xref: /openbmc/qemu/hw/net/virtio-net.c (revision 00e7b1299599384dfdda2a2a4570a0fb2d69eb6b)
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/atomic.h"
16 #include "qemu/iov.h"
17 #include "qemu/main-loop.h"
18 #include "qemu/module.h"
19 #include "hw/virtio/virtio.h"
20 #include "net/net.h"
21 #include "net/checksum.h"
22 #include "net/tap.h"
23 #include "qemu/error-report.h"
24 #include "qemu/timer.h"
25 #include "qemu/option.h"
26 #include "qemu/option_int.h"
27 #include "qemu/config-file.h"
28 #include "qapi/qmp/qdict.h"
29 #include "hw/virtio/virtio-net.h"
30 #include "net/vhost_net.h"
31 #include "net/announce.h"
32 #include "hw/virtio/virtio-bus.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-events-net.h"
35 #include "hw/qdev-properties.h"
36 #include "qapi/qapi-types-migration.h"
37 #include "qapi/qapi-events-migration.h"
38 #include "hw/virtio/virtio-access.h"
39 #include "migration/misc.h"
40 #include "standard-headers/linux/ethtool.h"
41 #include "sysemu/sysemu.h"
42 #include "trace.h"
43 #include "monitor/qdev.h"
44 #include "hw/pci/pci.h"
45 #include "net_rx_pkt.h"
46 #include "hw/virtio/vhost.h"
47 
48 #define VIRTIO_NET_VM_VERSION    11
49 
50 #define MAC_TABLE_ENTRIES    64
51 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
52 
53 /* previously fixed value */
54 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
55 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
56 
57 /* for now, only allow larger queues; with virtio-1, guest can downsize */
58 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
59 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
60 
61 #define VIRTIO_NET_IP4_ADDR_SIZE   8        /* ipv4 saddr + daddr */
62 
63 #define VIRTIO_NET_TCP_FLAG         0x3F
64 #define VIRTIO_NET_TCP_HDR_LENGTH   0xF000
65 
66 /* IPv4 max payload, 16 bits in the header */
67 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
68 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
69 
70 /* header length value in ip header without option */
71 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
72 
73 #define VIRTIO_NET_IP6_ADDR_SIZE   32      /* ipv6 saddr + daddr */
74 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
75 
76 /* Purge coalesced packets timer interval, This value affects the performance
77    a lot, and should be tuned carefully, '300000'(300us) is the recommended
78    value to pass the WHQL test, '50000' can gain 2x netperf throughput with
79    tso/gso/gro 'off'. */
80 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
81 
82 #define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \
83                                          VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \
84                                          VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \
85                                          VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \
86                                          VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \
87                                          VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \
88                                          VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \
89                                          VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \
90                                          VIRTIO_NET_RSS_HASH_TYPE_UDP_EX)
91 
92 static VirtIOFeature feature_sizes[] = {
93     {.flags = 1ULL << VIRTIO_NET_F_MAC,
94      .end = endof(struct virtio_net_config, mac)},
95     {.flags = 1ULL << VIRTIO_NET_F_STATUS,
96      .end = endof(struct virtio_net_config, status)},
97     {.flags = 1ULL << VIRTIO_NET_F_MQ,
98      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
99     {.flags = 1ULL << VIRTIO_NET_F_MTU,
100      .end = endof(struct virtio_net_config, mtu)},
101     {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX,
102      .end = endof(struct virtio_net_config, duplex)},
103     {.flags = (1ULL << VIRTIO_NET_F_RSS) | (1ULL << VIRTIO_NET_F_HASH_REPORT),
104      .end = endof(struct virtio_net_config, supported_hash_types)},
105     {}
106 };
107 
108 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
109 {
110     VirtIONet *n = qemu_get_nic_opaque(nc);
111 
112     return &n->vqs[nc->queue_index];
113 }
114 
115 static int vq2q(int queue_index)
116 {
117     return queue_index / 2;
118 }
119 
120 /* TODO
121  * - we could suppress RX interrupt if we were so inclined.
122  */
123 
124 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
125 {
126     VirtIONet *n = VIRTIO_NET(vdev);
127     struct virtio_net_config netcfg;
128     NetClientState *nc = qemu_get_queue(n->nic);
129 
130     int ret = 0;
131     memset(&netcfg, 0 , sizeof(struct virtio_net_config));
132     virtio_stw_p(vdev, &netcfg.status, n->status);
133     virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues);
134     virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
135     memcpy(netcfg.mac, n->mac, ETH_ALEN);
136     virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed);
137     netcfg.duplex = n->net_conf.duplex;
138     netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE;
139     virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length,
140                  virtio_host_has_feature(vdev, VIRTIO_NET_F_RSS) ?
141                  VIRTIO_NET_RSS_MAX_TABLE_LEN : 1);
142     virtio_stl_p(vdev, &netcfg.supported_hash_types,
143                  VIRTIO_NET_RSS_SUPPORTED_HASHES);
144     memcpy(config, &netcfg, n->config_size);
145 
146     /*
147      * Is this VDPA? No peer means not VDPA: there's no way to
148      * disconnect/reconnect a VDPA peer.
149      */
150     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
151         ret = vhost_net_get_config(get_vhost_net(nc->peer), (uint8_t *)&netcfg,
152                                    n->config_size);
153         if (ret != -1) {
154             memcpy(config, &netcfg, n->config_size);
155         }
156     }
157 }
158 
159 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
160 {
161     VirtIONet *n = VIRTIO_NET(vdev);
162     struct virtio_net_config netcfg = {};
163     NetClientState *nc = qemu_get_queue(n->nic);
164 
165     memcpy(&netcfg, config, n->config_size);
166 
167     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
168         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
169         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
170         memcpy(n->mac, netcfg.mac, ETH_ALEN);
171         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
172     }
173 
174     /*
175      * Is this VDPA? No peer means not VDPA: there's no way to
176      * disconnect/reconnect a VDPA peer.
177      */
178     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
179         vhost_net_set_config(get_vhost_net(nc->peer),
180                              (uint8_t *)&netcfg, 0, n->config_size,
181                              VHOST_SET_CONFIG_TYPE_MASTER);
182       }
183 }
184 
185 static bool virtio_net_started(VirtIONet *n, uint8_t status)
186 {
187     VirtIODevice *vdev = VIRTIO_DEVICE(n);
188     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
189         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
190 }
191 
192 static void virtio_net_announce_notify(VirtIONet *net)
193 {
194     VirtIODevice *vdev = VIRTIO_DEVICE(net);
195     trace_virtio_net_announce_notify();
196 
197     net->status |= VIRTIO_NET_S_ANNOUNCE;
198     virtio_notify_config(vdev);
199 }
200 
201 static void virtio_net_announce_timer(void *opaque)
202 {
203     VirtIONet *n = opaque;
204     trace_virtio_net_announce_timer(n->announce_timer.round);
205 
206     n->announce_timer.round--;
207     virtio_net_announce_notify(n);
208 }
209 
210 static void virtio_net_announce(NetClientState *nc)
211 {
212     VirtIONet *n = qemu_get_nic_opaque(nc);
213     VirtIODevice *vdev = VIRTIO_DEVICE(n);
214 
215     /*
216      * Make sure the virtio migration announcement timer isn't running
217      * If it is, let it trigger announcement so that we do not cause
218      * confusion.
219      */
220     if (n->announce_timer.round) {
221         return;
222     }
223 
224     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
225         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
226             virtio_net_announce_notify(n);
227     }
228 }
229 
230 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
231 {
232     VirtIODevice *vdev = VIRTIO_DEVICE(n);
233     NetClientState *nc = qemu_get_queue(n->nic);
234     int queues = n->multiqueue ? n->max_queues : 1;
235 
236     if (!get_vhost_net(nc->peer)) {
237         return;
238     }
239 
240     if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
241         !!n->vhost_started) {
242         return;
243     }
244     if (!n->vhost_started) {
245         int r, i;
246 
247         if (n->needs_vnet_hdr_swap) {
248             error_report("backend does not support %s vnet headers; "
249                          "falling back on userspace virtio",
250                          virtio_is_big_endian(vdev) ? "BE" : "LE");
251             return;
252         }
253 
254         /* Any packets outstanding? Purge them to avoid touching rings
255          * when vhost is running.
256          */
257         for (i = 0;  i < queues; i++) {
258             NetClientState *qnc = qemu_get_subqueue(n->nic, i);
259 
260             /* Purge both directions: TX and RX. */
261             qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
262             qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
263         }
264 
265         if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
266             r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
267             if (r < 0) {
268                 error_report("%uBytes MTU not supported by the backend",
269                              n->net_conf.mtu);
270 
271                 return;
272             }
273         }
274 
275         n->vhost_started = 1;
276         r = vhost_net_start(vdev, n->nic->ncs, queues);
277         if (r < 0) {
278             error_report("unable to start vhost net: %d: "
279                          "falling back on userspace virtio", -r);
280             n->vhost_started = 0;
281         }
282     } else {
283         vhost_net_stop(vdev, n->nic->ncs, queues);
284         n->vhost_started = 0;
285     }
286 }
287 
288 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
289                                           NetClientState *peer,
290                                           bool enable)
291 {
292     if (virtio_is_big_endian(vdev)) {
293         return qemu_set_vnet_be(peer, enable);
294     } else {
295         return qemu_set_vnet_le(peer, enable);
296     }
297 }
298 
299 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
300                                        int queues, bool enable)
301 {
302     int i;
303 
304     for (i = 0; i < queues; i++) {
305         if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
306             enable) {
307             while (--i >= 0) {
308                 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
309             }
310 
311             return true;
312         }
313     }
314 
315     return false;
316 }
317 
318 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
319 {
320     VirtIODevice *vdev = VIRTIO_DEVICE(n);
321     int queues = n->multiqueue ? n->max_queues : 1;
322 
323     if (virtio_net_started(n, status)) {
324         /* Before using the device, we tell the network backend about the
325          * endianness to use when parsing vnet headers. If the backend
326          * can't do it, we fallback onto fixing the headers in the core
327          * virtio-net code.
328          */
329         n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
330                                                             queues, true);
331     } else if (virtio_net_started(n, vdev->status)) {
332         /* After using the device, we need to reset the network backend to
333          * the default (guest native endianness), otherwise the guest may
334          * lose network connectivity if it is rebooted into a different
335          * endianness.
