 0) {
1018             if (i == 0)
1019                 return -1;
1020             error_report("virtio-net unexpected empty queue: "
1021                     "i %zd mergeable %d offset %zd, size %zd, "
1022                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
1023                     i, n->mergeable_rx_bufs, offset, size,
1024                     n->guest_hdr_len, n->host_hdr_len, vdev->guest_features);
1025             exit(1);
1026         }
1027 
1028         if (elem.in_num < 1) {
1029             error_report("virtio-net receive queue contains no in buffers");
1030             exit(1);
1031         }
1032 
1033         if (i == 0) {
1034             assert(offset == 0);
1035             if (n->mergeable_rx_bufs) {
1036                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1037                                     sg, elem.in_num,
1038                                     offsetof(typeof(mhdr), num_buffers),
1039                                     sizeof(mhdr.num_buffers));
1040             }
1041 
1042             receive_header(n, sg, elem.in_num, buf, size);
1043             offset = n->host_hdr_len;
1044             total += n->guest_hdr_len;
1045             guest_offset = n->guest_hdr_len;
1046         } else {
1047             guest_offset = 0;
1048         }
1049 
1050         /* copy in packet.  ugh */
1051         len = iov_from_buf(sg, elem.in_num, guest_offset,
1052                            buf + offset, size - offset);
1053         total += len;
1054         offset += len;
1055         /* If buffers can't be merged, at this point we
1056          * must have consumed the complete packet.
1057          * Otherwise, drop it. */
1058         if (!n->mergeable_rx_bufs && offset < size) {
1059 #if 0
1060             error_report("virtio-net truncated non-mergeable packet: "
1061                          "i %zd mergeable %d offset %zd, size %zd, "
1062                          "guest hdr len %zd, host hdr len %zd",
1063                          i, n->mergeable_rx_bufs,
1064                          offset, size, n->guest_hdr_len, n->host_hdr_len);
1065 #endif
1066             return size;
1067         }
1068 
1069         /* signal other side */
1070         virtqueue_fill(q->rx_vq, &elem, total, i++);
1071     }
1072 
1073     if (mhdr_cnt) {
1074         virtio_stw_p(vdev, &mhdr.num_buffers, i);
1075         iov_from_buf(mhdr_sg, mhdr_cnt,
1076                      0,
1077                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1078     }
1079 
1080     virtqueue_flush(q->rx_vq, i);
1081     virtio_notify(vdev, q->rx_vq);
1082 
1083     return size;
1084 }
1085 
1086 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1087 
1088 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1089 {
1090     VirtIONet *n = qemu_get_nic_opaque(nc);
1091     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1092     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1093 
1094     virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
1095     virtio_notify(vdev, q->tx_vq);
1096 
1097     q->async_tx.elem.out_num = q->async_tx.len = 0;
1098 
1099     virtio_queue_set_notification(q->tx_vq, 1);
1100     virtio_net_flush_tx(q);
1101 }
1102 
1103 /* TX */
1104 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
1105 {
1106     VirtIONet *n = q->n;
1107     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1108     VirtQueueElement elem;
1109     int32_t num_packets = 0;
1110     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
1111     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1112         return num_packets;
1113     }
1114 
1115     assert(vdev->vm_running);
1116 
1117     if (q->async_tx.elem.out_num) {
1118         virtio_queue_set_notification(q->tx_vq, 0);
1119         return num_packets;
1120     }
1121 
1122     while (virtqueue_pop(q->tx_vq, &elem)) {
1123         ssize_t ret, len;
1124         unsigned int out_num = elem.out_num;
1125         struct iovec *out_sg = &elem.out_sg[0];
1126         struct iovec sg[VIRTQUEUE_MAX_SIZE];
1127 
1128         if (out_num < 1) {
1129             error_report("virtio-net header not in first element");
1130             exit(1);
1131         }
1132 
1133         if (n->has_vnet_hdr) {
1134             if (out_sg[0].iov_len < n->guest_hdr_len) {
1135                 error_report("virtio-net header incorrect");
1136                 exit(1);
1137             }
1138             virtio_net_hdr_swap(vdev, (void *) out_sg[0].iov_base);
1139         }
1140 
1141         /*
1142          * If host wants to see the guest header as is, we can
1143          * pass it on unchanged. Otherwise, copy just the parts
1144          * that host is interested in.
