xref: /openbmc/qemu/hw/net/virtio-net.c (revision c2b38b27)
1 /*
2  * Virtio Network Device
3  *
4  * Copyright IBM, Corp. 2007
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/osdep.h"
15 #include "qemu/iov.h"
16 #include "hw/virtio/virtio.h"
17 #include "net/net.h"
18 #include "net/checksum.h"
19 #include "net/tap.h"
20 #include "qemu/error-report.h"
21 #include "qemu/timer.h"
22 #include "hw/virtio/virtio-net.h"
23 #include "net/vhost_net.h"
24 #include "hw/virtio/virtio-bus.h"
25 #include "qapi/qmp/qjson.h"
26 #include "qapi-event.h"
27 #include "hw/virtio/virtio-access.h"
28 
29 #define VIRTIO_NET_VM_VERSION    11
30 
31 #define MAC_TABLE_ENTRIES    64
32 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
33 
34 /* previously fixed value */
35 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
36 /* for now, only allow larger queues; with virtio-1, guest can downsize */
37 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
38 
39 /*
40  * Calculate the number of bytes up to and including the given 'field' of
41  * 'container'.
42  */
43 #define endof(container, field) \
44     (offsetof(container, field) + sizeof(((container *)0)->field))
45 
46 typedef struct VirtIOFeature {
47     uint32_t flags;
48     size_t end;
49 } VirtIOFeature;
50 
51 static VirtIOFeature feature_sizes[] = {
52     {.flags = 1 << VIRTIO_NET_F_MAC,
53      .end = endof(struct virtio_net_config, mac)},
54     {.flags = 1 << VIRTIO_NET_F_STATUS,
55      .end = endof(struct virtio_net_config, status)},
56     {.flags = 1 << VIRTIO_NET_F_MQ,
57      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
58     {.flags = 1 << VIRTIO_NET_F_MTU,
59      .end = endof(struct virtio_net_config, mtu)},
60     {}
61 };
62 
63 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
64 {
65     VirtIONet *n = qemu_get_nic_opaque(nc);
66 
67     return &n->vqs[nc->queue_index];
68 }
69 
70 static int vq2q(int queue_index)
71 {
72     return queue_index / 2;
73 }
74 
75 /* TODO
76  * - we could suppress RX interrupt if we were so inclined.
77  */
78 
79 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
80 {
81     VirtIONet *n = VIRTIO_NET(vdev);
82     struct virtio_net_config netcfg;
83 
84     virtio_stw_p(vdev, &netcfg.status, n->status);
85     virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues);
86     virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
87     memcpy(netcfg.mac, n->mac, ETH_ALEN);
88     memcpy(config, &netcfg, n->config_size);
89 }
90 
91 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
92 {
93     VirtIONet *n = VIRTIO_NET(vdev);
94     struct virtio_net_config netcfg = {};
95 
96     memcpy(&netcfg, config, n->config_size);
97 
98     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
99         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
100         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
101         memcpy(n->mac, netcfg.mac, ETH_ALEN);
102         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
103     }
104 }
105 
106 static bool virtio_net_started(VirtIONet *n, uint8_t status)
107 {
108     VirtIODevice *vdev = VIRTIO_DEVICE(n);
109     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
110         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
111 }
112 
113 static void virtio_net_announce_timer(void *opaque)
114 {
115     VirtIONet *n = opaque;
116     VirtIODevice *vdev = VIRTIO_DEVICE(n);
117 
118     n->announce_counter--;
119     n->status |= VIRTIO_NET_S_ANNOUNCE;
120     virtio_notify_config(vdev);
121 }
122 
123 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
124 {
125     VirtIODevice *vdev = VIRTIO_DEVICE(n);
126     NetClientState *nc = qemu_get_queue(n->nic);
127     int queues = n->multiqueue ? n->max_queues : 1;
128 
129     if (!get_vhost_net(nc->peer)) {
130         return;
131     }
132 
133     if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
134         !!n->vhost_started) {
135         return;
136     }
137     if (!n->vhost_started) {
138         int r, i;
139 
140         if (n->needs_vnet_hdr_swap) {
141             error_report("backend does not support %s vnet headers; "
142                          "falling back on userspace virtio",
143                          virtio_is_big_endian(vdev) ? "BE" : "LE");
144             return;
145         }
146 
147         /* Any packets outstanding? Purge them to avoid touching rings
148          * when vhost is running.
149          */
150         for (i = 0;  i < queues; i++) {
151             NetClientState *qnc = qemu_get_subqueue(n->nic, i);
152 
153             /* Purge both directions: TX and RX. */
154             qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
155             qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
156         }
157 
158         if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
159             r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
160             if (r < 0) {
161                 error_report("%uBytes MTU not supported by the backend",
162                              n->net_conf.mtu);
163 
164                 return;
165             }
166         }
167 
168         n->vhost_started = 1;
169         r = vhost_net_start(vdev, n->nic->ncs, queues);
170         if (r < 0) {
171             error_report("unable to start vhost net: %d: "
172                          "falling back on userspace virtio", -r);
173             n->vhost_started = 0;
174         }
175     } else {
176         vhost_net_stop(vdev, n->nic->ncs, queues);
177         n->vhost_started = 0;
178     }
179 }
180 
181 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
182                                           NetClientState *peer,
183                                           bool enable)
184 {
185     if (virtio_is_big_endian(vdev)) {
186         return qemu_set_vnet_be(peer, enable);
187     } else {
188         return qemu_set_vnet_le(peer, enable);
189     }
190 }
191 
192 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
193                                        int queues, bool enable)
194 {
195     int i;
196 
197     for (i = 0; i < queues; i++) {
198         if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
199             enable) {
200             while (--i >= 0) {
201                 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
202             }
203 
204             return true;
205         }
206     }
207 
208     return false;
209 }
210 
211 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
212 {
213     VirtIODevice *vdev = VIRTIO_DEVICE(n);
214     int queues = n->multiqueue ? n->max_queues : 1;
215 
216     if (virtio_net_started(n, status)) {
217         /* Before using the device, we tell the network backend about the
218          * endianness to use when parsing vnet headers. If the backend
219          * can't do it, we fallback onto fixing the headers in the core
220          * virtio-net code.
221          */
222         n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
223                                                             queues, true);
224     } else if (virtio_net_started(n, vdev->status)) {
225         /* After using the device, we need to reset the network backend to
226          * the default (guest native endianness), otherwise the guest may
227          * lose network connectivity if it is rebooted into a different
228          * endianness.