336          */
337         virtio_net_set_vnet_endian(vdev, n->nic->ncs, queues, false);
338     }
339 }
340 
341 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq)
342 {
343     unsigned int dropped = virtqueue_drop_all(vq);
344     if (dropped) {
345         virtio_notify(vdev, vq);
346     }
347 }
348 
349 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
350 {
351     VirtIONet *n = VIRTIO_NET(vdev);
352     VirtIONetQueue *q;
353     int i;
354     uint8_t queue_status;
355 
356     virtio_net_vnet_endian_status(n, status);
357     virtio_net_vhost_status(n, status);
358 
359     for (i = 0; i < n->max_queues; i++) {
360         NetClientState *ncs = qemu_get_subqueue(n->nic, i);
361         bool queue_started;
362         q = &n->vqs[i];
363 
364         if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
365             queue_status = 0;
366         } else {
367             queue_status = status;
368         }
369         queue_started =
370             virtio_net_started(n, queue_status) && !n->vhost_started;
371 
372         if (queue_started) {
373             qemu_flush_queued_packets(ncs);
374         }
375 
376         if (!q->tx_waiting) {
377             continue;
378         }
379 
380         if (queue_started) {
381             if (q->tx_timer) {
382                 timer_mod(q->tx_timer,
383                                qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
384             } else {
385                 qemu_bh_schedule(q->tx_bh);
386             }
387         } else {
388             if (q->tx_timer) {
389                 timer_del(q->tx_timer);
390             } else {
391                 qemu_bh_cancel(q->tx_bh);
392             }
393             if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 &&
394                 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) &&
395                 vdev->vm_running) {
396                 /* if tx is waiting we are likely have some packets in tx queue
397                  * and disabled notification */
398                 q->tx_waiting = 0;
399                 virtio_queue_set_notification(q->tx_vq, 1);
400                 virtio_net_drop_tx_queue_data(vdev, q->tx_vq);
401             }
402         }
403     }
404 }
405 
406 static void virtio_net_set_link_status(NetClientState *nc)
407 {
408     VirtIONet *n = qemu_get_nic_opaque(nc);
409     VirtIODevice *vdev = VIRTIO_DEVICE(n);
410     uint16_t old_status = n->status;
411 
412     if (nc->link_down)
413         n->status &= ~VIRTIO_NET_S_LINK_UP;
414     else
415         n->status |= VIRTIO_NET_S_LINK_UP;
416 
417     if (n->status != old_status)
418         virtio_notify_config(vdev);
419 
420     virtio_net_set_status(vdev, vdev->status);
421 }
422 
423 static void rxfilter_notify(NetClientState *nc)
424 {
425     VirtIONet *n = qemu_get_nic_opaque(nc);
426 
427     if (nc->rxfilter_notify_enabled) {
428         char *path = object_get_canonical_path(OBJECT(n->qdev));
429         qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
430                                               n->netclient_name, path);
431         g_free(path);
432 
433         /* disable event notification to avoid events flooding */
434         nc->rxfilter_notify_enabled = 0;
435     }
436 }
437 
438 static intList *get_vlan_table(VirtIONet *n)
439 {
440     intList *list;
441     int i, j;
442 
443     list = NULL;
444     for (i = 0; i < MAX_VLAN >> 5; i++) {
445         for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
446             if (n->vlans[i] & (1U << j)) {
447                 QAPI_LIST_PREPEND(list, (i << 5) + j);
448             }
449         }
450     }
451 
452     return list;
453 }
454 
455 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
456 {
457     VirtIONet *n = qemu_get_nic_opaque(nc);
458     VirtIODevice *vdev = VIRTIO_DEVICE(n);
459     RxFilterInfo *info;
460     strList *str_list;
461     int i;
462 
463     info = g_malloc0(sizeof(*info));
464     info->name = g_strdup(nc->name);
465     info->promiscuous = n->promisc;
466 
467     if (n->nouni) {
468         info->unicast = RX_STATE_NONE;
469     } else if (n->alluni) {
470         info->unicast = RX_STATE_ALL;
471     } else {
472         info->unicast = RX_STATE_NORMAL;
473     }
474 
475     if (n->nomulti) {
476         info->multicast = RX_STATE_NONE;
477     } else if (n->allmulti) {
478         info->multicast = RX_STATE_ALL;
479     } else {
480         info->multicast = RX_STATE_NORMAL;
481     }
482 
483     info->broadcast_allowed = n->nobcast;
484     info->multicast_overflow = n->mac_table.multi_overflow;
485     info->unicast_overflow = n->mac_table.uni_overflow;
486 
487     info->main_mac = qemu_mac_strdup_printf(n->mac);
488 
489     str_list = NULL;
490     for (i = 0; i < n->mac_table.first_multi; i++) {
491         QAPI_LIST_PREPEND(str_list,
492                       qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN));
493     }
494     info->unicast_table = str_list;
495 
496     str_list = NULL;
497     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
498         QAPI_LIST_PREPEND(str_list,
499                       qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN));
500     }
501     info->multicast_table = str_list;
502     info->vlan_table = get_vlan_table(n);
503 
504     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
505         info->vlan = RX_STATE_ALL;
506     } else if (!info->vlan_table) {
507         info->vlan = RX_STATE_NONE;
508     } else {
509         info->vlan = RX_STATE_NORMAL;
510     }
511 
512     /* enable event notification after query */
513     nc->rxfilter_notify_enabled = 1;
514 
515     return info;
516 }
517 
518 static void virtio_net_reset(VirtIODevice *vdev)
519 {
520     VirtIONet *n = VIRTIO_NET(vdev);
521     int i;
522 
523     /* Reset back to compatibility mode */
524     n->promisc = 1;
525     n->allmulti = 0;
526     n->alluni = 0;
527     n->nomulti = 0;
528     n->nouni = 0;
529     n->nobcast = 0;
530     /* multiqueue is disabled by default */
531     n->curr_queues = 1;
532     timer_del(n->announce_timer.tm);
533     n->announce_timer.round = 0;
534     n->status &= ~VIRTIO_NET_S_ANNOUNCE;
535 
536     /* Flush any MAC and VLAN filter table state */
537     n->mac_table.in_use = 0;
538     n->mac_table.first_multi = 0;
539     n->mac_table.multi_overflow = 0;
540     n->mac_table.uni_overflow = 0;
541     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
542     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
543     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
544     memset(n->vlans, 0, MAX_VLAN >> 3);
545 
546     /* Flush any async TX */
547     for (i = 0;  i < n->max_queues; i++) {
548         NetClientState *nc = qemu_get_subqueue(n->nic, i);
549 
550         if (nc->peer) {
551             qemu_flush_or_purge_queued_packets(nc->peer, true);
552             assert(!virtio_net_get_subqueue(nc)->async_tx.elem);
553         }
554     }
555 }
556 
557 static void peer_test_vnet_hdr(VirtIONet *n)
558 {
559     NetClientState *nc = qemu_get_queue(n->nic);
560     if (!nc->peer) {
561         return;
562     }
563 
564     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
565 }
566 
567 static int peer_has_vnet_hdr(VirtIONet *n)
568 {
569     return n->has_vnet_hdr;
570 }
571 
572 static int peer_has_ufo(VirtIONet *n)
573 {
574     if (!peer_has_vnet_hdr(n))
575         return 0;
576 
577     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
578 
579     return n->has_ufo;
580 }
581 
582 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
583                                        int version_1, int hash_report)
584 {
585     int i;
586     NetClientState *nc;
587 
588     n->mergeable_rx_bufs = mergeable_rx_bufs;
589 
590     if (version_1) {
591         n->guest_hdr_len = hash_report ?
592             sizeof(struct virtio_net_hdr_v1_hash) :
593             sizeof(struct virtio_net_hdr_mrg_rxbuf);
594         n->rss_data.populate_hash = !!hash_report;
595     } else {
596         n->guest_hdr_len = n->mergeable_rx_bufs ?
597             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
598             sizeof(struct virtio_net_hdr);
599     }
600 
601     for (i = 0; i < n->max_queues; i++) {
602         nc = qemu_get_subqueue(n->nic, i);
603 
604         if (peer_has_vnet_hdr(n) &&
605             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
606             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
607             n->host_hdr_len = n->guest_hdr_len;
608         }
609     }
610 }
611 
612 static int virtio_net_max_tx_queue_size(VirtIONet *n)
613 {
614     NetClientState *peer = n->nic_conf.peers.ncs[0];
615 
616     /*
617      * Backends other than vhost-user don't support max queue size.
618      */
619     if (!peer) {
620         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
621     }
622 
623     if (peer->info->type != NET_CLIENT_DRIVER_VHOST_USER) {
624         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
625     }
626 
627     return VIRTQUEUE_MAX_SIZE;
628 }
629 
630 static int peer_attach(VirtIONet *n, int index)
631 {
632     NetClientState *nc = qemu_get_subqueue(n->nic, index);
633 
634     if (!nc->peer) {
635         return 0;
636     }
637 
638     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
639         vhost_set_vring_enable(nc->peer, 1);
640     }
641 
642     if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
643         return 0;
644     }
645 
646     if (n->max_queues == 1) {
647         return 0;
648     }
649 
650     return tap_enable(nc->peer);
651 }
652 
653 static int peer_detach(VirtIONet *n, int index)
654 {
655     NetClientState *nc = qemu_get_subqueue(n->nic, index);
656 
657     if (!nc->peer) {
658         return 0;
659     }
660 
661     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
662         vhost_set_vring_enable(nc->peer, 0);
663     }
664 
665     if (nc->peer->info->type !=  NET_CLIENT_DRIVER_TAP) {
666         return 0;
667     }
668 
669     return tap_disable(nc->peer);
670 }
671 
672 static void virtio_net_set_queues(VirtIONet *n)
673 {
674     int i;
675     int r;
676 
677     if (n->nic->peer_deleted) {
678         return;
679     }
680 
681     for (i = 0; i < n->max_queues; i++) {
682         if (i < n->curr_queues) {
683             r = peer_attach(n, i);
684             assert(!r);
685         } else {
686             r = peer_detach(n, i);
687             assert(!r);
688         }
689     }
690 }
691 
692 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
693 
694 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
695                                         Error **errp)
696 {
697     VirtIONet *n = VIRTIO_NET(vdev);
698     NetClientState *nc = qemu_get_queue(n->nic);
699 
700     /* Firstly sync all virtio-net possible supported features */
701     features |= n->host_features;
702 
703     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
704 
705     if (!peer_has_vnet_hdr(n)) {
706         virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
707         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
708         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
709         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
710 
711         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
712         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
713         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
714         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
715 
716         virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT);
717     }
718 
719     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
720         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
721         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
722     }
723 
724     if (!get_vhost_net(nc->peer)) {
725         return features;
726     }
727 
728     virtio_clear_feature(&features, VIRTIO_NET_F_RSS);
729     virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT);
730     features = vhost_net_get_features(get_vhost_net(nc->peer), features);
731     vdev->backend_features = features;
732 
733     if (n->mtu_bypass_backend &&
734             (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) {
735         features |= (1ULL << VIRTIO_NET_F_MTU);
736     }
737 
738     return features;
739 }
740 
741 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
742 {
743     uint64_t features = 0;
744 
745     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
746      * but also these: */
747     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
748     virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
749     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
750     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
751     virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
752 
753     return features;
754 }
755 
756 static void virtio_net_apply_guest_offloads(VirtIONet *n)
757 {
758     qemu_set_offload(qemu_get_queue(n->nic)->peer,
759             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
760             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
761             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
762             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
763             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
764 }
765 
766 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
767 {
768     static const uint64_t guest_offloads_mask =
769         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
770         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
771         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
772         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
773         (1ULL << VIRTIO_NET_F_GUEST_UFO);
774 
775     return guest_offloads_mask & features;
776 }
777 
778 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
779 {
780     VirtIODevice *vdev = VIRTIO_DEVICE(n);
781     return virtio_net_guest_offloads_by_features(vdev->guest_features);
782 }
783 
784 typedef struct {
785     VirtIONet *n;
786     char *id;
787 } FailoverId;
788 
789 /**
790  * Set the id of the failover primary device
791  *
792  * @opaque: FailoverId to setup