1145          */
1146         assert(n->host_hdr_len <= n->guest_hdr_len);
1147         if (n->host_hdr_len != n->guest_hdr_len) {
1148             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
1149                                        out_sg, out_num,
1150                                        0, n->host_hdr_len);
1151             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
1152                              out_sg, out_num,
1153                              n->guest_hdr_len, -1);
1154             out_num = sg_num;
1155             out_sg = sg;
1156         }
1157 
1158         len = n->guest_hdr_len;
1159 
1160         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
1161                                       out_sg, out_num, virtio_net_tx_complete);
1162         if (ret == 0) {
1163             virtio_queue_set_notification(q->tx_vq, 0);
1164             q->async_tx.elem = elem;
1165             q->async_tx.len  = len;
1166             return -EBUSY;
1167         }
1168 
1169         len += ret;
1170 
1171         virtqueue_push(q->tx_vq, &elem, 0);
1172         virtio_notify(vdev, q->tx_vq);
1173 
1174         if (++num_packets >= n->tx_burst) {
1175             break;
1176         }
1177     }
1178     return num_packets;
1179 }
1180 
1181 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
1182 {
1183     VirtIONet *n = VIRTIO_NET(vdev);
1184     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1185 
1186     /* This happens when device was stopped but VCPU wasn't. */
1187     if (!vdev->vm_running) {
1188         q->tx_waiting = 1;
1189         return;
1190     }
1191 
1192     if (q->tx_waiting) {
1193         virtio_queue_set_notification(vq, 1);
1194         timer_del(q->tx_timer);
1195         q->tx_waiting = 0;
1196         virtio_net_flush_tx(q);
1197     } else {
1198         timer_mod(q->tx_timer,
1199                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
1200         q->tx_waiting = 1;
1201         virtio_queue_set_notification(vq, 0);
1202     }
1203 }
1204 
1205 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1206 {
1207     VirtIONet *n = VIRTIO_NET(vdev);
1208     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1209 
1210     if (unlikely(q->tx_waiting)) {
1211         return;
1212     }
1213     q->tx_waiting = 1;
1214     /* This happens when device was stopped but VCPU wasn't. */
1215     if (!vdev->vm_running) {
1216         return;
1217     }
1218     virtio_queue_set_notification(vq, 0);
1219     qemu_bh_schedule(q->tx_bh);
1220 }
1221 
1222 static void virtio_net_tx_timer(void *opaque)
1223 {
1224     VirtIONetQueue *q = opaque;
1225     VirtIONet *n = q->n;
1226     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1227     assert(vdev->vm_running);
1228 
1229     q->tx_waiting = 0;
1230 
1231     /* Just in case the driver is not ready on more */
1232     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1233         return;
1234     }
1235 
1236     virtio_queue_set_notification(q->tx_vq, 1);
1237     virtio_net_flush_tx(q);
1238 }
1239 
1240 static void virtio_net_tx_bh(void *opaque)
1241 {
1242     VirtIONetQueue *q = opaque;
1243     VirtIONet *n = q->n;
1244     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1245     int32_t ret;
1246 
1247     assert(vdev->vm_running);
1248 
1249     q->tx_waiting = 0;
1250 
1251     /* Just in case the driver is not ready on more */
1252     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1253         return;
1254     }
1255 
1256     ret = virtio_net_flush_tx(q);
1257     if (ret == -EBUSY) {
1258         return; /* Notification re-enable handled by tx_complete */
1259     }
1260 
1261     /* If we flush a full burst of packets, assume there are
1262      * more coming and immediately reschedule */
1263     if (ret >= n->tx_burst) {
1264         qemu_bh_schedule(q->tx_bh);
1265         q->tx_waiting = 1;
1266         return;
1267     }
1268 
1269     /* If less than a full burst, re-enable notification and flush
1270      * anything that may have come in while we weren't looking.  If
1271      * we find something, assume the guest is still active and reschedule */
1272     virtio_queue_set_notification(q->tx_vq, 1);
1273     if (virtio_net_flush_tx(q) > 0) {
1274         virtio_queue_set_notification(q->tx_vq, 0);
1275         qemu_bh_schedule(q->tx_bh);
1276         q->tx_waiting = 1;
1277     }
1278 }
1279 
1280 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1281 {
1282     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1283     int i, max = multiqueue ? n->max_queues : 1;