229          */
230         virtio_net_set_vnet_endian(vdev, n->nic->ncs, queues, false);
231     }
232 }
233 
234 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq)
235 {
236     unsigned int dropped = virtqueue_drop_all(vq);
237     if (dropped) {
238         virtio_notify(vdev, vq);
239     }
240 }
241 
242 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
243 {
244     VirtIONet *n = VIRTIO_NET(vdev);
245     VirtIONetQueue *q;
246     int i;
247     uint8_t queue_status;
248 
249     virtio_net_vnet_endian_status(n, status);
250     virtio_net_vhost_status(n, status);
251 
252     for (i = 0; i < n->max_queues; i++) {
253         NetClientState *ncs = qemu_get_subqueue(n->nic, i);
254         bool queue_started;
255         q = &n->vqs[i];
256 
257         if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
258             queue_status = 0;
259         } else {
260             queue_status = status;
261         }
262         queue_started =
263             virtio_net_started(n, queue_status) && !n->vhost_started;
264 
265         if (queue_started) {
266             qemu_flush_queued_packets(ncs);
267         }
268 
269         if (!q->tx_waiting) {
270             continue;
271         }
272 
273         if (queue_started) {
274             if (q->tx_timer) {
275                 timer_mod(q->tx_timer,
276                                qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
277             } else {
278                 qemu_bh_schedule(q->tx_bh);
279             }
280         } else {
281             if (q->tx_timer) {
282                 timer_del(q->tx_timer);
283             } else {
284                 qemu_bh_cancel(q->tx_bh);
285             }
286             if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 &&
287                 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK)) {
288                 /* if tx is waiting we are likely have some packets in tx queue
289                  * and disabled notification */
290                 q->tx_waiting = 0;
291                 virtio_queue_set_notification(q->tx_vq, 1);
292                 virtio_net_drop_tx_queue_data(vdev, q->tx_vq);
293             }
294         }
295     }
296 }
297 
298 static void virtio_net_set_link_status(NetClientState *nc)
299 {
300     VirtIONet *n = qemu_get_nic_opaque(nc);
301     VirtIODevice *vdev = VIRTIO_DEVICE(n);
302     uint16_t old_status = n->status;
303 
304     if (nc->link_down)
305         n->status &= ~VIRTIO_NET_S_LINK_UP;
306     else
307         n->status |= VIRTIO_NET_S_LINK_UP;
308 
309     if (n->status != old_status)
310         virtio_notify_config(vdev);
311 
312     virtio_net_set_status(vdev, vdev->status);
313 }
314 
315 static void rxfilter_notify(NetClientState *nc)
316 {
317     VirtIONet *n = qemu_get_nic_opaque(nc);
318 
319     if (nc->rxfilter_notify_enabled) {
320         gchar *path = object_get_canonical_path(OBJECT(n->qdev));
321         qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
322                                               n->netclient_name, path, &error_abort);
323         g_free(path);
324 
325         /* disable event notification to avoid events flooding */
326         nc->rxfilter_notify_enabled = 0;
327     }
328 }
329 
330 static intList *get_vlan_table(VirtIONet *n)
331 {
332     intList *list, *entry;
333     int i, j;
334 
335     list = NULL;
336     for (i = 0; i < MAX_VLAN >> 5; i++) {
337         for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
338             if (n->vlans[i] & (1U << j)) {
339                 entry = g_malloc0(sizeof(*entry));
340                 entry->value = (i << 5) + j;
341                 entry->next = list;
342                 list = entry;
343             }
344         }
345     }
346 
347     return list;
348 }
349 
350 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
351 {
352     VirtIONet *n = qemu_get_nic_opaque(nc);
353     VirtIODevice *vdev = VIRTIO_DEVICE(n);
354     RxFilterInfo *info;
355     strList *str_list, *entry;
356     int i;
357 
358     info = g_malloc0(sizeof(*info));
359     info->name = g_strdup(nc->name);
360     info->promiscuous = n->promisc;
361 
362     if (n->nouni) {
363         info->unicast = RX_STATE_NONE;
364     } else if (n->alluni) {
365         info->unicast = RX_STATE_ALL;
366     } else {
367         info->unicast = RX_STATE_NORMAL;
368     }
369 
370     if (n->nomulti) {
371         info->multicast = RX_STATE_NONE;
372     } else if (n->allmulti) {
373         info->multicast = RX_STATE_ALL;
374     } else {
375         info->multicast = RX_STATE_NORMAL;
376     }
377 
378     info->broadcast_allowed = n->nobcast;
379     info->multicast_overflow = n->mac_table.multi_overflow;
380     info->unicast_overflow = n->mac_table.uni_overflow;
381 
382     info->main_mac = qemu_mac_strdup_printf(n->mac);
383 
384     str_list = NULL;
385     for (i = 0; i < n->mac_table.first_multi; i++) {
386         entry = g_malloc0(sizeof(*entry));
387         entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
388         entry->next = str_list;
389         str_list = entry;
390     }
391     info->unicast_table = str_list;
392 
393     str_list = NULL;
394     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
395         entry = g_malloc0(sizeof(*entry));
396         entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
397         entry->next = str_list;
398         str_list = entry;
399     }
400     info->multicast_table = str_list;
401     info->vlan_table = get_vlan_table(n);
402 
403     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
404         info->vlan = RX_STATE_ALL;
405     } else if (!info->vlan_table) {
406         info->vlan = RX_STATE_NONE;
407     } else {
408         info->vlan = RX_STATE_NORMAL;
409     }
410 
411     /* enable event notification after query */
412     nc->rxfilter_notify_enabled = 1;
413 
414     return info;
415 }
416 
417 static void virtio_net_reset(VirtIODevice *vdev)
418 {
419     VirtIONet *n = VIRTIO_NET(vdev);
420 
421     /* Reset back to compatibility mode */
422     n->promisc = 1;
423     n->allmulti = 0;
424     n->alluni = 0;
425     n->nomulti = 0;
426     n->nouni = 0;
427     n->nobcast = 0;
428     /* multiqueue is disabled by default */
429     n->curr_queues = 1;
430     timer_del(n->announce_timer);
431     n->announce_counter = 0;
432     n->status &= ~VIRTIO_NET_S_ANNOUNCE;
433 
434     /* Flush any MAC and VLAN filter table state */
435     n->mac_table.in_use = 0;
436     n->mac_table.first_multi = 0;
437     n->mac_table.multi_overflow = 0;
438     n->mac_table.uni_overflow = 0;
439     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
440     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
441     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
442     memset(n->vlans, 0, MAX_VLAN >> 3);
443 }
444 
445 static void peer_test_vnet_hdr(VirtIONet *n)
446 {
447     NetClientState *nc = qemu_get_queue(n->nic);
448     if (!nc->peer) {
449         return;
450     }
451 
452     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
453 }
454 
455 static int peer_has_vnet_hdr(VirtIONet *n)
456 {
457     return n->has_vnet_hdr;
458 }
459 
460 static int peer_has_ufo(VirtIONet *n)
461 {
462     if (!peer_has_vnet_hdr(n))
463         return 0;
464 
465     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
466 
467     return n->has_ufo;
468 }
469 
470 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
471                                        int version_1)
472 {
473     int i;
474     NetClientState *nc;
475 
476     n->mergeable_rx_bufs = mergeable_rx_bufs;
477 
478     if (version_1) {
479         n->guest_hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
480     } else {
481         n->guest_hdr_len = n->mergeable_rx_bufs ?