793  * @opts: opts for device we are handling
794  * @errp: returns an error if this function fails
795  */
796 static int failover_set_primary(void *opaque, QemuOpts *opts, Error **errp)
797 {
798     FailoverId *fid = opaque;
799     const char *standby_id = qemu_opt_get(opts, "failover_pair_id");
800 
801     if (g_strcmp0(standby_id, fid->n->netclient_name) == 0) {
802         fid->id = g_strdup(opts->id);
803         return 1;
804     }
805 
806     return 0;
807 }
808 
809 /**
810  * Find the primary device id for this failover virtio-net
811  *
812  * @n: VirtIONet device
813  * @errp: returns an error if this function fails
814  */
815 static char *failover_find_primary_device_id(VirtIONet *n)
816 {
817     Error *err = NULL;
818     FailoverId fid;
819 
820     fid.n = n;
821     if (!qemu_opts_foreach(qemu_find_opts("device"),
822                            failover_set_primary, &fid, &err)) {
823         return NULL;
824     }
825     return fid.id;
826 }
827 
828 /**
829  * Find the primary device for this failover virtio-net
830  *
831  * @n: VirtIONet device
832  * @errp: returns an error if this function fails
833  */
834 static DeviceState *failover_find_primary_device(VirtIONet *n)
835 {
836     char *id = failover_find_primary_device_id(n);
837 
838     if (!id) {
839         return NULL;
840     }
841 
842     return qdev_find_recursive(sysbus_get_default(), id);
843 }
844 
845 static void failover_add_primary(VirtIONet *n, Error **errp)
846 {
847     Error *err = NULL;
848     QemuOpts *opts;
849     char *id;
850     DeviceState *dev = failover_find_primary_device(n);
851 
852     if (dev) {
853         return;
854     }
855 
856     id = failover_find_primary_device_id(n);
857     if (!id) {
858         return;
859     }
860     opts = qemu_opts_find(qemu_find_opts("device"), id);
861     if (opts) {
862         dev = qdev_device_add(opts, &err);
863         if (err) {
864             qemu_opts_del(opts);
865         } else {
866             object_unref(OBJECT(dev));
867         }
868     } else {
869         error_setg(errp, "Primary device not found");
870         error_append_hint(errp, "Virtio-net failover will not work. Make "
871                           "sure primary device has parameter"
872                           " failover_pair_id=<virtio-net-id>\n");
873     }
874     error_propagate(errp, err);
875 }
876 
877 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
878 {
879     VirtIONet *n = VIRTIO_NET(vdev);
880     Error *err = NULL;
881     int i;
882 
883     if (n->mtu_bypass_backend &&
884             !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) {
885         features &= ~(1ULL << VIRTIO_NET_F_MTU);
886     }
887 
888     virtio_net_set_multiqueue(n,
889                               virtio_has_feature(features, VIRTIO_NET_F_RSS) ||
890                               virtio_has_feature(features, VIRTIO_NET_F_MQ));
891 
892     virtio_net_set_mrg_rx_bufs(n,
893                                virtio_has_feature(features,
894                                                   VIRTIO_NET_F_MRG_RXBUF),
895                                virtio_has_feature(features,
896                                                   VIRTIO_F_VERSION_1),
897                                virtio_has_feature(features,
898                                                   VIRTIO_NET_F_HASH_REPORT));
899 
900     n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
901         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4);
902     n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
903         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6);
904     n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS);
905 
906     if (n->has_vnet_hdr) {
907         n->curr_guest_offloads =
908             virtio_net_guest_offloads_by_features(features);
909         virtio_net_apply_guest_offloads(n);
910     }
911 
912     for (i = 0;  i < n->max_queues; i++) {
913         NetClientState *nc = qemu_get_subqueue(n->nic, i);
914 
915         if (!get_vhost_net(nc->peer)) {
916             continue;
917         }
918         vhost_net_ack_features(get_vhost_net(nc->peer), features);
919     }
920 
921     if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
922         memset(n->vlans, 0, MAX_VLAN >> 3);
923     } else {
924         memset(n->vlans, 0xff, MAX_VLAN >> 3);
925     }
926 
927     if (virtio_has_feature(features, VIRTIO_NET_F_STANDBY)) {
928         qapi_event_send_failover_negotiated(n->netclient_name);
929         qatomic_set(&n->failover_primary_hidden, false);
930         failover_add_primary(n, &err);
931         if (err) {
932             warn_report_err(err);
933         }
934     }
935 }
936 
937 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
938                                      struct iovec *iov, unsigned int iov_cnt)
939 {
940     uint8_t on;
941     size_t s;
942     NetClientState *nc = qemu_get_queue(n->nic);
943 
944     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
945     if (s != sizeof(on)) {
946         return VIRTIO_NET_ERR;
947     }
948 
949     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
950         n->promisc = on;
951     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
952         n->allmulti = on;
953     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
954         n->alluni = on;
955     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
956         n->nomulti = on;
957     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
958         n->nouni = on;
959     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
960         n->nobcast = on;
961     } else {
962         return VIRTIO_NET_ERR;
963     }
964 
965     rxfilter_notify(nc);
966 
967     return VIRTIO_NET_OK;
968 }
969 
970 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
971                                      struct iovec *iov, unsigned int iov_cnt)
972 {
973     VirtIODevice *vdev = VIRTIO_DEVICE(n);
974     uint64_t offloads;
975     size_t s;
976 
977     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
978         return VIRTIO_NET_ERR;
979     }
980 
981     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
982     if (s != sizeof(offloads)) {
983         return VIRTIO_NET_ERR;
984     }
985 
986     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
987         uint64_t supported_offloads;
988 
989         offloads = virtio_ldq_p(vdev, &offloads);
990 
991         if (!n->has_vnet_hdr) {
992             return VIRTIO_NET_ERR;
993         }
994 
995         n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
996             virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4);
997         n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
998             virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6);
999         virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT);
1000 
1001         supported_offloads = virtio_net_supported_guest_offloads(n);
1002         if (offloads & ~supported_offloads) {
1003             return VIRTIO_NET_ERR;
1004         }
1005 
1006         n->curr_guest_offloads = offloads;
1007         virtio_net_apply_guest_offloads(n);
1008 
1009         return VIRTIO_NET_OK;
1010     } else {
1011         return VIRTIO_NET_ERR;
1012     }
1013 }
1014 
1015 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
1016                                  struct iovec *iov, unsigned int iov_cnt)
1017 {
1018     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1019     struct virtio_net_ctrl_mac mac_data;
1020     size_t s;
1021     NetClientState *nc = qemu_get_queue(n->nic);
1022 
1023     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
1024         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
1025             return VIRTIO_NET_ERR;
1026         }
1027         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
1028         assert(s == sizeof(n->mac));
1029         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
1030         rxfilter_notify(nc);
1031 
1032         return VIRTIO_NET_OK;
1033     }
1034 
1035     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
1036         return VIRTIO_NET_ERR;
1037     }
1038 
1039     int in_use = 0;
1040     int first_multi = 0;
1041     uint8_t uni_overflow = 0;
1042     uint8_t multi_overflow = 0;
1043     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1044 
1045     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1046                    sizeof(mac_data.entries));
1047     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1048     if (s != sizeof(mac_data.entries)) {
1049         goto error;
1050     }
1051     iov_discard_front(&iov, &iov_cnt, s);
1052 
1053     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
1054         goto error;
1055     }
1056 
1057     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
1058         s = iov_to_buf(iov, iov_cnt, 0, macs,
1059                        mac_data.entries * ETH_ALEN);
1060         if (s != mac_data.entries * ETH_ALEN) {
1061             goto error;
1062         }
1063         in_use += mac_data.entries;
1064     } else {
1065         uni_overflow = 1;
1066     }
1067 
1068     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
1069 
1070     first_multi = in_use;
1071 
1072     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1073                    sizeof(mac_data.entries));
1074     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1075     if (s != sizeof(mac_data.entries)) {
1076         goto error;
1077     }
1078 
1079     iov_discard_front(&iov, &iov_cnt, s);
1080 
1081     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
1082         goto error;
1083     }
1084 
1085     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
1086         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
1087                        mac_data.entries * ETH_ALEN);
1088         if (s != mac_data.entries * ETH_ALEN) {
1089             goto error;
1090         }
1091         in_use += mac_data.entries;
1092     } else {
1093         multi_overflow = 1;
1094     }
1095 
1096     n->mac_table.in_use = in_use;
1097     n->mac_table.first_multi = first_multi;
1098     n->mac_table.uni_overflow = uni_overflow;
1099     n->mac_table.multi_overflow = multi_overflow;
1100     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
1101     g_free(macs);
1102     rxfilter_notify(nc);
1103 
1104     return VIRTIO_NET_OK;
1105 
1106 error:
1107     g_free(macs);
1108     return VIRTIO_NET_ERR;
1109 }
1110 
1111 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
1112                                         struct iovec *iov, unsigned int iov_cnt)
1113 {
1114     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1115     uint16_t vid;
1116     size_t s;
1117     NetClientState *nc = qemu_get_queue(n->nic);
1118 
1119     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
1120     vid = virtio_lduw_p(vdev, &vid);
1121     if (s != sizeof(vid)) {
1122         return VIRTIO_NET_ERR;
1123     }
1124 
1125     if (vid >= MAX_VLAN)
1126         return VIRTIO_NET_ERR;
1127 
1128     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
1129         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
1130     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
1131         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
1132     else
1133         return VIRTIO_NET_ERR;
1134 
1135     rxfilter_notify(nc);
1136 
1137     return VIRTIO_NET_OK;
1138 }
1139 
1140 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
1141                                       struct iovec *iov, unsigned int iov_cnt)
1142 {
1143     trace_virtio_net_handle_announce(n->announce_timer.round);
1144     if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
1145         n->status & VIRTIO_NET_S_ANNOUNCE) {
1146         n->status &= ~VIRTIO_NET_S_ANNOUNCE;
1147         if (n->announce_timer.round) {
1148             qemu_announce_timer_step(&n->announce_timer);
1149         }
1150         return VIRTIO_NET_OK;
1151     } else {
1152         return VIRTIO_NET_ERR;
1153     }
1154 }
1155 
1156 static void virtio_net_disable_rss(VirtIONet *n)
1157 {
1158     if (n->rss_data.enabled) {
1159         trace_virtio_net_rss_disable();
1160     }
1161     n->rss_data.enabled = false;
1162 }
1163 
1164 static uint16_t virtio_net_handle_rss(VirtIONet *n,
1165                                       struct iovec *iov,
1166                                       unsigned int iov_cnt,
1167                                       bool do_rss)
1168 {
1169     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1170     struct virtio_net_rss_config cfg;
1171     size_t s, offset = 0, size_get;
1172     uint16_t queues, i;
1173     struct {
1174         uint16_t us;
1175         uint8_t b;
1176     } QEMU_PACKED temp;
1177     const char *err_msg = "";
1178     uint32_t err_value = 0;
1179 
1180     if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) {
1181         err_msg = "RSS is not negotiated";
1182         goto error;
1183     }
1184     if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) {
1185         err_msg = "Hash report is not negotiated";
1186         goto error;
1187     }
1188     size_get = offsetof(struct virtio_net_rss_config, indirection_table);
1189     s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get);
1190     if (s != size_get) {
1191         err_msg = "Short command buffer";
1192         err_value = (uint32_t)s;
1193         goto error;
1194     }
1195     n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types);
1196     n->rss_data.indirections_len =
1197         virtio_lduw_p(vdev, &cfg.indirection_table_mask);
1198     n->rss_data.indirections_len++;
1199     if (!do_rss) {
1200         n->rss_data.indirections_len = 1;
1201     }
1202     if (!is_power_of_2(n->rss_data.indirections_len)) {
1203         err_msg = "Invalid size of indirection table";
1204         err_value = n->rss_data.indirections_len;
1205         goto error;
1206     }
1207     if (n->rss_data.indirections_len > VIRTIO_NET_RSS_MAX_TABLE_LEN) {
1208         err_msg = "Too large indirection table";
1209         err_value = n->rss_data.indirections_len;
1210         goto error;
1211     }
1212     n->rss_data.default_queue = do_rss ?