1284 
1285     n->multiqueue = multiqueue;
1286 
1287     for (i = 2; i <= n->max_queues * 2 + 1; i++) {
1288         virtio_del_queue(vdev, i);
1289     }
1290 
1291     for (i = 1; i < max; i++) {
1292         n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1293         if (n->vqs[i].tx_timer) {
1294             n->vqs[i].tx_vq =
1295                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1296             n->vqs[i].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
1297                                                    virtio_net_tx_timer,
1298                                                    &n->vqs[i]);
1299         } else {
1300             n->vqs[i].tx_vq =
1301                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1302             n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1303         }
1304 
1305         n->vqs[i].tx_waiting = 0;
1306         n->vqs[i].n = n;
1307     }
1308 
1309     /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack
1310      * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid
1311      * breaking them.
1312      */
1313     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1314 
1315     virtio_net_set_queues(n);
1316 }
1317 
1318 static void virtio_net_save(QEMUFile *f, void *opaque)
1319 {
1320     VirtIONet *n = opaque;
1321     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1322 
1323     /* At this point, backend must be stopped, otherwise
1324      * it might keep writing to memory. */
1325     assert(!n->vhost_started);
1326     virtio_save(vdev, f);
1327 }
1328 
1329 static void virtio_net_save_device(VirtIODevice *vdev, QEMUFile *f)
1330 {
1331     VirtIONet *n = VIRTIO_NET(vdev);
1332     int i;
1333 
1334     qemu_put_buffer(f, n->mac, ETH_ALEN);
1335     qemu_put_be32(f, n->vqs[0].tx_waiting);
1336     qemu_put_be32(f, n->mergeable_rx_bufs);
1337     qemu_put_be16(f, n->status);
1338     qemu_put_byte(f, n->promisc);
1339     qemu_put_byte(f, n->allmulti);
1340     qemu_put_be32(f, n->mac_table.in_use);
1341     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1342     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1343     qemu_put_be32(f, n->has_vnet_hdr);
1344     qemu_put_byte(f, n->mac_table.multi_overflow);
1345     qemu_put_byte(f, n->mac_table.uni_overflow);
1346     qemu_put_byte(f, n->alluni);
1347     qemu_put_byte(f, n->nomulti);
1348     qemu_put_byte(f, n->nouni);
1349     qemu_put_byte(f, n->nobcast);
1350     qemu_put_byte(f, n->has_ufo);
1351     if (n->max_queues > 1) {
1352         qemu_put_be16(f, n->max_queues);
1353         qemu_put_be16(f, n->curr_queues);
1354         for (i = 1; i < n->curr_queues; i++) {
1355             qemu_put_be32(f, n->vqs[i].tx_waiting);
1356         }
1357     }
1358 
1359     if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) {
1360         qemu_put_be64(f, n->curr_guest_offloads);
1361     }
1362 }
1363 
1364 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1365 {
1366     VirtIONet *n = opaque;
1367     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1368 
1369     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1370         return -EINVAL;
1371 
1372     return virtio_load(vdev, f, version_id);
1373 }
1374 
1375 static int virtio_net_load_device(VirtIODevice *vdev, QEMUFile *f,
1376                                   int version_id)
1377 {
1378     VirtIONet *n = VIRTIO_NET(vdev);
1379     int i, link_down;
1380 
1381     qemu_get_buffer(f, n->mac, ETH_ALEN);
1382     n->vqs[0].tx_waiting = qemu_get_be32(f);
1383 
1384     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1385 
1386     if (version_id >= 3)
1387         n->status = qemu_get_be16(f);
1388 
1389     if (version_id >= 4) {
1390         if (version_id < 8) {
1391             n->promisc = qemu_get_be32(f);
1392             n->allmulti = qemu_get_be32(f);
1393         } else {
1394             n->promisc = qemu_get_byte(f);
1395             n->allmulti = qemu_get_byte(f);
1396         }
1397     }
1398 
1399     if (version_id >= 5) {
1400         n->mac_table.in_use = qemu_get_be32(f);
1401         /* MAC_TABLE_ENTRIES may be different from the saved image */
1402         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1403             qemu_get_buffer(f, n->mac_table.macs,
1404                             n->mac_table.in_use * ETH_ALEN);
1405         } else {
1406             int64_t i;
1407 
1408             /* Overflow detected - can happen if source has a larger MAC table.