482             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
483             sizeof(struct virtio_net_hdr);
484     }
485 
486     for (i = 0; i < n->max_queues; i++) {
487         nc = qemu_get_subqueue(n->nic, i);
488 
489         if (peer_has_vnet_hdr(n) &&
490             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
491             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
492             n->host_hdr_len = n->guest_hdr_len;
493         }
494     }
495 }
496 
497 static int peer_attach(VirtIONet *n, int index)
498 {
499     NetClientState *nc = qemu_get_subqueue(n->nic, index);
500 
501     if (!nc->peer) {
502         return 0;
503     }
504 
505     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
506         vhost_set_vring_enable(nc->peer, 1);
507     }
508 
509     if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
510         return 0;
511     }
512 
513     return tap_enable(nc->peer);
514 }
515 
516 static int peer_detach(VirtIONet *n, int index)
517 {
518     NetClientState *nc = qemu_get_subqueue(n->nic, index);
519 
520     if (!nc->peer) {
521         return 0;
522     }
523 
524     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
525         vhost_set_vring_enable(nc->peer, 0);
526     }
527 
528     if (nc->peer->info->type !=  NET_CLIENT_DRIVER_TAP) {
529         return 0;
530     }
531 
532     return tap_disable(nc->peer);
533 }
534 
535 static void virtio_net_set_queues(VirtIONet *n)
536 {
537     int i;
538     int r;
539 
540     if (n->nic->peer_deleted) {
541         return;
542     }
543 
544     for (i = 0; i < n->max_queues; i++) {
545         if (i < n->curr_queues) {
546             r = peer_attach(n, i);
547             assert(!r);
548         } else {
549             r = peer_detach(n, i);
550             assert(!r);
551         }
552     }
553 }
554 
555 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
556 
557 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
558                                         Error **errp)
559 {
560     VirtIONet *n = VIRTIO_NET(vdev);
561     NetClientState *nc = qemu_get_queue(n->nic);
562 
563     /* Firstly sync all virtio-net possible supported features */
564     features |= n->host_features;
565 
566     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
567 
568     if (!peer_has_vnet_hdr(n)) {
569         virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
570         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
571         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
572         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
573 
574         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
575         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
576         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
577         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
578     }
579 
580     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
581         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
582         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
583     }
584 
585     if (!get_vhost_net(nc->peer)) {
586         return features;
587     }
588     return vhost_net_get_features(get_vhost_net(nc->peer), features);
589 }
590 
591 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
592 {
593     uint64_t features = 0;
594 
595     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
596      * but also these: */
597     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
598     virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
599     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
600     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
601     virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
602 
603     return features;
604 }
605 
606 static void virtio_net_apply_guest_offloads(VirtIONet *n)
607 {
608     qemu_set_offload(qemu_get_queue(n->nic)->peer,
609             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
610             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
611             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
612             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
613             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
614 }
615 
616 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
617 {
618     static const uint64_t guest_offloads_mask =
619         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
620         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
621         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
622         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
623         (1ULL << VIRTIO_NET_F_GUEST_UFO);
624 
625     return guest_offloads_mask & features;
626 }
627 
628 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
629 {
630     VirtIODevice *vdev = VIRTIO_DEVICE(n);
631     return virtio_net_guest_offloads_by_features(vdev->guest_features);
632 }
633 
634 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
635 {
636     VirtIONet *n = VIRTIO_NET(vdev);
637     int i;
638 
639     virtio_net_set_multiqueue(n,
640                               virtio_has_feature(features, VIRTIO_NET_F_MQ));
641 
642     virtio_net_set_mrg_rx_bufs(n,
643                                virtio_has_feature(features,
644                                                   VIRTIO_NET_F_MRG_RXBUF),
645                                virtio_has_feature(features,
646                                                   VIRTIO_F_VERSION_1));
647 
648     if (n->has_vnet_hdr) {
649         n->curr_guest_offloads =
650             virtio_net_guest_offloads_by_features(features);
651         virtio_net_apply_guest_offloads(n);
652     }
653 
654     for (i = 0;  i < n->max_queues; i++) {
655         NetClientState *nc = qemu_get_subqueue(n->nic, i);
656 
657         if (!get_vhost_net(nc->peer)) {
658             continue;
659         }
660         vhost_net_ack_features(get_vhost_net(nc->peer), features);
661     }
662 
663     if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
664         memset(n->vlans, 0, MAX_VLAN >> 3);
665     } else {
666         memset(n->vlans, 0xff, MAX_VLAN >> 3);
667     }
668 }
669 
670 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
671                                      struct iovec *iov, unsigned int iov_cnt)
672 {
673     uint8_t on;
674     size_t s;
675     NetClientState *nc = qemu_get_queue(n->nic);
676 
677     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
678     if (s != sizeof(on)) {
679         return VIRTIO_NET_ERR;
680     }
681 
682     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
683         n->promisc = on;
684     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
685         n->allmulti = on;
686     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
687         n->alluni = on;
688     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
689         n->nomulti = on;
690     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
691         n->nouni = on;
692     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
693         n->nobcast = on;
694     } else {
695         return VIRTIO_NET_ERR;
696     }
697 
698     rxfilter_notify(nc);
699 
700     return VIRTIO_NET_OK;
701 }
702 
703 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
704                                      struct iovec *iov, unsigned int iov_cnt)
705 {
706     VirtIODevice *vdev = VIRTIO_DEVICE(n);
707     uint64_t offloads;
708     size_t s;
709 
710     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
711         return VIRTIO_NET_ERR;
712     }
713 
714     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
715     if (s != sizeof(offloads)) {
716         return VIRTIO_NET_ERR;
717     }
718 
719     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
720         uint64_t supported_offloads;
721 
722         if (!n->has_vnet_hdr) {
723             return VIRTIO_NET_ERR;
724         }
725 
726         supported_offloads = virtio_net_supported_guest_offloads(n);
727         if (offloads & ~supported_offloads) {
728             return VIRTIO_NET_ERR;
729         }
730 
731         n->curr_guest_offloads = offloads;
732         virtio_net_apply_guest_offloads(n);
733 
734         return VIRTIO_NET_OK;
735     } else {
736         return VIRTIO_NET_ERR;
737     }
738 }
739 
740 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
741                                  struct iovec *iov, unsigned int iov_cnt)
742 {
743     VirtIODevice *vdev = VIRTIO_DEVICE(n);
744     struct virtio_net_ctrl_mac mac_data;
745     size_t s;
746     NetClientState *nc = qemu_get_queue(n->nic);
747 
748     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
749         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
750             return VIRTIO_NET_ERR;
751         }
752         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
753         assert(s == sizeof(n->mac));
754         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
755         rxfilter_notify(nc);
756 
757         return VIRTIO_NET_OK;
758     }
759 
760     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
761         return VIRTIO_NET_ERR;
762     }
763 
764     int in_use = 0;
765     int first_multi = 0;
766     uint8_t uni_overflow = 0;
767     uint8_t multi_overflow = 0;
768     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
769 
770     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
771                    sizeof(mac_data.entries));
772     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
773     if (s != sizeof(mac_data.entries)) {
774         goto error;
775     }
776     iov_discard_front(&iov, &iov_cnt, s);
777 
778     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
779         goto error;
780     }
781 
782     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
783         s = iov_to_buf(iov, iov_cnt, 0, macs,
784                        mac_data.entries * ETH_ALEN);
785         if (s != mac_data.entries * ETH_ALEN) {
786             goto error;
787         }
788         in_use += mac_data.entries;
789     } else {
790         uni_overflow = 1;
791     }
792 
793     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
794 
795     first_multi = in_use;
796 
797     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
798                    sizeof(mac_data.entries));
799     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
800     if (s != sizeof(mac_data.entries)) {
801         goto error;
802     }
803 
804     iov_discard_front(&iov, &iov_cnt, s);
805 
806     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
807         goto error;
808     }
809 
810     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
811         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
812                        mac_data.entries * ETH_ALEN);
813         if (s != mac_data.entries * ETH_ALEN) {
814             goto error;
815         }
816         in_use += mac_data.entries;
817     } else {
818         multi_overflow = 1;
819     }
820 
821     n->mac_table.in_use = in_use;
822     n->mac_table.first_multi = first_multi;
823     n->mac_table.uni_overflow = uni_overflow;
824     n->mac_table.multi_overflow = multi_overflow;
825     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
826     g_free(macs);
827     rxfilter_notify(nc);
828 
829     return VIRTIO_NET_OK;
830 
831 error:
832     g_free(macs);
833     return VIRTIO_NET_ERR;
834 }
835 
836 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
837                                         struct iovec *iov, unsigned int iov_cnt)
838 {
839     VirtIODevice *vdev = VIRTIO_DEVICE(n);
840     uint16_t vid;
841     size_t s;
842     NetClientState *nc = qemu_get_queue(n->nic);
843 
844     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
845     vid = virtio_lduw_p(vdev, &vid);
846     if (s != sizeof(vid)) {
847         return VIRTIO_NET_ERR;
848     }
849 
850     if (vid >= MAX_VLAN)
851         return VIRTIO_NET_ERR;
852 
853     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
854         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
855     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
856         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
857     else
858         return VIRTIO_NET_ERR;
859 
860     rxfilter_notify(nc);
861 
862     return VIRTIO_NET_OK;
863 }
864 
865 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
866                                       struct iovec *iov, unsigned int iov_cnt)