1213         virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0;
1214     if (n->rss_data.default_queue >= n->max_queues) {
1215         err_msg = "Invalid default queue";
1216         err_value = n->rss_data.default_queue;
1217         goto error;
1218     }
1219     offset += size_get;
1220     size_get = sizeof(uint16_t) * n->rss_data.indirections_len;
1221     g_free(n->rss_data.indirections_table);
1222     n->rss_data.indirections_table = g_malloc(size_get);
1223     if (!n->rss_data.indirections_table) {
1224         err_msg = "Can't allocate indirections table";
1225         err_value = n->rss_data.indirections_len;
1226         goto error;
1227     }
1228     s = iov_to_buf(iov, iov_cnt, offset,
1229                    n->rss_data.indirections_table, size_get);
1230     if (s != size_get) {
1231         err_msg = "Short indirection table buffer";
1232         err_value = (uint32_t)s;
1233         goto error;
1234     }
1235     for (i = 0; i < n->rss_data.indirections_len; ++i) {
1236         uint16_t val = n->rss_data.indirections_table[i];
1237         n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val);
1238     }
1239     offset += size_get;
1240     size_get = sizeof(temp);
1241     s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get);
1242     if (s != size_get) {
1243         err_msg = "Can't get queues";
1244         err_value = (uint32_t)s;
1245         goto error;
1246     }
1247     queues = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queues;
1248     if (queues == 0 || queues > n->max_queues) {
1249         err_msg = "Invalid number of queues";
1250         err_value = queues;
1251         goto error;
1252     }
1253     if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) {
1254         err_msg = "Invalid key size";
1255         err_value = temp.b;
1256         goto error;
1257     }
1258     if (!temp.b && n->rss_data.hash_types) {
1259         err_msg = "No key provided";
1260         err_value = 0;
1261         goto error;
1262     }
1263     if (!temp.b && !n->rss_data.hash_types) {
1264         virtio_net_disable_rss(n);
1265         return queues;
1266     }
1267     offset += size_get;
1268     size_get = temp.b;
1269     s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get);
1270     if (s != size_get) {
1271         err_msg = "Can get key buffer";
1272         err_value = (uint32_t)s;
1273         goto error;
1274     }
1275     n->rss_data.enabled = true;
1276     trace_virtio_net_rss_enable(n->rss_data.hash_types,
1277                                 n->rss_data.indirections_len,
1278                                 temp.b);
1279     return queues;
1280 error:
1281     trace_virtio_net_rss_error(err_msg, err_value);
1282     virtio_net_disable_rss(n);
1283     return 0;
1284 }
1285 
1286 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
1287                                 struct iovec *iov, unsigned int iov_cnt)
1288 {
1289     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1290     uint16_t queues;
1291 
1292     virtio_net_disable_rss(n);
1293     if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) {
1294         queues = virtio_net_handle_rss(n, iov, iov_cnt, false);
1295         return queues ? VIRTIO_NET_OK : VIRTIO_NET_ERR;
1296     }
1297     if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
1298         queues = virtio_net_handle_rss(n, iov, iov_cnt, true);
1299     } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
1300         struct virtio_net_ctrl_mq mq;
1301         size_t s;
1302         if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) {
1303             return VIRTIO_NET_ERR;
1304         }
1305         s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
1306         if (s != sizeof(mq)) {
1307             return VIRTIO_NET_ERR;
1308         }
1309         queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
1310 
1311     } else {
1312         return VIRTIO_NET_ERR;
1313     }
1314 
1315     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1316         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1317         queues > n->max_queues ||
1318         !n->multiqueue) {
1319         return VIRTIO_NET_ERR;
1320     }
1321 
1322     n->curr_queues = queues;
1323     /* stop the backend before changing the number of queues to avoid handling a
1324      * disabled queue */
1325     virtio_net_set_status(vdev, vdev->status);
1326     virtio_net_set_queues(n);
1327 
1328     return VIRTIO_NET_OK;
1329 }
1330 
1331 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
1332 {
1333     VirtIONet *n = VIRTIO_NET(vdev);
1334     struct virtio_net_ctrl_hdr ctrl;
1335     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
1336     VirtQueueElement *elem;
1337     size_t s;
1338     struct iovec *iov, *iov2;
1339     unsigned int iov_cnt;
1340 
1341     for (;;) {
1342         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
1343         if (!elem) {
1344             break;
1345         }
1346         if (iov_size(elem->in_sg, elem->in_num) < sizeof(status) ||
1347             iov_size(elem->out_sg, elem->out_num) < sizeof(ctrl)) {
1348             virtio_error(vdev, "virtio-net ctrl missing headers");
1349             virtqueue_detach_element(vq, elem, 0);
1350             g_free(elem);
1351             break;
1352         }
1353 
1354         iov_cnt = elem->out_num;
1355         iov2 = iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num);
1356         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
1357         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
1358         if (s != sizeof(ctrl)) {
1359             status = VIRTIO_NET_ERR;
1360         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
1361             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
1362         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
1363             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
1364         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
1365             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
1366         } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
1367             status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
1368         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
1369             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
1370         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
1371             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
1372         }
1373 
1374         s = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, sizeof(status));
1375         assert(s == sizeof(status));
1376 
1377         virtqueue_push(vq, elem, sizeof(status));
1378         virtio_notify(vdev, vq);
1379         g_free(iov2);
1380         g_free(elem);
1381     }
1382 }
1383 
1384 /* RX */
1385 
1386 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
1387 {
1388     VirtIONet *n = VIRTIO_NET(vdev);
1389     int queue_index = vq2q(virtio_get_queue_index(vq));
1390 
1391     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
1392 }
1393 
1394 static bool virtio_net_can_receive(NetClientState *nc)
1395 {
1396     VirtIONet *n = qemu_get_nic_opaque(nc);
1397     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1398     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1399 
1400     if (!vdev->vm_running) {
1401         return false;
1402     }
1403 
1404     if (nc->queue_index >= n->curr_queues) {
1405         return false;
1406     }
1407 
1408     if (!virtio_queue_ready(q->rx_vq) ||
1409         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1410         return false;
1411     }
1412 
1413     return true;
1414 }
1415 
1416 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1417 {
1418     VirtIONet *n = q->n;
1419     if (virtio_queue_empty(q->rx_vq) ||
1420         (n->mergeable_rx_bufs &&
1421          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1422         virtio_queue_set_notification(q->rx_vq, 1);
1423 
1424         /* To avoid a race condition where the guest has made some buffers
1425          * available after the above check but before notification was
1426          * enabled, check for available buffers again.
1427          */
1428         if (virtio_queue_empty(q->rx_vq) ||
1429             (n->mergeable_rx_bufs &&
1430              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1431             return 0;
1432         }
1433     }
1434 
1435     virtio_queue_set_notification(q->rx_vq, 0);
1436     return 1;
1437 }
1438 
1439 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1440 {
1441     virtio_tswap16s(vdev, &hdr->hdr_len);
1442     virtio_tswap16s(vdev, &hdr->gso_size);
1443     virtio_tswap16s(vdev, &hdr->csum_start);
1444     virtio_tswap16s(vdev, &hdr->csum_offset);
1445 }
1446 
1447 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1448  * it never finds out that the packets don't have valid checksums.  This
1449  * causes dhclient to get upset.  Fedora's carried a patch for ages to
1450  * fix this with Xen but it hasn't appeared in an upstream release of
1451  * dhclient yet.
1452  *
1453  * To avoid breaking existing guests, we catch udp packets and add
1454  * checksums.  This is terrible but it's better than hacking the guest
1455  * kernels.
1456  *
1457  * N.B. if we introduce a zero-copy API, this operation is no longer free so
1458  * we should provide a mechanism to disable it to avoid polluting the host
1459  * cache.
1460  */
1461 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1462                                         uint8_t *buf, size_t size)
1463 {
1464     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1465         (size > 27 && size < 1500) && /* normal sized MTU */
1466         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1467         (buf[23] == 17) && /* ip.protocol == UDP */
1468         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1469         net_checksum_calculate(buf, size, CSUM_UDP);
1470         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1471     }
1472 }
1473 
1474 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1475                            const void *buf, size_t size)
1476 {
1477     if (n->has_vnet_hdr) {
1478         /* FIXME this cast is evil */
1479         void *wbuf = (void *)buf;
1480         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1481                                     size - n->host_hdr_len);
1482 
1483         if (n->needs_vnet_hdr_swap) {
1484             virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1485         }
1486         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1487     } else {
1488         struct virtio_net_hdr hdr = {
1489             .flags = 0,
1490             .gso_type = VIRTIO_NET_HDR_GSO_NONE
1491         };
1492         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1493     }
1494 }
1495 
1496 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1497 {
1498     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1499     static const uint8_t vlan[] = {0x81, 0x00};
1500     uint8_t *ptr = (uint8_t *)buf;
1501     int i;
1502 
1503     if (n->promisc)
1504         return 1;
1505 
1506     ptr += n->host_hdr_len;
1507 
1508     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1509         int vid = lduw_be_p(ptr + 14) & 0xfff;
1510         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1511             return 0;
1512     }
1513 
1514     if (ptr[0] & 1) { // multicast
1515         if (!memcmp(ptr, bcast, sizeof(bcast))) {
1516             return !n->nobcast;
1517         } else if (n->nomulti) {
1518             return 0;
1519         } else if (n->allmulti || n->mac_table.multi_overflow) {
1520             return 1;
1521         }
1522 
1523         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1524             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1525                 return 1;
1526             }
1527         }
1528     } else { // unicast
1529         if (n->nouni) {
1530             return 0;
1531         } else if (n->alluni || n->mac_table.uni_overflow) {
1532             return 1;
1533         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1534             return 1;
1535         }
1536 
1537         for (i = 0; i < n->mac_table.first_multi; i++) {
1538             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1539                 return 1;
1540             }
1541         }
1542     }
1543 
1544     return 0;
1545 }
1546 
1547 static uint8_t virtio_net_get_hash_type(bool isip4,
1548                                         bool isip6,
1549                                         bool isudp,
1550                                         bool istcp,
1551                                         uint32_t types)
1552 {
1553     if (isip4) {
1554         if (istcp && (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4)) {
1555             return NetPktRssIpV4Tcp;
1556         }
1557         if (isudp && (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4)) {
1558             return NetPktRssIpV4Udp;
1559         }
1560         if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
1561             return NetPktRssIpV4;
1562         }
1563     } else if (isip6) {
1564         uint32_t mask = VIRTIO_NET_RSS_HASH_TYPE_TCP_EX |
1565                         VIRTIO_NET_RSS_HASH_TYPE_TCPv6;
1566 
1567         if (istcp && (types & mask)) {
1568             return (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) ?