1409              * We simply set overflow flag so there's no need to maintain the
1410              * table of addresses, discard them all.
1411              * Note: 64 bit math to avoid integer overflow.
1412              */
1413             for (i = 0; i < (int64_t)n->mac_table.in_use * ETH_ALEN; ++i) {
1414                 qemu_get_byte(f);
1415             }
1416             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1417             n->mac_table.in_use = 0;
1418         }
1419     }
1420 
1421     if (version_id >= 6)
1422         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1423 
1424     if (version_id >= 7) {
1425         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1426             error_report("virtio-net: saved image requires vnet_hdr=on");
1427             return -1;
1428         }
1429     }
1430 
1431     if (version_id >= 9) {
1432         n->mac_table.multi_overflow = qemu_get_byte(f);
1433         n->mac_table.uni_overflow = qemu_get_byte(f);
1434     }
1435 
1436     if (version_id >= 10) {
1437         n->alluni = qemu_get_byte(f);
1438         n->nomulti = qemu_get_byte(f);
1439         n->nouni = qemu_get_byte(f);
1440         n->nobcast = qemu_get_byte(f);
1441     }
1442 
1443     if (version_id >= 11) {
1444         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1445             error_report("virtio-net: saved image requires TUN_F_UFO support");
1446             return -1;
1447         }
1448     }
1449 
1450     if (n->max_queues > 1) {
1451         if (n->max_queues != qemu_get_be16(f)) {
1452             error_report("virtio-net: different max_queues ");
1453             return -1;
1454         }
1455 
1456         n->curr_queues = qemu_get_be16(f);
1457         if (n->curr_queues > n->max_queues) {
1458             error_report("virtio-net: curr_queues %x > max_queues %x",
1459                          n->curr_queues, n->max_queues);
1460             return -1;
1461         }
1462         for (i = 1; i < n->curr_queues; i++) {
1463             n->vqs[i].tx_waiting = qemu_get_be32(f);
1464         }
1465     }
1466 
1467     if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) {
1468         n->curr_guest_offloads = qemu_get_be64(f);
1469     } else {
1470         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
1471     }
1472 
1473     if (peer_has_vnet_hdr(n)) {
1474         virtio_net_apply_guest_offloads(n);
1475     }
1476 
1477     virtio_net_set_queues(n);
1478 
1479     /* Find the first multicast entry in the saved MAC filter */
1480     for (i = 0; i < n->mac_table.in_use; i++) {
1481         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1482             break;
1483         }
1484     }
1485     n->mac_table.first_multi = i;
1486 
1487     /* nc.link_down can't be migrated, so infer link_down according
1488      * to link status bit in n->status */
1489     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1490     for (i = 0; i < n->max_queues; i++) {
1491         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1492     }
1493 
1494     if (vdev->guest_features & (0x1 << VIRTIO_NET_F_GUEST_ANNOUNCE) &&
1495         vdev->guest_features & (0x1 << VIRTIO_NET_F_CTRL_VQ)) {
1496         n->announce_counter = SELF_ANNOUNCE_ROUNDS;
1497         timer_mod(n->announce_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL));
1498     }
1499 
1500     return 0;
1501 }
1502 
1503 static void virtio_net_cleanup(NetClientState *nc)
1504 {
1505     VirtIONet *n = qemu_get_nic_opaque(nc);
1506 
1507     n->nic = NULL;
1508 }
1509 
1510 static NetClientInfo net_virtio_info = {
1511     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1512     .size = sizeof(NICState),
1513     .can_receive = virtio_net_can_receive,