867 {
868     if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
869         n->status & VIRTIO_NET_S_ANNOUNCE) {
870         n->status &= ~VIRTIO_NET_S_ANNOUNCE;
871         if (n->announce_counter) {
872             timer_mod(n->announce_timer,
873                       qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
874                       self_announce_delay(n->announce_counter));
875         }
876         return VIRTIO_NET_OK;
877     } else {
878         return VIRTIO_NET_ERR;
879     }
880 }
881 
882 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
883                                 struct iovec *iov, unsigned int iov_cnt)
884 {
885     VirtIODevice *vdev = VIRTIO_DEVICE(n);
886     struct virtio_net_ctrl_mq mq;
887     size_t s;
888     uint16_t queues;
889 
890     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
891     if (s != sizeof(mq)) {
892         return VIRTIO_NET_ERR;
893     }
894 
895     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
896         return VIRTIO_NET_ERR;
897     }
898 
899     queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
900 
901     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
902         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
903         queues > n->max_queues ||
904         !n->multiqueue) {
905         return VIRTIO_NET_ERR;
906     }
907 
908     n->curr_queues = queues;
909     /* stop the backend before changing the number of queues to avoid handling a
910      * disabled queue */
911     virtio_net_set_status(vdev, vdev->status);
912     virtio_net_set_queues(n);
913 
914     return VIRTIO_NET_OK;
915 }
916 
917 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
918 {
919     VirtIONet *n = VIRTIO_NET(vdev);
920     struct virtio_net_ctrl_hdr ctrl;
921     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
922     VirtQueueElement *elem;
923     size_t s;
924     struct iovec *iov, *iov2;
925     unsigned int iov_cnt;
926 
927     for (;;) {
928         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
929         if (!elem) {
930             break;
931         }
932         if (iov_size(elem->in_sg, elem->in_num) < sizeof(status) ||
933             iov_size(elem->out_sg, elem->out_num) < sizeof(ctrl)) {
934             virtio_error(vdev, "virtio-net ctrl missing headers");
935             virtqueue_detach_element(vq, elem, 0);
936             g_free(elem);
937             break;
938         }
939 
940         iov_cnt = elem->out_num;
941         iov2 = iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num);
942         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
943         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
944         if (s != sizeof(ctrl)) {
945             status = VIRTIO_NET_ERR;
946         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
947             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
948         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
949             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
950         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
951             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
952         } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
953             status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
954         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
955             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
956         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
957             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
958         }
959 
960         s = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, sizeof(status));
961         assert(s == sizeof(status));
962 
963         virtqueue_push(vq, elem, sizeof(status));
964         virtio_notify(vdev, vq);
965         g_free(iov2);
966         g_free(elem);
967     }
968 }
969 
970 /* RX */
971 
972 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
973 {
974     VirtIONet *n = VIRTIO_NET(vdev);
975     int queue_index = vq2q(virtio_get_queue_index(vq));
976 
977     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
978 }
979 
980 static int virtio_net_can_receive(NetClientState *nc)
981 {
982     VirtIONet *n = qemu_get_nic_opaque(nc);
983     VirtIODevice *vdev = VIRTIO_DEVICE(n);
984     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
985 
986     if (!vdev->vm_running) {
987         return 0;
988     }
989 
990     if (nc->queue_index >= n->curr_queues) {
991         return 0;
992     }
993 
994     if (!virtio_queue_ready(q->rx_vq) ||
995         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
996         return 0;
997     }
998 
999     return 1;
1000 }
1001 
1002 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1003 {
1004     VirtIONet *n = q->n;
1005     if (virtio_queue_empty(q->rx_vq) ||
1006         (n->mergeable_rx_bufs &&
1007          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1008         virtio_queue_set_notification(q->rx_vq, 1);
1009 
1010         /* To avoid a race condition where the guest has made some buffers
1011          * available after the above check but before notification was
1012          * enabled, check for available buffers again.
1013          */
1014         if (virtio_queue_empty(q->rx_vq) ||
1015             (n->mergeable_rx_bufs &&
1016              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1017             return 0;
1018         }
1019     }
1020 
1021     virtio_queue_set_notification(q->rx_vq, 0);
1022     return 1;
1023 }
1024 
1025 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1026 {
1027     virtio_tswap16s(vdev, &hdr->hdr_len);
1028     virtio_tswap16s(vdev, &hdr->gso_size);
1029     virtio_tswap16s(vdev, &hdr->csum_start);
1030     virtio_tswap16s(vdev, &hdr->csum_offset);
1031 }
1032 
1033 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1034  * it never finds out that the packets don't have valid checksums.  This
1035  * causes dhclient to get upset.  Fedora's carried a patch for ages to
1036  * fix this with Xen but it hasn't appeared in an upstream release of
1037  * dhclient yet.
1038  *
1039  * To avoid breaking existing guests, we catch udp packets and add
1040  * checksums.  This is terrible but it's better than hacking the guest
1041  * kernels.
1042  *
1043  * N.B. if we introduce a zero-copy API, this operation is no longer free so
1044  * we should provide a mechanism to disable it to avoid polluting the host
1045  * cache.
1046  */
1047 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1048                                         uint8_t *buf, size_t size)
1049 {
1050     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1051         (size > 27 && size < 1500) && /* normal sized MTU */
1052         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1053         (buf[23] == 17) && /* ip.protocol == UDP */
1054         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1055         net_checksum_calculate(buf, size);
1056         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1057     }
1058 }
1059 
1060 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1061                            const void *buf, size_t size)
1062 {
1063     if (n->has_vnet_hdr) {
1064         /* FIXME this cast is evil */
1065         void *wbuf = (void *)buf;
1066         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1067                                     size - n->host_hdr_len);
1068 
1069         if (n->needs_vnet_hdr_swap) {
1070             virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1071         }
1072         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1073     } else {
1074         struct virtio_net_hdr hdr = {
1075             .flags = 0,
1076             .gso_type = VIRTIO_NET_HDR_GSO_NONE
1077         };
1078         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1079     }
1080 }
1081 
1082 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1083 {
1084     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1085     static const uint8_t vlan[] = {0x81, 0x00};
1086     uint8_t *ptr = (uint8_t *)buf;
1087     int i;
1088 
1089     if (n->promisc)
1090         return 1;
1091 
1092     ptr += n->host_hdr_len;
1093 
1094     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1095         int vid = lduw_be_p(ptr + 14) & 0xfff;
1096         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1097             return 0;
1098     }
1099 
1100     if (ptr[0] & 1) { // multicast
1101         if (!memcmp(ptr, bcast, sizeof(bcast))) {
1102             return !n->nobcast;
1103         } else if (n->nomulti) {
1104             return 0;
1105         } else if (n->allmulti || n->mac_table.multi_overflow) {
1106             return 1;
1107         }
1108 
1109         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1110             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1111                 return 1;
1112             }
1113         }
1114     } else { // unicast
1115         if (n->nouni) {
1116             return 0;
1117         } else if (n->alluni || n->mac_table.uni_overflow) {
1118             return 1;
1119         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1120             return 1;
1121         }
1122 
1123         for (i = 0; i < n->mac_table.first_multi; i++) {
1124             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1125                 return 1;
1126             }
1127         }
1128     }
1129 
1130     return 0;
1131 }
1132 
1133 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
1134                                       size_t size)
1135 {
1136     VirtIONet *n = qemu_get_nic_opaque(nc);
1137     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1138     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1139     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1140     struct virtio_net_hdr_mrg_rxbuf mhdr;
1141     unsigned mhdr_cnt = 0;
1142     size_t offset, i, guest_offset;
1143 
1144     if (!virtio_net_can_receive(nc)) {
1145         return -1;
1146     }
1147 
1148     /* hdr_len refers to the header we supply to the guest */
1149     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1150         return 0;
1151     }
1152 
1153     if (!receive_filter(n, buf, size))
1154         return size;
1155 
1156     offset = i = 0;
1157 
1158     while (offset < size) {
1159         VirtQueueElement *elem;
1160         int len, total;
1161         const struct iovec *sg;
1162 
1163         total = 0;
1164 
1165         elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1166         if (!elem) {
1167             if (i) {
1168                 virtio_error(vdev, "virtio-net unexpected empty queue: "
1169                              "i %zd mergeable %d offset %zd, size %zd, "
1170                              "guest hdr len %zd, host hdr len %zd "
1171                              "guest features 0x%" PRIx64,
1172                              i, n->mergeable_rx_bufs, offset, size,
1173                              n->guest_hdr_len, n->host_hdr_len,
1174                              vdev->guest_features);
1175             }
1176             return -1;
1177         }
1178 
1179         if (elem->in_num < 1) {
1180             virtio_error(vdev,
1181                          "virtio-net receive queue contains no in buffers");
1182             virtqueue_detach_element(q->rx_vq, elem, 0);
1183             g_free(elem);
1184             return -1;
1185         }
1186 
1187         sg = elem->in_sg;
1188         if (i == 0) {
1189             assert(offset == 0);
1190             if (n->mergeable_rx_bufs) {
1191                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1192                                     sg, elem->in_num,
1193                                     offsetof(typeof(mhdr), num_buffers),
1194                                     sizeof(mhdr.num_buffers));
1195             }
1196 
1197             receive_header(n, sg, elem->in_num, buf, size);
1198             offset = n->host_hdr_len;
1199             total += n->guest_hdr_len;
1200             guest_offset = n->guest_hdr_len;
1201         } else {
1202             guest_offset = 0;
1203         }
1204 
1205         /* copy in packet.  ugh */
1206         len = iov_from_buf(sg, elem->in_num, guest_offset,
1207                            buf + offset, size - offset);
1208         total += len;
1209         offset += len;
1210         /* If buffers can't be merged, at this point we
1211          * must have consumed the complete packet.