1569                 NetPktRssIpV6TcpEx : NetPktRssIpV6Tcp;
1570         }
1571         mask = VIRTIO_NET_RSS_HASH_TYPE_UDP_EX | VIRTIO_NET_RSS_HASH_TYPE_UDPv6;
1572         if (isudp && (types & mask)) {
1573             return (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) ?
1574                 NetPktRssIpV6UdpEx : NetPktRssIpV6Udp;
1575         }
1576         mask = VIRTIO_NET_RSS_HASH_TYPE_IP_EX | VIRTIO_NET_RSS_HASH_TYPE_IPv6;
1577         if (types & mask) {
1578             return (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) ?
1579                 NetPktRssIpV6Ex : NetPktRssIpV6;
1580         }
1581     }
1582     return 0xff;
1583 }
1584 
1585 static void virtio_set_packet_hash(const uint8_t *buf, uint8_t report,
1586                                    uint32_t hash)
1587 {
1588     struct virtio_net_hdr_v1_hash *hdr = (void *)buf;
1589     hdr->hash_value = hash;
1590     hdr->hash_report = report;
1591 }
1592 
1593 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf,
1594                                   size_t size)
1595 {
1596     VirtIONet *n = qemu_get_nic_opaque(nc);
1597     unsigned int index = nc->queue_index, new_index = index;
1598     struct NetRxPkt *pkt = n->rx_pkt;
1599     uint8_t net_hash_type;
1600     uint32_t hash;
1601     bool isip4, isip6, isudp, istcp;
1602     static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = {
1603         VIRTIO_NET_HASH_REPORT_IPv4,
1604         VIRTIO_NET_HASH_REPORT_TCPv4,
1605         VIRTIO_NET_HASH_REPORT_TCPv6,
1606         VIRTIO_NET_HASH_REPORT_IPv6,
1607         VIRTIO_NET_HASH_REPORT_IPv6_EX,
1608         VIRTIO_NET_HASH_REPORT_TCPv6_EX,
1609         VIRTIO_NET_HASH_REPORT_UDPv4,
1610         VIRTIO_NET_HASH_REPORT_UDPv6,
1611         VIRTIO_NET_HASH_REPORT_UDPv6_EX
1612     };
1613 
1614     net_rx_pkt_set_protocols(pkt, buf + n->host_hdr_len,
1615                              size - n->host_hdr_len);
1616     net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
1617     if (isip4 && (net_rx_pkt_get_ip4_info(pkt)->fragment)) {
1618         istcp = isudp = false;
1619     }
1620     if (isip6 && (net_rx_pkt_get_ip6_info(pkt)->fragment)) {
1621         istcp = isudp = false;
1622     }
1623     net_hash_type = virtio_net_get_hash_type(isip4, isip6, isudp, istcp,
1624                                              n->rss_data.hash_types);
1625     if (net_hash_type > NetPktRssIpV6UdpEx) {
1626         if (n->rss_data.populate_hash) {
1627             virtio_set_packet_hash(buf, VIRTIO_NET_HASH_REPORT_NONE, 0);
1628         }
1629         return n->rss_data.redirect ? n->rss_data.default_queue : -1;
1630     }
1631 
1632     hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key);
1633 
1634     if (n->rss_data.populate_hash) {
1635         virtio_set_packet_hash(buf, reports[net_hash_type], hash);
1636     }
1637 
1638     if (n->rss_data.redirect) {
1639         new_index = hash & (n->rss_data.indirections_len - 1);
1640         new_index = n->rss_data.indirections_table[new_index];
1641     }
1642 
1643     return (index == new_index) ? -1 : new_index;
1644 }
1645 
1646 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
1647                                       size_t size, bool no_rss)
1648 {
1649     VirtIONet *n = qemu_get_nic_opaque(nc);
1650     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1651     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1652     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1653     struct virtio_net_hdr_mrg_rxbuf mhdr;
1654     unsigned mhdr_cnt = 0;
1655     size_t offset, i, guest_offset;
1656 
1657     if (!virtio_net_can_receive(nc)) {
1658         return -1;
1659     }
1660 
1661     if (!no_rss && n->rss_data.enabled) {
1662         int index = virtio_net_process_rss(nc, buf, size);
1663         if (index >= 0) {
1664             NetClientState *nc2 = qemu_get_subqueue(n->nic, index);
1665             return virtio_net_receive_rcu(nc2, buf, size, true);
1666         }
1667     }
1668 
1669     /* hdr_len refers to the header we supply to the guest */
1670     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1671         return 0;
1672     }
1673 
1674     if (!receive_filter(n, buf, size))
1675         return size;
1676 
1677     offset = i = 0;
1678 
1679     while (offset < size) {
1680         VirtQueueElement *elem;
1681         int len, total;
1682         const struct iovec *sg;
1683 
1684         total = 0;
1685 
1686         elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1687         if (!elem) {
1688             if (i) {
1689                 virtio_error(vdev, "virtio-net unexpected empty queue: "
1690                              "i %zd mergeable %d offset %zd, size %zd, "
1691                              "guest hdr len %zd, host hdr len %zd "
1692                              "guest features 0x%" PRIx64,
1693                              i, n->mergeable_rx_bufs, offset, size,
1694                              n->guest_hdr_len, n->host_hdr_len,
1695                              vdev->guest_features);
1696             }
1697             return -1;
1698         }
1699 
1700         if (elem->in_num < 1) {
1701             virtio_error(vdev,
1702                          "virtio-net receive queue contains no in buffers");
1703             virtqueue_detach_element(q->rx_vq, elem, 0);
1704             g_free(elem);
1705             return -1;
1706         }
1707 
1708         sg = elem->in_sg;
1709         if (i == 0) {
1710             assert(offset == 0);
1711             if (n->mergeable_rx_bufs) {
1712                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1713                                     sg, elem->in_num,
1714                                     offsetof(typeof(mhdr), num_buffers),
1715                                     sizeof(mhdr.num_buffers));
1716             }
1717 
1718             receive_header(n, sg, elem->in_num, buf, size);
1719             if (n->rss_data.populate_hash) {
1720                 offset = sizeof(mhdr);
1721                 iov_from_buf(sg, elem->in_num, offset,
1722                              buf + offset, n->host_hdr_len - sizeof(mhdr));
1723             }
1724             offset = n->host_hdr_len;
1725             total += n->guest_hdr_len;
1726             guest_offset = n->guest_hdr_len;
1727         } else {
1728             guest_offset = 0;
1729         }
1730 
1731         /* copy in packet.  ugh */
1732         len = iov_from_buf(sg, elem->in_num, guest_offset,
1733                            buf + offset, size - offset);
1734         total += len;
1735         offset += len;
1736         /* If buffers can't be merged, at this point we
1737          * must have consumed the complete packet.
1738          * Otherwise, drop it. */
1739         if (!n->mergeable_rx_bufs && offset < size) {
1740             virtqueue_unpop(q->rx_vq, elem, total);
1741             g_free(elem);
1742             return size;
1743         }
1744 
1745         /* signal other side */
1746         virtqueue_fill(q->rx_vq, elem, total, i++);
1747         g_free(elem);
1748     }
1749 
1750     if (mhdr_cnt) {
1751         virtio_stw_p(vdev, &mhdr.num_buffers, i);
1752         iov_from_buf(mhdr_sg, mhdr_cnt,
1753                      0,
1754                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1755     }
1756 
1757     virtqueue_flush(q->rx_vq, i);
1758     virtio_notify(vdev, q->rx_vq);
1759 
1760     return size;
1761 }
1762 
1763 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf,
1764                                   size_t size)
1765 {
1766     RCU_READ_LOCK_GUARD();
1767 
1768     return virtio_net_receive_rcu(nc, buf, size, false);
1769 }
1770 
1771 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain,
1772                                          const uint8_t *buf,
1773                                          VirtioNetRscUnit *unit)
1774 {
1775     uint16_t ip_hdrlen;
1776     struct ip_header *ip;
1777 
1778     ip = (struct ip_header *)(buf + chain->n->guest_hdr_len
1779                               + sizeof(struct eth_header));
1780     unit->ip = (void *)ip;
1781     ip_hdrlen = (ip->ip_ver_len & 0xF) << 2;
1782     unit->ip_plen = &ip->ip_len;
1783     unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen);
1784     unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1785     unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen;
1786 }
1787 
1788 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain,
1789                                          const uint8_t *buf,
1790                                          VirtioNetRscUnit *unit)
1791 {
1792     struct ip6_header *ip6;
1793 
1794     ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len
1795                                  + sizeof(struct eth_header));
1796     unit->ip = ip6;
1797     unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
1798     unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip)
1799                                         + sizeof(struct ip6_header));
1800     unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1801 
1802     /* There is a difference between payload lenght in ipv4 and v6,
1803        ip header is excluded in ipv6 */
1804     unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen;
1805 }
1806 
1807 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain,
1808                                        VirtioNetRscSeg *seg)
1809 {
1810     int ret;
1811     struct virtio_net_hdr_v1 *h;
1812 
1813     h = (struct virtio_net_hdr_v1 *)seg->buf;
1814     h->flags = 0;
1815     h->gso_type = VIRTIO_NET_HDR_GSO_NONE;
1816 
1817     if (seg->is_coalesced) {
1818         h->rsc.segments = seg->packets;
1819         h->rsc.dup_acks = seg->dup_ack;
1820         h->flags = VIRTIO_NET_HDR_F_RSC_INFO;
1821         if (chain->proto == ETH_P_IP) {
1822             h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1823         } else {
1824             h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1825         }
1826     }
1827 
1828     ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size);
1829     QTAILQ_REMOVE(&chain->buffers, seg, next);
1830     g_free(seg->buf);
1831     g_free(seg);
1832 
1833     return ret;
1834 }
1835 
1836 static void virtio_net_rsc_purge(void *opq)
1837 {
1838     VirtioNetRscSeg *seg, *rn;
1839     VirtioNetRscChain *chain = (VirtioNetRscChain *)opq;
1840 
1841     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) {
1842         if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
1843             chain->stat.purge_failed++;
1844             continue;
1845         }
1846     }
1847 
1848     chain->stat.timer++;
1849     if (!QTAILQ_EMPTY(&chain->buffers)) {
1850         timer_mod(chain->drain_timer,
1851               qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
1852     }
1853 }
1854 
1855 static void virtio_net_rsc_cleanup(VirtIONet *n)
1856 {
1857     VirtioNetRscChain *chain, *rn_chain;
1858     VirtioNetRscSeg *seg, *rn_seg;
1859 
1860     QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) {
1861         QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) {
1862             QTAILQ_REMOVE(&chain->buffers, seg, next);
1863             g_free(seg->buf);
1864             g_free(seg);
1865         }
1866 
1867         timer_free(chain->drain_timer);
1868         QTAILQ_REMOVE(&n->rsc_chains, chain, next);
1869         g_free(chain);
1870     }
1871 }
1872 
1873 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain,
1874                                      NetClientState *nc,
1875                                      const uint8_t *buf, size_t size)
1876 {
1877     uint16_t hdr_len;
1878     VirtioNetRscSeg *seg;
1879 
1880     hdr_len = chain->n->guest_hdr_len;
1881     seg = g_malloc(sizeof(VirtioNetRscSeg));
1882     seg->buf = g_malloc(hdr_len + sizeof(struct eth_header)
1883         + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD);
1884     memcpy(seg->buf, buf, size);
1885     seg->size = size;
1886     seg->packets = 1;
1887     seg->dup_ack = 0;
1888     seg->is_coalesced = 0;
1889     seg->nc = nc;
1890 
1891     QTAILQ_INSERT_TAIL(&chain->buffers, seg, next);
1892     chain->stat.cache++;
1893 
1894     switch (chain->proto) {
1895     case ETH_P_IP:
1896         virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
1897         break;
1898     case ETH_P_IPV6:
1899         virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit);
1900         break;
1901     default:
1902         g_assert_not_reached();
1903     }
1904 }
1905 
1906 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain,
1907                                          VirtioNetRscSeg *seg,
1908                                          const uint8_t *buf,
1909                                          struct tcp_header *n_tcp,
1910                                          struct tcp_header *o_tcp)
1911 {
1912     uint32_t nack, oack;
1913     uint16_t nwin, owin;
1914 
1915     nack = htonl(n_tcp->th_ack);
1916     nwin = htons(n_tcp->th_win);
1917     oack = htonl(o_tcp->th_ack);
1918     owin = htons(o_tcp->th_win);
1919 
1920     if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) {
1921         chain->stat.ack_out_of_win++;
1922         return RSC_FINAL;
1923     } else if (nack == oack) {
1924         /* duplicated ack or window probe */
1925         if (nwin == owin) {
1926             /* duplicated ack, add dup ack count due to whql test up to 1 */
1927             chain->stat.dup_ack++;
1928             return RSC_FINAL;
1929         } else {
1930             /* Coalesce window update */