1514     .receive = virtio_net_receive,
1515     .cleanup = virtio_net_cleanup,
1516     .link_status_changed = virtio_net_set_link_status,
1517     .query_rx_filter = virtio_net_query_rxfilter,
1518 };
1519 
1520 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1521 {
1522     VirtIONet *n = VIRTIO_NET(vdev);
1523     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1524     assert(n->vhost_started);
1525     return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
1526 }
1527 
1528 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1529                                            bool mask)
1530 {
1531     VirtIONet *n = VIRTIO_NET(vdev);
1532     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1533     assert(n->vhost_started);
1534     vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
1535                              vdev, idx, mask);
1536 }
1537 
1538 void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features)
1539 {
1540     int i, config_size = 0;
1541     host_features |= (1 << VIRTIO_NET_F_MAC);
1542     for (i = 0; feature_sizes[i].flags != 0; i++) {
1543         if (host_features & feature_sizes[i].flags) {
1544             config_size = MAX(feature_sizes[i].end, config_size);
1545         }
1546     }
1547     n->config_size = config_size;
1548 }
1549 
1550 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
1551                                    const char *type)
1552 {
1553     /*
1554      * The name can be NULL, the netclient name will be type.x.
1555      */
1556     assert(type != NULL);
1557 
1558     g_free(n->netclient_name);
1559     g_free(n->netclient_type);
1560     n->netclient_name = g_strdup(name);
1561     n->netclient_type = g_strdup(type);
1562 }
1563 
1564 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
1565 {
1566     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1567     VirtIONet *n = VIRTIO_NET(dev);
1568     NetClientState *nc;
1569     int i;
1570 
1571     virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
1572 
1573     n->max_queues = MAX(n->nic_conf.peers.queues, 1);
1574     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1575     n->vqs[0].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1576     n->curr_queues = 1;
1577     n->vqs[0].n = n;
1578     n->tx_timeout = n->net_conf.txtimer;
1579 
1580     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1581                        && strcmp(n->net_conf.tx, "bh")) {
1582         error_report("virtio-net: "
1583                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1584                      n->net_conf.tx);
1585         error_report("Defaulting to \"bh\"");
1586     }
1587 
1588     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1589         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1590                                            virtio_net_handle_tx_timer);
1591         n->vqs[0].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, virtio_net_tx_timer,
1592                                                &n->vqs[0]);
1593     } else {
1594         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1595                                            virtio_net_handle_tx_bh);
1596         n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]);
1597     }
1598     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1599     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1600     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1601     n->status = VIRTIO_NET_S_LINK_UP;
1602     n->announce_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
1603                                      virtio_net_announce_timer, n);
1604 
1605     if (n->netclient_type) {
1606         /*
1607          * Happen when virtio_net_set_netclient_name has been called.