1212          * Otherwise, drop it. */
1213         if (!n->mergeable_rx_bufs && offset < size) {
1214             virtqueue_unpop(q->rx_vq, elem, total);
1215             g_free(elem);
1216             return size;
1217         }
1218 
1219         /* signal other side */
1220         virtqueue_fill(q->rx_vq, elem, total, i++);
1221         g_free(elem);
1222     }
1223 
1224     if (mhdr_cnt) {
1225         virtio_stw_p(vdev, &mhdr.num_buffers, i);
1226         iov_from_buf(mhdr_sg, mhdr_cnt,
1227                      0,
1228                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1229     }
1230 
1231     virtqueue_flush(q->rx_vq, i);
1232     virtio_notify(vdev, q->rx_vq);
1233 
1234     return size;
1235 }
1236 
1237 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf,
1238                                   size_t size)
1239 {
1240     ssize_t r;
1241 
1242     rcu_read_lock();
1243     r = virtio_net_receive_rcu(nc, buf, size);
1244     rcu_read_unlock();
1245     return r;
1246 }
1247 
1248 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1249 
1250 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1251 {
1252     VirtIONet *n = qemu_get_nic_opaque(nc);
1253     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1254     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1255 
1256     virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
1257     virtio_notify(vdev, q->tx_vq);
1258 
1259     g_free(q->async_tx.elem);
1260     q->async_tx.elem = NULL;
1261 
1262     virtio_queue_set_notification(q->tx_vq, 1);
1263     virtio_net_flush_tx(q);
1264 }
1265 
1266 /* TX */
1267 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
1268 {
1269     VirtIONet *n = q->n;
1270     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1271     VirtQueueElement *elem;
1272     int32_t num_packets = 0;
1273     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
1274     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1275         return num_packets;
1276     }
1277 
1278     if (q->async_tx.elem) {
1279         virtio_queue_set_notification(q->tx_vq, 0);
1280         return num_packets;
1281     }
1282 
1283     for (;;) {
1284         ssize_t ret;
1285         unsigned int out_num;
1286         struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
1287         struct virtio_net_hdr_mrg_rxbuf mhdr;
1288 
1289         elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
1290         if (!elem) {
1291             break;
1292         }
1293 
1294         out_num = elem->out_num;
1295         out_sg = elem->out_sg;
1296         if (out_num < 1) {
1297             virtio_error(vdev, "virtio-net header not in first element");
1298             virtqueue_detach_element(q->tx_vq, elem, 0);
1299             g_free(elem);
1300             return -EINVAL;
1301         }
1302 
1303         if (n->has_vnet_hdr) {
1304             if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
1305                 n->guest_hdr_len) {
1306                 virtio_error(vdev, "virtio-net header incorrect");
1307                 virtqueue_detach_element(q->tx_vq, elem, 0);
1308                 g_free(elem);
1309                 return -EINVAL;
1310             }
1311             if (n->needs_vnet_hdr_swap) {
1312                 virtio_net_hdr_swap(vdev, (void *) &mhdr);
1313                 sg2[0].iov_base = &mhdr;
1314                 sg2[0].iov_len = n->guest_hdr_len;
1315                 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
1316                                    out_sg, out_num,
1317                                    n->guest_hdr_len, -1);
1318                 if (out_num == VIRTQUEUE_MAX_SIZE) {
1319                     goto drop;
1320 		}
1321                 out_num += 1;
1322                 out_sg = sg2;
1323 	    }
1324         }
1325         /*
1326          * If host wants to see the guest header as is, we can
1327          * pass it on unchanged. Otherwise, copy just the parts
1328          * that host is interested in.
1329          */
1330         assert(n->host_hdr_len <= n->guest_hdr_len);
1331         if (n->host_hdr_len != n->guest_hdr_len) {
1332             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
1333                                        out_sg, out_num,
1334                                        0, n->host_hdr_len);
1335             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
1336                              out_sg, out_num,
1337                              n->guest_hdr_len, -1);
1338             out_num = sg_num;
1339             out_sg = sg;
1340         }
1341 
1342         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
1343                                       out_sg, out_num, virtio_net_tx_complete);
1344         if (ret == 0) {
1345             virtio_queue_set_notification(q->tx_vq, 0);
1346             q->async_tx.elem = elem;
1347             return -EBUSY;
1348         }
1349 
1350 drop:
1351         virtqueue_push(q->tx_vq, elem, 0);
1352         virtio_notify(vdev, q->tx_vq);
1353         g_free(elem);
1354 
1355         if (++num_packets >= n->tx_burst) {
1356             break;
1357         }
1358     }
1359     return num_packets;
1360 }
1361 
1362 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
1363 {
1364     VirtIONet *n = VIRTIO_NET(vdev);
1365     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1366 
1367     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
1368         virtio_net_drop_tx_queue_data(vdev, vq);
1369         return;
1370     }
1371 
1372     /* This happens when device was stopped but VCPU wasn't. */
1373     if (!vdev->vm_running) {
1374         q->tx_waiting = 1;
1375         return;
1376     }
1377 
1378     if (q->tx_waiting) {
1379         virtio_queue_set_notification(vq, 1);
1380         timer_del(q->tx_timer);
1381         q->tx_waiting = 0;
1382         if (virtio_net_flush_tx(q) == -EINVAL) {
1383             return;
1384         }
1385     } else {
1386         timer_mod(q->tx_timer,
1387                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
1388         q->tx_waiting = 1;
1389         virtio_queue_set_notification(vq, 0);
1390     }
1391 }
1392 
1393 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1394 {
1395     VirtIONet *n = VIRTIO_NET(vdev);
1396     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1397 
1398     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
1399         virtio_net_drop_tx_queue_data(vdev, vq);
1400         return;
1401     }
1402 
1403     if (unlikely(q->tx_waiting)) {
1404         return;
1405     }
1406     q->tx_waiting = 1;
1407     /* This happens when device was stopped but VCPU wasn't. */
1408     if (!vdev->vm_running) {
1409         return;
1410     }
1411     virtio_queue_set_notification(vq, 0);
1412     qemu_bh_schedule(q->tx_bh);
1413 }
1414 
1415 static void virtio_net_tx_timer(void *opaque)
1416 {
1417     VirtIONetQueue *q = opaque;
1418     VirtIONet *n = q->n;
1419     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1420     /* This happens when device was stopped but BH wasn't. */
1421     if (!vdev->vm_running) {
1422         /* Make sure tx waiting is set, so we'll run when restarted. */
1423         assert(q->tx_waiting);
1424         return;
1425     }
1426 
1427     q->tx_waiting = 0;
1428 
1429     /* Just in case the driver is not ready on more */
1430     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1431         return;
1432     }
1433 
1434     virtio_queue_set_notification(q->tx_vq, 1);
1435     virtio_net_flush_tx(q);
1436 }
1437 
1438 static void virtio_net_tx_bh(void *opaque)
1439 {
1440     VirtIONetQueue *q = opaque;
1441     VirtIONet *n = q->n;
1442     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1443     int32_t ret;
1444 
1445     /* This happens when device was stopped but BH wasn't. */