1931             o_tcp->th_win = n_tcp->th_win;
1932             chain->stat.win_update++;
1933             return RSC_COALESCE;
1934         }
1935     } else {
1936         /* pure ack, go to 'C', finalize*/
1937         chain->stat.pure_ack++;
1938         return RSC_FINAL;
1939     }
1940 }
1941 
1942 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain,
1943                                             VirtioNetRscSeg *seg,
1944                                             const uint8_t *buf,
1945                                             VirtioNetRscUnit *n_unit)
1946 {
1947     void *data;
1948     uint16_t o_ip_len;
1949     uint32_t nseq, oseq;
1950     VirtioNetRscUnit *o_unit;
1951 
1952     o_unit = &seg->unit;
1953     o_ip_len = htons(*o_unit->ip_plen);
1954     nseq = htonl(n_unit->tcp->th_seq);
1955     oseq = htonl(o_unit->tcp->th_seq);
1956 
1957     /* out of order or retransmitted. */
1958     if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) {
1959         chain->stat.data_out_of_win++;
1960         return RSC_FINAL;
1961     }
1962 
1963     data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen;
1964     if (nseq == oseq) {
1965         if ((o_unit->payload == 0) && n_unit->payload) {
1966             /* From no payload to payload, normal case, not a dup ack or etc */
1967             chain->stat.data_after_pure_ack++;
1968             goto coalesce;
1969         } else {
1970             return virtio_net_rsc_handle_ack(chain, seg, buf,
1971                                              n_unit->tcp, o_unit->tcp);
1972         }
1973     } else if ((nseq - oseq) != o_unit->payload) {
1974         /* Not a consistent packet, out of order */
1975         chain->stat.data_out_of_order++;
1976         return RSC_FINAL;
1977     } else {
1978 coalesce:
1979         if ((o_ip_len + n_unit->payload) > chain->max_payload) {
1980             chain->stat.over_size++;
1981             return RSC_FINAL;
1982         }
1983 
1984         /* Here comes the right data, the payload length in v4/v6 is different,
1985            so use the field value to update and record the new data len */
1986         o_unit->payload += n_unit->payload; /* update new data len */
1987 
1988         /* update field in ip header */
1989         *o_unit->ip_plen = htons(o_ip_len + n_unit->payload);
1990 
1991         /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
1992            for windows guest, while this may change the behavior for linux
1993            guest (only if it uses RSC feature). */
1994         o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags;
1995 
1996         o_unit->tcp->th_ack = n_unit->tcp->th_ack;
1997         o_unit->tcp->th_win = n_unit->tcp->th_win;
1998 
1999         memmove(seg->buf + seg->size, data, n_unit->payload);
2000         seg->size += n_unit->payload;
2001         seg->packets++;
2002         chain->stat.coalesced++;
2003         return RSC_COALESCE;
2004     }
2005 }
2006 
2007 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain,
2008                                         VirtioNetRscSeg *seg,
2009                                         const uint8_t *buf, size_t size,
2010                                         VirtioNetRscUnit *unit)
2011 {
2012     struct ip_header *ip1, *ip2;
2013 
2014     ip1 = (struct ip_header *)(unit->ip);
2015     ip2 = (struct ip_header *)(seg->unit.ip);
2016     if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst)
2017         || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2018         || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2019         chain->stat.no_match++;
2020         return RSC_NO_MATCH;
2021     }
2022 
2023     return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2024 }
2025 
2026 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain,
2027                                         VirtioNetRscSeg *seg,
2028                                         const uint8_t *buf, size_t size,
2029                                         VirtioNetRscUnit *unit)
2030 {
2031     struct ip6_header *ip1, *ip2;
2032 
2033     ip1 = (struct ip6_header *)(unit->ip);
2034     ip2 = (struct ip6_header *)(seg->unit.ip);
2035     if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address))
2036         || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address))
2037         || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2038         || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2039             chain->stat.no_match++;
2040             return RSC_NO_MATCH;
2041     }
2042 
2043     return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2044 }
2045 
2046 /* Packets with 'SYN' should bypass, other flag should be sent after drain
2047  * to prevent out of order */
2048 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain,
2049                                          struct tcp_header *tcp)
2050 {
2051     uint16_t tcp_hdr;
2052     uint16_t tcp_flag;
2053 
2054     tcp_flag = htons(tcp->th_offset_flags);
2055     tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10;
2056     tcp_flag &= VIRTIO_NET_TCP_FLAG;
2057     if (tcp_flag & TH_SYN) {
2058         chain->stat.tcp_syn++;
2059         return RSC_BYPASS;
2060     }
2061 
2062     if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) {
2063         chain->stat.tcp_ctrl_drain++;
2064         return RSC_FINAL;
2065     }
2066 
2067     if (tcp_hdr > sizeof(struct tcp_header)) {
2068         chain->stat.tcp_all_opt++;
2069         return RSC_FINAL;
2070     }
2071 
2072     return RSC_CANDIDATE;
2073 }
2074 
2075 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain,
2076                                          NetClientState *nc,
2077                                          const uint8_t *buf, size_t size,
2078                                          VirtioNetRscUnit *unit)
2079 {
2080     int ret;
2081     VirtioNetRscSeg *seg, *nseg;
2082 
2083     if (QTAILQ_EMPTY(&chain->buffers)) {
2084         chain->stat.empty_cache++;
2085         virtio_net_rsc_cache_buf(chain, nc, buf, size);
2086         timer_mod(chain->drain_timer,
2087               qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
2088         return size;
2089     }
2090 
2091     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2092         if (chain->proto == ETH_P_IP) {
2093             ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
2094         } else {
2095             ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit);
2096         }
2097 
2098         if (ret == RSC_FINAL) {
2099             if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2100                 /* Send failed */
2101                 chain->stat.final_failed++;
2102                 return 0;
2103             }
2104 
2105             /* Send current packet */
2106             return virtio_net_do_receive(nc, buf, size);
2107         } else if (ret == RSC_NO_MATCH) {
2108             continue;
2109         } else {
2110             /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
2111             seg->is_coalesced = 1;
2112             return size;
2113         }
2114     }
2115 
2116     chain->stat.no_match_cache++;
2117     virtio_net_rsc_cache_buf(chain, nc, buf, size);
2118     return size;
2119 }
2120 
2121 /* Drain a connection data, this is to avoid out of order segments */
2122 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain,
2123                                         NetClientState *nc,
2124                                         const uint8_t *buf, size_t size,
2125                                         uint16_t ip_start, uint16_t ip_size,
2126                                         uint16_t tcp_port)
2127 {
2128     VirtioNetRscSeg *seg, *nseg;
2129     uint32_t ppair1, ppair2;
2130 
2131     ppair1 = *(uint32_t *)(buf + tcp_port);
2132     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2133         ppair2 = *(uint32_t *)(seg->buf + tcp_port);
2134         if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size)
2135             || (ppair1 != ppair2)) {
2136             continue;
2137         }
2138         if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2139             chain->stat.drain_failed++;
2140         }
2141 
2142         break;
2143     }
2144 
2145     return virtio_net_do_receive(nc, buf, size);
2146 }
2147 
2148 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain,
2149                                             struct ip_header *ip,
2150                                             const uint8_t *buf, size_t size)
2151 {
2152     uint16_t ip_len;
2153 
2154     /* Not an ipv4 packet */
2155     if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) {
2156         chain->stat.ip_option++;
2157         return RSC_BYPASS;
2158     }
2159 
2160     /* Don't handle packets with ip option */
2161     if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) {
2162         chain->stat.ip_option++;
2163         return RSC_BYPASS;
2164     }
2165 
2166     if (ip->ip_p != IPPROTO_TCP) {
2167         chain->stat.bypass_not_tcp++;
2168         return RSC_BYPASS;
2169     }
2170 
2171     /* Don't handle packets with ip fragment */
2172     if (!(htons(ip->ip_off) & IP_DF)) {
2173         chain->stat.ip_frag++;
2174         return RSC_BYPASS;
2175     }
2176 
2177     /* Don't handle packets with ecn flag */
2178     if (IPTOS_ECN(ip->ip_tos)) {
2179         chain->stat.ip_ecn++;
2180         return RSC_BYPASS;
2181     }
2182 
2183     ip_len = htons(ip->ip_len);
2184     if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header))
2185         || ip_len > (size - chain->n->guest_hdr_len -
2186                      sizeof(struct eth_header))) {
2187         chain->stat.ip_hacked++;
2188         return RSC_BYPASS;
2189     }
2190 
2191     return RSC_CANDIDATE;
2192 }
2193 
2194 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain,
2195                                       NetClientState *nc,
2196                                       const uint8_t *buf, size_t size)
2197 {
2198     int32_t ret;
2199     uint16_t hdr_len;
2200     VirtioNetRscUnit unit;
2201 
2202     hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2203 
2204     if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)
2205         + sizeof(struct tcp_header))) {
2206         chain->stat.bypass_not_tcp++;
2207         return virtio_net_do_receive(nc, buf, size);
2208     }
2209 
2210     virtio_net_rsc_extract_unit4(chain, buf, &unit);
2211     if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size)
2212         != RSC_CANDIDATE) {
2213         return virtio_net_do_receive(nc, buf, size);
2214     }
2215 
2216     ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2217     if (ret == RSC_BYPASS) {
2218         return virtio_net_do_receive(nc, buf, size);
2219     } else if (ret == RSC_FINAL) {
2220         return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2221                 ((hdr_len + sizeof(struct eth_header)) + 12),
2222                 VIRTIO_NET_IP4_ADDR_SIZE,
2223                 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header));
2224     }
2225 
2226     return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2227 }
2228 
2229 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain,
2230                                             struct ip6_header *ip6,
2231                                             const uint8_t *buf, size_t size)
2232 {
2233     uint16_t ip_len;
2234 
2235     if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4)
2236         != IP_HEADER_VERSION_6) {
2237         return RSC_BYPASS;
2238     }
2239 
2240     /* Both option and protocol is checked in this */
2241     if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) {
2242         chain->stat.bypass_not_tcp++;
2243         return RSC_BYPASS;
2244     }
2245 
2246     ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
2247     if (ip_len < sizeof(struct tcp_header) ||
2248         ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header)
2249                   - sizeof(struct ip6_header))) {
2250         chain->stat.ip_hacked++;
2251         return RSC_BYPASS;
2252     }
2253 
2254     /* Don't handle packets with ecn flag */
2255     if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) {
2256         chain->stat.ip_ecn++;
2257         return RSC_BYPASS;
2258     }
2259 
2260     return RSC_CANDIDATE;
2261 }
2262 
2263 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc,
2264                                       const uint8_t *buf, size_t size)
2265 {
2266     int32_t ret;
2267     uint16_t hdr_len;
2268     VirtioNetRscChain *chain;
2269     VirtioNetRscUnit unit;
2270 
2271     chain = (VirtioNetRscChain *)opq;
2272     hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2273 
2274     if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header)
2275         + sizeof(tcp_header))) {
2276         return virtio_net_do_receive(nc, buf, size);
2277     }
2278 
2279     virtio_net_rsc_extract_unit6(chain, buf, &unit);
2280     if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain,
2281                                                  unit.ip, buf, size)) {
2282         return virtio_net_do_receive(nc, buf, size);
2283     }
2284 
2285     ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2286     if (ret == RSC_BYPASS) {
2287         return virtio_net_do_receive(nc, buf, size);
2288     } else if (ret == RSC_FINAL) {
2289         return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2290                 ((hdr_len + sizeof(struct eth_header)) + 8),
2291                 VIRTIO_NET_IP6_ADDR_SIZE,