1608          */
1609         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1610                               n->netclient_type, n->netclient_name, n);
1611     } else {
1612         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1613                               object_get_typename(OBJECT(dev)), dev->id, n);
1614     }
1615 
1616     peer_test_vnet_hdr(n);
1617     if (peer_has_vnet_hdr(n)) {
1618         for (i = 0; i < n->max_queues; i++) {
1619             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1620         }
1621         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1622     } else {
1623         n->host_hdr_len = 0;
1624     }
1625 
1626     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1627 
1628     n->vqs[0].tx_waiting = 0;
1629     n->tx_burst = n->net_conf.txburst;
1630     virtio_net_set_mrg_rx_bufs(n, 0);
1631     n->promisc = 1; /* for compatibility */
1632 
1633     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1634 
1635     n->vlans = g_malloc0(MAX_VLAN >> 3);
1636 
1637     nc = qemu_get_queue(n->nic);
1638     nc->rxfilter_notify_enabled = 1;
1639 
1640     n->qdev = dev;
1641     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1642                     virtio_net_save, virtio_net_load, n);
1643 
1644     add_boot_device_path(n->nic_conf.bootindex, dev, "/ethernet-phy@0");
1645 }
1646 
1647 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
1648 {
1649     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1650     VirtIONet *n = VIRTIO_NET(dev);
1651     int i;
1652 
1653     /* This will stop vhost backend if appropriate. */
1654     virtio_net_set_status(vdev, 0);
1655 
1656     unregister_savevm(dev, "virtio-net", n);
1657 
1658     g_free(n->netclient_name);
1659     n->netclient_name = NULL;
1660     g_free(n->netclient_type);
1661     n->netclient_type = NULL;
1662 
1663     g_free(n->mac_table.macs);
1664     g_free(n->vlans);
1665 
1666     for (i = 0; i < n->max_queues; i++) {
1667         VirtIONetQueue *q = &n->vqs[i];
1668         NetClientState *nc = qemu_get_subqueue(n->nic, i);
1669 
1670         qemu_purge_queued_packets(nc);
1671 
1672         if (q->tx_timer) {
1673             timer_del(q->tx_timer);
1674             timer_free(q->tx_timer);
1675         } else if (q->tx_bh) {
1676             qemu_bh_delete(q->tx_bh);
1677         }
1678     }
1679 
1680     timer_del(n->announce_timer);
1681     timer_free(n->announce_timer);
1682     g_free(n->vqs);
1683     qemu_del_nic(n->nic);
1684     virtio_cleanup(vdev);
1685 }
1686 
1687 static void virtio_net_instance_init(Object *obj)
1688 {
1689     VirtIONet *n = VIRTIO_NET(obj);
1690 
1691     /*
1692      * The default config_size is sizeof(struct virtio_net_config).
1693      * Can be overriden with virtio_net_set_config_size.
1694      */
1695     n->config_size = sizeof(struct virtio_net_config);
1696 }
1697 
1698 static Property virtio_net_properties[] = {
1699     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
1700     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
1701                                                TX_TIMER_INTERVAL),
1702     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
1703     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
1704     DEFINE_PROP_END_OF_LIST(),
1705 };
1706 
1707 static void virtio_net_class_init(ObjectClass *klass, void *data)
1708 {
1709     DeviceClass *dc = DEVICE_CLASS(klass);
1710     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1711 
1712     dc->props = virtio_net_properties;
1713     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
1714     vdc->realize = virtio_net_device_realize;
1715     vdc->unrealize = virtio_net_device_unrealize;
1716     vdc->get_config = virtio_net_get_config;
1717     vdc->set_config = virtio_net_set_config;
1718     vdc->get_features = virtio_net_get_features;
1719     vdc->set_features = virtio_net_set_features;
1720     vdc->bad_features = virtio_net_bad_features;
1721     vdc->reset = virtio_net_reset;
1722     vdc->set_status = virtio_net_set_status;
1723     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
1724     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
1725     vdc->load = virtio_net_load_device;
1726     vdc->save = virtio_net_save_device;
1727 }
1728 
1729 static const TypeInfo virtio_net_info = {
1730     .name = TYPE_VIRTIO_NET,
1731     .parent = TYPE_VIRTIO_DEVICE,
1732     .instance_size = sizeof(VirtIONet),
1733     .instance_init = virtio_net_instance_init,
1734     .class_init = virtio_net_class_init,
1735 };
1736 
1737 static void virtio_register_types(void)
1738 {
1739     type_register_static(&virtio_net_info);
1740 }
1741 
1742 type_init(virtio_register_types)
1743