1446     if (!vdev->vm_running) {
1447         /* Make sure tx waiting is set, so we'll run when restarted. */
1448         assert(q->tx_waiting);
1449         return;
1450     }
1451 
1452     q->tx_waiting = 0;
1453 
1454     /* Just in case the driver is not ready on more */
1455     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1456         return;
1457     }
1458 
1459     ret = virtio_net_flush_tx(q);
1460     if (ret == -EBUSY || ret == -EINVAL) {
1461         return; /* Notification re-enable handled by tx_complete or device
1462                  * broken */
1463     }
1464 
1465     /* If we flush a full burst of packets, assume there are
1466      * more coming and immediately reschedule */
1467     if (ret >= n->tx_burst) {
1468         qemu_bh_schedule(q->tx_bh);
1469         q->tx_waiting = 1;
1470         return;
1471     }
1472 
1473     /* If less than a full burst, re-enable notification and flush
1474      * anything that may have come in while we weren't looking.  If
1475      * we find something, assume the guest is still active and reschedule */
1476     virtio_queue_set_notification(q->tx_vq, 1);
1477     ret = virtio_net_flush_tx(q);
1478     if (ret == -EINVAL) {
1479         return;
1480     } else if (ret > 0) {
1481         virtio_queue_set_notification(q->tx_vq, 0);
1482         qemu_bh_schedule(q->tx_bh);
1483         q->tx_waiting = 1;
1484     }
1485 }
1486 
1487 static void virtio_net_add_queue(VirtIONet *n, int index)
1488 {
1489     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1490 
1491     n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
1492                                            virtio_net_handle_rx);
1493     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1494         n->vqs[index].tx_vq =
1495             virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1496         n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
1497                                               virtio_net_tx_timer,
1498                                               &n->vqs[index]);
1499     } else {
1500         n->vqs[index].tx_vq =
1501             virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1502         n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
1503     }
1504 
1505     n->vqs[index].tx_waiting = 0;
1506     n->vqs[index].n = n;
1507 }
1508 
1509 static void virtio_net_del_queue(VirtIONet *n, int index)
1510 {
1511     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1512     VirtIONetQueue *q = &n->vqs[index];
1513     NetClientState *nc = qemu_get_subqueue(n->nic, index);
1514 
1515     qemu_purge_queued_packets(nc);
1516 
1517     virtio_del_queue(vdev, index * 2);
1518     if (q->tx_timer) {
1519         timer_del(q->tx_timer);
1520         timer_free(q->tx_timer);
1521     } else {
1522         qemu_bh_delete(q->tx_bh);
1523     }
1524     virtio_del_queue(vdev, index * 2 + 1);
1525 }
1526 
1527 static void virtio_net_change_num_queues(VirtIONet *n, int new_max_queues)
1528 {
1529     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1530     int old_num_queues = virtio_get_num_queues(vdev);
1531     int new_num_queues = new_max_queues * 2 + 1;
1532     int i;
1533 
1534     assert(old_num_queues >= 3);
1535     assert(old_num_queues % 2 == 1);
1536 
1537     if (old_num_queues == new_num_queues) {
1538         return;
1539     }
1540 
1541     /*
1542      * We always need to remove and add ctrl vq if
1543      * old_num_queues != new_num_queues. Remove ctrl_vq first,
1544      * and then we only enter one of the following too loops.
1545      */
1546     virtio_del_queue(vdev, old_num_queues - 1);
1547 
1548     for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
1549         /* new_num_queues < old_num_queues */
1550         virtio_net_del_queue(n, i / 2);
1551     }
1552 
1553     for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
1554         /* new_num_queues > old_num_queues */
1555         virtio_net_add_queue(n, i / 2);
1556     }
1557 
1558     /* add ctrl_vq last */
1559     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1560 }
1561 
1562 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1563 {
1564     int max = multiqueue ? n->max_queues : 1;
1565 
1566     n->multiqueue = multiqueue;
1567     virtio_net_change_num_queues(n, max);
1568 
1569     virtio_net_set_queues(n);
1570 }
1571 
1572 static int virtio_net_post_load_device(void *opaque, int version_id)
1573 {
1574     VirtIONet *n = opaque;
1575     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1576     int i, link_down;
1577 
1578     virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
1579                                virtio_vdev_has_feature(vdev,
1580                                                        VIRTIO_F_VERSION_1));
1581 
1582     /* MAC_TABLE_ENTRIES may be different from the saved image */
1583     if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
1584         n->mac_table.in_use = 0;
1585     }
1586 
1587     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1588         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
1589     }
1590 
1591     if (peer_has_vnet_hdr(n)) {
1592         virtio_net_apply_guest_offloads(n);
1593     }
1594 
1595     virtio_net_set_queues(n);
1596 
1597     /* Find the first multicast entry in the saved MAC filter */
1598     for (i = 0; i < n->mac_table.in_use; i++) {
1599         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1600             break;
1601         }
1602     }
1603     n->mac_table.first_multi = i;
1604 
1605     /* nc.link_down can't be migrated, so infer link_down according
1606      * to link status bit in n->status */
1607     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1608     for (i = 0; i < n->max_queues; i++) {
1609         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1610     }
1611 
1612     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
1613         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
1614         n->announce_counter = SELF_ANNOUNCE_ROUNDS;
1615         timer_mod(n->announce_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL));
1616     }
1617 
1618     return 0;
1619 }
1620 
1621 /* tx_waiting field of a VirtIONetQueue */
1622 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = {
1623     .name = "virtio-net-queue-tx_waiting",
1624     .fields = (VMStateField[]) {
1625         VMSTATE_UINT32(tx_waiting, VirtIONetQueue),
1626         VMSTATE_END_OF_LIST()
1627    },
1628 };
1629 
1630 static bool max_queues_gt_1(void *opaque, int version_id)
1631 {
1632     return VIRTIO_NET(opaque)->max_queues > 1;
1633 }
1634 
1635 static bool has_ctrl_guest_offloads(void *opaque, int version_id)
1636 {
1637     return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque),
1638                                    VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
1639 }
1640 
1641 static bool mac_table_fits(void *opaque, int version_id)
1642 {
1643     return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES;
1644 }
1645 
1646 static bool mac_table_doesnt_fit(void *opaque, int version_id)
1647 {
1648     return !mac_table_fits(opaque, version_id);
1649 }
1650 
1651 /* This temporary type is shared by all the WITH_TMP methods
1652  * although only some fields are used by each.
1653  */
1654 struct VirtIONetMigTmp {
1655     VirtIONet      *parent;
1656     VirtIONetQueue *vqs_1;
1657     uint16_t        curr_queues_1;
1658     uint8_t         has_ufo;
1659     uint32_t        has_vnet_hdr;
1660 };
1661 
1662 /* The 2nd and subsequent tx_waiting flags are loaded later than
1663  * the 1st entry in the queues and only if there's more than one
1664  * entry.  We use the tmp mechanism to calculate a temporary
1665  * pointer and count and also validate the count.