2292                 hdr_len + sizeof(struct eth_header)
2293                 + sizeof(struct ip6_header));
2294     }
2295 
2296     return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2297 }
2298 
2299 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n,
2300                                                       NetClientState *nc,
2301                                                       uint16_t proto)
2302 {
2303     VirtioNetRscChain *chain;
2304 
2305     if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) {
2306         return NULL;
2307     }
2308 
2309     QTAILQ_FOREACH(chain, &n->rsc_chains, next) {
2310         if (chain->proto == proto) {
2311             return chain;
2312         }
2313     }
2314 
2315     chain = g_malloc(sizeof(*chain));
2316     chain->n = n;
2317     chain->proto = proto;
2318     if (proto == (uint16_t)ETH_P_IP) {
2319         chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
2320         chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2321     } else {
2322         chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD;
2323         chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2324     }
2325     chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST,
2326                                       virtio_net_rsc_purge, chain);
2327     memset(&chain->stat, 0, sizeof(chain->stat));
2328 
2329     QTAILQ_INIT(&chain->buffers);
2330     QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next);
2331 
2332     return chain;
2333 }
2334 
2335 static ssize_t virtio_net_rsc_receive(NetClientState *nc,
2336                                       const uint8_t *buf,
2337                                       size_t size)
2338 {
2339     uint16_t proto;
2340     VirtioNetRscChain *chain;
2341     struct eth_header *eth;
2342     VirtIONet *n;
2343 
2344     n = qemu_get_nic_opaque(nc);
2345     if (size < (n->host_hdr_len + sizeof(struct eth_header))) {
2346         return virtio_net_do_receive(nc, buf, size);
2347     }
2348 
2349     eth = (struct eth_header *)(buf + n->guest_hdr_len);
2350     proto = htons(eth->h_proto);
2351 
2352     chain = virtio_net_rsc_lookup_chain(n, nc, proto);
2353     if (chain) {
2354         chain->stat.received++;
2355         if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) {
2356             return virtio_net_rsc_receive4(chain, nc, buf, size);
2357         } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) {
2358             return virtio_net_rsc_receive6(chain, nc, buf, size);
2359         }
2360     }
2361     return virtio_net_do_receive(nc, buf, size);
2362 }
2363 
2364 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf,
2365                                   size_t size)
2366 {
2367     VirtIONet *n = qemu_get_nic_opaque(nc);
2368     if ((n->rsc4_enabled || n->rsc6_enabled)) {
2369         return virtio_net_rsc_receive(nc, buf, size);
2370     } else {
2371         return virtio_net_do_receive(nc, buf, size);
2372     }
2373 }
2374 
2375 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
2376 
2377 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
2378 {
2379     VirtIONet *n = qemu_get_nic_opaque(nc);
2380     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
2381     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2382 
2383     virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
2384     virtio_notify(vdev, q->tx_vq);
2385 
2386     g_free(q->async_tx.elem);
2387     q->async_tx.elem = NULL;
2388 
2389     virtio_queue_set_notification(q->tx_vq, 1);
2390     virtio_net_flush_tx(q);
2391 }
2392 
2393 /* TX */
2394 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
2395 {
2396     VirtIONet *n = q->n;
2397     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2398     VirtQueueElement *elem;
2399     int32_t num_packets = 0;
2400     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
2401     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2402         return num_packets;
2403     }
2404 
2405     if (q->async_tx.elem) {
2406         virtio_queue_set_notification(q->tx_vq, 0);
2407         return num_packets;
2408     }
2409 
2410     for (;;) {
2411         ssize_t ret;
2412         unsigned int out_num;
2413         struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
2414         struct virtio_net_hdr_mrg_rxbuf mhdr;
2415 
2416         elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
2417         if (!elem) {
2418             break;
2419         }
2420 
2421         out_num = elem->out_num;
2422         out_sg = elem->out_sg;
2423         if (out_num < 1) {
2424             virtio_error(vdev, "virtio-net header not in first element");
2425             virtqueue_detach_element(q->tx_vq, elem, 0);
2426             g_free(elem);
2427             return -EINVAL;
2428         }
2429 
2430         if (n->has_vnet_hdr) {
2431             if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
2432                 n->guest_hdr_len) {
2433                 virtio_error(vdev, "virtio-net header incorrect");
2434                 virtqueue_detach_element(q->tx_vq, elem, 0);
2435                 g_free(elem);
2436                 return -EINVAL;
2437             }
2438             if (n->needs_vnet_hdr_swap) {
2439                 virtio_net_hdr_swap(vdev, (void *) &mhdr);
2440                 sg2[0].iov_base = &mhdr;
2441                 sg2[0].iov_len = n->guest_hdr_len;
2442                 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
2443                                    out_sg, out_num,
2444                                    n->guest_hdr_len, -1);
2445                 if (out_num == VIRTQUEUE_MAX_SIZE) {
2446                     goto drop;
2447                 }
2448                 out_num += 1;
2449                 out_sg = sg2;
2450             }
2451         }
2452         /*
2453          * If host wants to see the guest header as is, we can
2454          * pass it on unchanged. Otherwise, copy just the parts
2455          * that host is interested in.
2456          */
2457         assert(n->host_hdr_len <= n->guest_hdr_len);
2458         if (n->host_hdr_len != n->guest_hdr_len) {
2459             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
2460                                        out_sg, out_num,
2461                                        0, n->host_hdr_len);
2462             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
2463                              out_sg, out_num,
2464                              n->guest_hdr_len, -1);
2465             out_num = sg_num;
2466             out_sg = sg;
2467         }
2468 
2469         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
2470                                       out_sg, out_num, virtio_net_tx_complete);
2471         if (ret == 0) {
2472             virtio_queue_set_notification(q->tx_vq, 0);
2473             q->async_tx.elem = elem;
2474             return -EBUSY;
2475         }
2476 
2477 drop:
2478         virtqueue_push(q->tx_vq, elem, 0);
2479         virtio_notify(vdev, q->tx_vq);
2480         g_free(elem);
2481 
2482         if (++num_packets >= n->tx_burst) {
2483             break;
2484         }
2485     }
2486     return num_packets;
2487 }
2488 
2489 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
2490 {
2491     VirtIONet *n = VIRTIO_NET(vdev);
2492     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2493 
2494     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2495         virtio_net_drop_tx_queue_data(vdev, vq);
2496         return;
2497     }
2498 
2499     /* This happens when device was stopped but VCPU wasn't. */
2500     if (!vdev->vm_running) {
2501         q->tx_waiting = 1;
2502         return;
2503     }
2504 
2505     if (q->tx_waiting) {
2506         virtio_queue_set_notification(vq, 1);
2507         timer_del(q->tx_timer);
2508         q->tx_waiting = 0;
2509         if (virtio_net_flush_tx(q) == -EINVAL) {
2510             return;
2511         }
2512     } else {
2513         timer_mod(q->tx_timer,
2514                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2515         q->tx_waiting = 1;
2516         virtio_queue_set_notification(vq, 0);
2517     }
2518 }
2519 
2520 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
2521 {
2522     VirtIONet *n = VIRTIO_NET(vdev);
2523     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2524 
2525     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2526         virtio_net_drop_tx_queue_data(vdev, vq);
2527         return;
2528     }
2529 
2530     if (unlikely(q->tx_waiting)) {
2531         return;
2532     }
2533     q->tx_waiting = 1;
2534     /* This happens when device was stopped but VCPU wasn't. */
2535     if (!vdev->vm_running) {
2536         return;
2537     }
2538     virtio_queue_set_notification(vq, 0);
2539     qemu_bh_schedule(q->tx_bh);
2540 }
2541 
2542 static void virtio_net_tx_timer(void *opaque)
2543 {
2544     VirtIONetQueue *q = opaque;
2545     VirtIONet *n = q->n;
2546     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2547     /* This happens when device was stopped but BH wasn't. */
2548     if (!vdev->vm_running) {
2549         /* Make sure tx waiting is set, so we'll run when restarted. */
2550         assert(q->tx_waiting);
2551         return;
2552     }
2553 
2554     q->tx_waiting = 0;
2555 
2556     /* Just in case the driver is not ready on more */
2557     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2558         return;
2559     }
2560 
2561     virtio_queue_set_notification(q->tx_vq, 1);
2562     virtio_net_flush_tx(q);
2563 }
2564 
2565 static void virtio_net_tx_bh(void *opaque)
2566 {
2567     VirtIONetQueue *q = opaque;
2568     VirtIONet *n = q->n;
2569     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2570     int32_t ret;
2571 
2572     /* This happens when device was stopped but BH wasn't. */
2573     if (!vdev->vm_running) {
2574         /* Make sure tx waiting is set, so we'll run when restarted. */
2575         assert(q->tx_waiting);
2576         return;
2577     }
2578 
2579     q->tx_waiting = 0;
2580 
2581     /* Just in case the driver is not ready on more */
2582     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
2583         return;
2584     }
2585 
2586     ret = virtio_net_flush_tx(q);
2587     if (ret == -EBUSY || ret == -EINVAL) {
2588         return; /* Notification re-enable handled by tx_complete or device
2589                  * broken */
2590     }
2591 
2592     /* If we flush a full burst of packets, assume there are
2593      * more coming and immediately reschedule */
2594     if (ret >= n->tx_burst) {
2595         qemu_bh_schedule(q->tx_bh);
2596         q->tx_waiting = 1;
2597         return;
2598     }
2599 
2600     /* If less than a full burst, re-enable notification and flush
2601      * anything that may have come in while we weren't looking.  If
2602      * we find something, assume the guest is still active and reschedule */
2603     virtio_queue_set_notification(q->tx_vq, 1);
2604     ret = virtio_net_flush_tx(q);
2605     if (ret == -EINVAL) {
2606         return;
2607     } else if (ret > 0) {
2608         virtio_queue_set_notification(q->tx_vq, 0);
2609         qemu_bh_schedule(q->tx_bh);
2610         q->tx_waiting = 1;
2611     }
2612 }
2613 
2614 static void virtio_net_add_queue(VirtIONet *n, int index)
2615 {
2616     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2617 
2618     n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
2619                                            virtio_net_handle_rx);
2620 
2621     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
2622         n->vqs[index].tx_vq =
2623             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2624                              virtio_net_handle_tx_timer);
2625         n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
2626                                               virtio_net_tx_timer,
2627                                               &n->vqs[index]);
2628     } else {
2629         n->vqs[index].tx_vq =
2630             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2631                              virtio_net_handle_tx_bh);
2632         n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
2633     }
2634 
2635     n->vqs[index].tx_waiting = 0;
2636     n->vqs[index].n = n;
2637 }
2638 
2639 static void virtio_net_del_queue(VirtIONet *n, int index)
2640 {
2641     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2642     VirtIONetQueue *q = &n->vqs[index];
2643     NetClientState *nc = qemu_get_subqueue(n->nic, index);
2644 
2645     qemu_purge_queued_packets(nc);
2646 
2647     virtio_del_queue(vdev, index * 2);
2648     if (q->tx_timer) {
2649         timer_free(q->tx_timer);
2650         q->tx_timer = NULL;
2651     } else {
2652         qemu_bh_delete(q->tx_bh);
2653         q->tx_bh = NULL;
2654     }
2655     q->tx_waiting = 0;
2656     virtio_del_queue(vdev, index * 2 + 1);
2657 }
2658 
2659 static void virtio_net_change_num_queues(VirtIONet *n, int new_max_queues)
2660 {
2661     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2662     int old_num_queues = virtio_get_num_queues(vdev);
2663     int new_num_queues = new_max_queues * 2 + 1;
2664     int i;
2665 
2666     assert(old_num_queues >= 3);
2667     assert(old_num_queues % 2 == 1);
2668 
2669     if (old_num_queues == new_num_queues) {
2670         return;
2671     }
2672 
2673     /*
2674      * We always need to remove and add ctrl vq if
2675      * old_num_queues != new_num_queues. Remove ctrl_vq first,
2676      * and then we only enter one of the following two loops.