1666  */
1667 
1668 static void virtio_net_tx_waiting_pre_save(void *opaque)
1669 {
1670     struct VirtIONetMigTmp *tmp = opaque;
1671 
1672     tmp->vqs_1 = tmp->parent->vqs + 1;
1673     tmp->curr_queues_1 = tmp->parent->curr_queues - 1;
1674     if (tmp->parent->curr_queues == 0) {
1675         tmp->curr_queues_1 = 0;
1676     }
1677 }
1678 
1679 static int virtio_net_tx_waiting_pre_load(void *opaque)
1680 {
1681     struct VirtIONetMigTmp *tmp = opaque;
1682 
1683     /* Reuse the pointer setup from save */
1684     virtio_net_tx_waiting_pre_save(opaque);
1685 
1686     if (tmp->parent->curr_queues > tmp->parent->max_queues) {
1687         error_report("virtio-net: curr_queues %x > max_queues %x",
1688             tmp->parent->curr_queues, tmp->parent->max_queues);
1689 
1690         return -EINVAL;
1691     }
1692 
1693     return 0; /* all good */
1694 }
1695 
1696 static const VMStateDescription vmstate_virtio_net_tx_waiting = {
1697     .name      = "virtio-net-tx_waiting",
1698     .pre_load  = virtio_net_tx_waiting_pre_load,
1699     .pre_save  = virtio_net_tx_waiting_pre_save,
1700     .fields    = (VMStateField[]) {
1701         VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp,
1702                                      curr_queues_1,
1703                                      vmstate_virtio_net_queue_tx_waiting,
1704                                      struct VirtIONetQueue),
1705         VMSTATE_END_OF_LIST()
1706     },
1707 };
1708 
1709 /* the 'has_ufo' flag is just tested; if the incoming stream has the
1710  * flag set we need to check that we have it
1711  */
1712 static int virtio_net_ufo_post_load(void *opaque, int version_id)
1713 {
1714     struct VirtIONetMigTmp *tmp = opaque;
1715 
1716     if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) {
1717         error_report("virtio-net: saved image requires TUN_F_UFO support");
1718         return -EINVAL;
1719     }
1720 
1721     return 0;
1722 }
1723 
1724 static void virtio_net_ufo_pre_save(void *opaque)
1725 {
1726     struct VirtIONetMigTmp *tmp = opaque;
1727 
1728     tmp->has_ufo = tmp->parent->has_ufo;
1729 }
1730 
1731 static const VMStateDescription vmstate_virtio_net_has_ufo = {
1732     .name      = "virtio-net-ufo",
1733     .post_load = virtio_net_ufo_post_load,
1734     .pre_save  = virtio_net_ufo_pre_save,
1735     .fields    = (VMStateField[]) {
1736         VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp),
1737         VMSTATE_END_OF_LIST()
1738     },
1739 };
1740 
1741 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
1742  * flag set we need to check that we have it
1743  */
1744 static int virtio_net_vnet_post_load(void *opaque, int version_id)
1745 {
1746     struct VirtIONetMigTmp *tmp = opaque;
1747 
1748     if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) {
1749         error_report("virtio-net: saved image requires vnet_hdr=on");
1750         return -EINVAL;
1751     }
1752 
1753     return 0;
1754 }
1755 
1756 static void virtio_net_vnet_pre_save(void *opaque)
1757 {
1758     struct VirtIONetMigTmp *tmp = opaque;
1759 
1760     tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr;
1761 }
1762 
1763 static const VMStateDescription vmstate_virtio_net_has_vnet = {
1764     .name      = "virtio-net-vnet",
1765     .post_load = virtio_net_vnet_post_load,
1766     .pre_save  = virtio_net_vnet_pre_save,
1767     .fields    = (VMStateField[]) {
1768         VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp),
1769         VMSTATE_END_OF_LIST()
1770     },
1771 };
1772 
1773 static const VMStateDescription vmstate_virtio_net_device = {
1774     .name = "virtio-net-device",
1775     .version_id = VIRTIO_NET_VM_VERSION,
1776     .minimum_version_id = VIRTIO_NET_VM_VERSION,
1777     .post_load = virtio_net_post_load_device,
1778     .fields = (VMStateField[]) {
1779         VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN),
1780         VMSTATE_STRUCT_POINTER(vqs, VirtIONet,
1781                                vmstate_virtio_net_queue_tx_waiting,
1782                                VirtIONetQueue),
1783         VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet),
1784         VMSTATE_UINT16(status, VirtIONet),
1785         VMSTATE_UINT8(promisc, VirtIONet),
1786         VMSTATE_UINT8(allmulti, VirtIONet),
1787         VMSTATE_UINT32(mac_table.in_use, VirtIONet),
1788 
1789         /* Guarded pair: If it fits we load it, else we throw it away
1790          * - can happen if source has a larger MAC table.; post-load
1791          *  sets flags in this case.
1792          */
1793         VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet,
1794                                 0, mac_table_fits, mac_table.in_use,
1795                                  ETH_ALEN),
1796         VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0,
1797                                      mac_table.in_use, ETH_ALEN),
1798 
1799         /* Note: This is an array of uint32's that's always been saved as a
1800          * buffer; hold onto your endiannesses; it's actually used as a bitmap
1801          * but based on the uint.
1802          */
1803         VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3),
1804         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
1805                          vmstate_virtio_net_has_vnet),
1806         VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet),
1807         VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet),
1808         VMSTATE_UINT8(alluni, VirtIONet),
1809         VMSTATE_UINT8(nomulti, VirtIONet),
1810         VMSTATE_UINT8(nouni, VirtIONet),
1811         VMSTATE_UINT8(nobcast, VirtIONet),
1812         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
1813                          vmstate_virtio_net_has_ufo),
1814         VMSTATE_SINGLE_TEST(max_queues, VirtIONet, max_queues_gt_1, 0,
1815                             vmstate_info_uint16_equal, uint16_t),
1816         VMSTATE_UINT16_TEST(curr_queues, VirtIONet, max_queues_gt_1),
1817         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
1818                          vmstate_virtio_net_tx_waiting),
1819         VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet,
1820                             has_ctrl_guest_offloads),
1821         VMSTATE_END_OF_LIST()
1822    },
1823 };
1824 
1825 static NetClientInfo net_virtio_info = {
1826     .type = NET_CLIENT_DRIVER_NIC,
1827     .size = sizeof(NICState),
1828     .can_receive = virtio_net_can_receive,
1829     .receive = virtio_net_receive,
1830     .link_status_changed = virtio_net_set_link_status,
1831     .query_rx_filter = virtio_net_query_rxfilter,
1832 };
1833 
1834 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1835 {
1836     VirtIONet *n = VIRTIO_NET(vdev);
1837     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1838     assert(n->vhost_started);
1839     return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
1840 }
1841 
1842 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1843                                            bool mask)
1844 {
1845     VirtIONet *n = VIRTIO_NET(vdev);
1846     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1847     assert(n->vhost_started);
1848     vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
1849                              vdev, idx, mask);
1850 }
1851 
1852 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
1853 {
1854     int i, config_size = 0;
1855     virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
1856 
1857     for (i = 0; feature_sizes[i].flags != 0; i++) {
1858         if (host_features & feature_sizes[i].flags) {
1859             config_size = MAX(feature_sizes[i].end, config_size);
1860         }
1861     }
1862     n->config_size = config_size;
1863 }
1864 
1865 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
1866                                    const char *type)
1867 {
1868     /*
1869      * The name can be NULL, the netclient name will be type.x.
1870      */
1871     assert(type != NULL);
1872 
1873     g_free(n->netclient_name);
1874     g_free(n->netclient_type);
1875     n->netclient_name = g_strdup(name);
1876     n->netclient_type = g_strdup(type);
1877 }
1878 
1879 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
1880 {
1881     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1882     VirtIONet *n = VIRTIO_NET(dev);
1883     NetClientState *nc;
1884     int i;
1885 
1886     if (n->net_conf.mtu) {
1887         n->host_features |= (0x1 << VIRTIO_NET_F_MTU);
1888     }
1889 
1890     virtio_net_set_config_size(n, n->host_features);
1891     virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
1892 
1893     /*
1894      * We set a lower limit on RX queue size to what it always was.
1895      * Guests that want a smaller ring can always resize it without
1896      * help from us (using virtio 1 and up).
1897      */
1898     if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
1899         n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
1900         (n->net_conf.rx_queue_size & (n->net_conf.rx_queue_size - 1))) {
1901         error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
1902                    "must be a power of 2 between %d and %d.",
1903                    n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
1904                    VIRTQUEUE_MAX_SIZE);
1905         virtio_cleanup(vdev);
1906         return;
1907     }
1908 
1909     n->max_queues = MAX(n->nic_conf.peers.queues, 1);
1910     if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) {
1911         error_setg(errp, "Invalid number of queues (= %" PRIu32 "), "
1912                    "must be a positive integer less than %d.",
1913                    n->max_queues, (VIRTIO_QUEUE_MAX - 1) / 2);
1914         virtio_cleanup(vdev);
1915         return;
1916     }
1917     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1918     n->curr_queues = 1;
1919     n->tx_timeout = n->net_conf.txtimer;
1920 
1921     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1922                        && strcmp(n->net_conf.tx, "bh")) {
1923         error_report("virtio-net: "
1924                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1925                      n->net_conf.tx);
1926         error_report("Defaulting to \"bh\"");
1927     }
1928 
1929     for (i = 0; i < n->max_queues; i++) {
1930         virtio_net_add_queue(n, i);
1931     }
1932 
1933     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1934     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1935     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1936     n->status = VIRTIO_NET_S_LINK_UP;
1937     n->announce_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
1938                                      virtio_net_announce_timer, n);
1939 
1940     if (n->netclient_type) {
1941         /*
1942          * Happen when virtio_net_set_netclient_name has been called.
1943          */
1944         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1945                               n->netclient_type, n->netclient_name, n);
1946     } else {
1947         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1948                               object_get_typename(OBJECT(dev)), dev->id, n);
1949     }
1950 
1951     peer_test_vnet_hdr(n);
1952     if (peer_has_vnet_hdr(n)) {
1953         for (i = 0; i < n->max_queues; i++) {
1954             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1955         }
1956         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1957     } else {
1958         n->host_hdr_len = 0;
1959     }
1960 
1961     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1962 
1963     n->vqs[0].tx_waiting = 0;
1964     n->tx_burst = n->net_conf.txburst;
1965     virtio_net_set_mrg_rx_bufs(n, 0, 0);
1966     n->promisc = 1; /* for compatibility */
1967 
1968     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1969 
1970     n->vlans = g_malloc0(MAX_VLAN >> 3);
1971 
1972     nc = qemu_get_queue(n->nic);
1973     nc->rxfilter_notify_enabled = 1;
1974 
1975     n->qdev = dev;
1976 }
1977 
1978 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
1979 {
1980     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1981     VirtIONet *n = VIRTIO_NET(dev);
1982     int i, max_queues;
1983 
1984     /* This will stop vhost backend if appropriate. */
1985     virtio_net_set_status(vdev, 0);
1986 
1987     g_free(n->netclient_name);
1988     n->netclient_name = NULL;
1989     g_free(n->netclient_type);
1990     n->netclient_type = NULL;
1991 
1992     g_free(n->mac_table.macs);
1993     g_free(n->vlans);
1994 
1995     max_queues = n->multiqueue ? n->max_queues : 1;
1996     for (i = 0; i < max_queues; i++) {
1997         virtio_net_del_queue(n, i);
1998     }
1999 
2000     timer_del(n->announce_timer);
2001     timer_free(n->announce_timer);
2002     g_free(n->vqs);
2003     qemu_del_nic(n->nic);
2004     virtio_cleanup(vdev);
2005 }
2006 
2007 static void virtio_net_instance_init(Object *obj)
2008 {
2009     VirtIONet *n = VIRTIO_NET(obj);
2010 
2011     /*
2012      * The default config_size is sizeof(struct virtio_net_config).
2013      * Can be overriden with virtio_net_set_config_size.
2014      */
2015     n->config_size = sizeof(struct virtio_net_config);
2016     device_add_bootindex_property(obj, &n->nic_conf.bootindex,
2017                                   "bootindex", "/ethernet-phy@0",
2018                                   DEVICE(n), NULL);
2019 }
2020 
2021 static void virtio_net_pre_save(void *opaque)
2022 {
2023     VirtIONet *n = opaque;
2024 
2025     /* At this point, backend must be stopped, otherwise
2026      * it might keep writing to memory. */
2027     assert(!n->vhost_started);
2028 }
2029 
2030 static const VMStateDescription vmstate_virtio_net = {
2031     .name = "virtio-net",
2032     .minimum_version_id = VIRTIO_NET_VM_VERSION,
2033     .version_id = VIRTIO_NET_VM_VERSION,
2034     .fields = (VMStateField[]) {
2035         VMSTATE_VIRTIO_DEVICE,
2036         VMSTATE_END_OF_LIST()
2037     },
2038     .pre_save = virtio_net_pre_save,
2039 };
2040 
2041 static Property virtio_net_properties[] = {
2042     DEFINE_PROP_BIT("csum", VirtIONet, host_features, VIRTIO_NET_F_CSUM, true),
2043     DEFINE_PROP_BIT("guest_csum", VirtIONet, host_features,
2044                     VIRTIO_NET_F_GUEST_CSUM, true),
2045     DEFINE_PROP_BIT("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
2046     DEFINE_PROP_BIT("guest_tso4", VirtIONet, host_features,
2047                     VIRTIO_NET_F_GUEST_TSO4, true),
2048     DEFINE_PROP_BIT("guest_tso6", VirtIONet, host_features,
2049                     VIRTIO_NET_F_GUEST_TSO6, true),
2050     DEFINE_PROP_BIT("guest_ecn", VirtIONet, host_features,
2051                     VIRTIO_NET_F_GUEST_ECN, true),
2052     DEFINE_PROP_BIT("guest_ufo", VirtIONet, host_features,
2053                     VIRTIO_NET_F_GUEST_UFO, true),
2054     DEFINE_PROP_BIT("guest_announce", VirtIONet, host_features,
2055                     VIRTIO_NET_F_GUEST_ANNOUNCE, true),
2056     DEFINE_PROP_BIT("host_tso4", VirtIONet, host_features,
2057                     VIRTIO_NET_F_HOST_TSO4, true),
2058     DEFINE_PROP_BIT("host_tso6", VirtIONet, host_features,
2059                     VIRTIO_NET_F_HOST_TSO6, true),
2060     DEFINE_PROP_BIT("host_ecn", VirtIONet, host_features,
2061                     VIRTIO_NET_F_HOST_ECN, true),
2062     DEFINE_PROP_BIT("host_ufo", VirtIONet, host_features,
2063                     VIRTIO_NET_F_HOST_UFO, true),
2064     DEFINE_PROP_BIT("mrg_rxbuf", VirtIONet, host_features,
2065                     VIRTIO_NET_F_MRG_RXBUF, true),
2066     DEFINE_PROP_BIT("status", VirtIONet, host_features,
2067                     VIRTIO_NET_F_STATUS, true),
2068     DEFINE_PROP_BIT("ctrl_vq", VirtIONet, host_features,
2069                     VIRTIO_NET_F_CTRL_VQ, true),
2070     DEFINE_PROP_BIT("ctrl_rx", VirtIONet, host_features,
2071                     VIRTIO_NET_F_CTRL_RX, true),
2072     DEFINE_PROP_BIT("ctrl_vlan", VirtIONet, host_features,
2073                     VIRTIO_NET_F_CTRL_VLAN, true),
2074     DEFINE_PROP_BIT("ctrl_rx_extra", VirtIONet, host_features,
2075                     VIRTIO_NET_F_CTRL_RX_EXTRA, true),
2076     DEFINE_PROP_BIT("ctrl_mac_addr", VirtIONet, host_features,
2077                     VIRTIO_NET_F_CTRL_MAC_ADDR, true),
2078     DEFINE_PROP_BIT("ctrl_guest_offloads", VirtIONet, host_features,
2079                     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
2080     DEFINE_PROP_BIT("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
2081     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
2082     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
2083                        TX_TIMER_INTERVAL),
2084     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
2085     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
2086     DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
2087                        VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
2088     DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
2089     DEFINE_PROP_END_OF_LIST(),
2090 };
2091 
2092 static void virtio_net_class_init(ObjectClass *klass, void *data)
2093 {
2094     DeviceClass *dc = DEVICE_CLASS(klass);
2095     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
2096 
2097     dc->props = virtio_net_properties;
2098     dc->vmsd = &vmstate_virtio_net;
2099     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
2100     vdc->realize = virtio_net_device_realize;
2101     vdc->unrealize = virtio_net_device_unrealize;
2102     vdc->get_config = virtio_net_get_config;
2103     vdc->set_config = virtio_net_set_config;
2104     vdc->get_features = virtio_net_get_features;
2105     vdc->set_features = virtio_net_set_features;
2106     vdc->bad_features = virtio_net_bad_features;
2107     vdc->reset = virtio_net_reset;
2108     vdc->set_status = virtio_net_set_status;
2109     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
2110     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
2111     vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
2112     vdc->vmsd = &vmstate_virtio_net_device;
2113 }
2114 
2115 static const TypeInfo virtio_net_info = {
2116     .name = TYPE_VIRTIO_NET,
2117     .parent = TYPE_VIRTIO_DEVICE,
2118     .instance_size = sizeof(VirtIONet),
2119     .instance_init = virtio_net_instance_init,
2120     .class_init = virtio_net_class_init,
2121 };
2122 
2123 static void virtio_register_types(void)
2124 {
2125     type_register_static(&virtio_net_info);
2126 }
2127 
2128 type_init(virtio_register_types)
2129