2677      */
2678     virtio_del_queue(vdev, old_num_queues - 1);
2679 
2680     for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
2681         /* new_num_queues < old_num_queues */
2682         virtio_net_del_queue(n, i / 2);
2683     }
2684 
2685     for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
2686         /* new_num_queues > old_num_queues */
2687         virtio_net_add_queue(n, i / 2);
2688     }
2689 
2690     /* add ctrl_vq last */
2691     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
2692 }
2693 
2694 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
2695 {
2696     int max = multiqueue ? n->max_queues : 1;
2697 
2698     n->multiqueue = multiqueue;
2699     virtio_net_change_num_queues(n, max);
2700 
2701     virtio_net_set_queues(n);
2702 }
2703 
2704 static int virtio_net_post_load_device(void *opaque, int version_id)
2705 {
2706     VirtIONet *n = opaque;
2707     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2708     int i, link_down;
2709 
2710     trace_virtio_net_post_load_device();
2711     virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
2712                                virtio_vdev_has_feature(vdev,
2713                                                        VIRTIO_F_VERSION_1),
2714                                virtio_vdev_has_feature(vdev,
2715                                                        VIRTIO_NET_F_HASH_REPORT));
2716 
2717     /* MAC_TABLE_ENTRIES may be different from the saved image */
2718     if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
2719         n->mac_table.in_use = 0;
2720     }
2721 
2722     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
2723         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
2724     }
2725 
2726     /*
2727      * curr_guest_offloads will be later overwritten by the
2728      * virtio_set_features_nocheck call done from the virtio_load.
2729      * Here we make sure it is preserved and restored accordingly
2730      * in the virtio_net_post_load_virtio callback.
2731      */
2732     n->saved_guest_offloads = n->curr_guest_offloads;
2733 
2734     virtio_net_set_queues(n);
2735 
2736     /* Find the first multicast entry in the saved MAC filter */
2737     for (i = 0; i < n->mac_table.in_use; i++) {
2738         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
2739             break;
2740         }
2741     }
2742     n->mac_table.first_multi = i;
2743 
2744     /* nc.link_down can't be migrated, so infer link_down according
2745      * to link status bit in n->status */
2746     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
2747     for (i = 0; i < n->max_queues; i++) {
2748         qemu_get_subqueue(n->nic, i)->link_down = link_down;
2749     }
2750 
2751     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
2752         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
2753         qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
2754                                   QEMU_CLOCK_VIRTUAL,
2755                                   virtio_net_announce_timer, n);
2756         if (n->announce_timer.round) {
2757             timer_mod(n->announce_timer.tm,
2758                       qemu_clock_get_ms(n->announce_timer.type));
2759         } else {
2760             qemu_announce_timer_del(&n->announce_timer, false);
2761         }
2762     }
2763 
2764     if (n->rss_data.enabled) {
2765         trace_virtio_net_rss_enable(n->rss_data.hash_types,
2766                                     n->rss_data.indirections_len,
2767                                     sizeof(n->rss_data.key));
2768     } else {
2769         trace_virtio_net_rss_disable();
2770     }
2771     return 0;
2772 }
2773 
2774 static int virtio_net_post_load_virtio(VirtIODevice *vdev)
2775 {
2776     VirtIONet *n = VIRTIO_NET(vdev);
2777     /*
2778      * The actual needed state is now in saved_guest_offloads,
2779      * see virtio_net_post_load_device for detail.
2780      * Restore it back and apply the desired offloads.
2781      */
2782     n->curr_guest_offloads = n->saved_guest_offloads;
2783     if (peer_has_vnet_hdr(n)) {
2784         virtio_net_apply_guest_offloads(n);
2785     }
2786 
2787     return 0;
2788 }
2789 
2790 /* tx_waiting field of a VirtIONetQueue */
2791 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = {
2792     .name = "virtio-net-queue-tx_waiting",
2793     .fields = (VMStateField[]) {
2794         VMSTATE_UINT32(tx_waiting, VirtIONetQueue),
2795         VMSTATE_END_OF_LIST()
2796    },
2797 };
2798 
2799 static bool max_queues_gt_1(void *opaque, int version_id)
2800 {
2801     return VIRTIO_NET(opaque)->max_queues > 1;
2802 }
2803 
2804 static bool has_ctrl_guest_offloads(void *opaque, int version_id)
2805 {
2806     return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque),
2807                                    VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
2808 }
2809 
2810 static bool mac_table_fits(void *opaque, int version_id)
2811 {
2812     return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES;
2813 }
2814 
2815 static bool mac_table_doesnt_fit(void *opaque, int version_id)
2816 {
2817     return !mac_table_fits(opaque, version_id);
2818 }
2819 
2820 /* This temporary type is shared by all the WITH_TMP methods
2821  * although only some fields are used by each.
2822  */
2823 struct VirtIONetMigTmp {
2824     VirtIONet      *parent;
2825     VirtIONetQueue *vqs_1;
2826     uint16_t        curr_queues_1;
2827     uint8_t         has_ufo;
2828     uint32_t        has_vnet_hdr;
2829 };
2830 
2831 /* The 2nd and subsequent tx_waiting flags are loaded later than
2832  * the 1st entry in the queues and only if there's more than one
2833  * entry.  We use the tmp mechanism to calculate a temporary
2834  * pointer and count and also validate the count.
2835  */
2836 
2837 static int virtio_net_tx_waiting_pre_save(void *opaque)
2838 {
2839     struct VirtIONetMigTmp *tmp = opaque;
2840 
2841     tmp->vqs_1 = tmp->parent->vqs + 1;
2842     tmp->curr_queues_1 = tmp->parent->curr_queues - 1;
2843     if (tmp->parent->curr_queues == 0) {
2844         tmp->curr_queues_1 = 0;
2845     }
2846 
2847     return 0;
2848 }
2849 
2850 static int virtio_net_tx_waiting_pre_load(void *opaque)
2851 {
2852     struct VirtIONetMigTmp *tmp = opaque;
2853 
2854     /* Reuse the pointer setup from save */
2855     virtio_net_tx_waiting_pre_save(opaque);
2856 
2857     if (tmp->parent->curr_queues > tmp->parent->max_queues) {
2858         error_report("virtio-net: curr_queues %x > max_queues %x",
2859             tmp->parent->curr_queues, tmp->parent->max_queues);
2860 
2861         return -EINVAL;
2862     }
2863 
2864     return 0; /* all good */
2865 }
2866 
2867 static const VMStateDescription vmstate_virtio_net_tx_waiting = {
2868     .name      = "virtio-net-tx_waiting",
2869     .pre_load  = virtio_net_tx_waiting_pre_load,
2870     .pre_save  = virtio_net_tx_waiting_pre_save,
2871     .fields    = (VMStateField[]) {
2872         VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp,
2873                                      curr_queues_1,
2874                                      vmstate_virtio_net_queue_tx_waiting,
2875                                      struct VirtIONetQueue),
2876         VMSTATE_END_OF_LIST()
2877     },
2878 };
2879 
2880 /* the 'has_ufo' flag is just tested; if the incoming stream has the
2881  * flag set we need to check that we have it
2882  */
2883 static int virtio_net_ufo_post_load(void *opaque, int version_id)
2884 {
2885     struct VirtIONetMigTmp *tmp = opaque;
2886 
2887     if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) {
2888         error_report("virtio-net: saved image requires TUN_F_UFO support");
2889         return -EINVAL;
2890     }
2891 
2892     return 0;
2893 }
2894 
2895 static int virtio_net_ufo_pre_save(void *opaque)
2896 {
2897     struct VirtIONetMigTmp *tmp = opaque;
2898 
2899     tmp->has_ufo = tmp->parent->has_ufo;
2900 
2901     return 0;
2902 }
2903 
2904 static const VMStateDescription vmstate_virtio_net_has_ufo = {
2905     .name      = "virtio-net-ufo",
2906     .post_load = virtio_net_ufo_post_load,
2907     .pre_save  = virtio_net_ufo_pre_save,
2908     .fields    = (VMStateField[]) {
2909         VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp),
2910         VMSTATE_END_OF_LIST()
2911     },
2912 };
2913 
2914 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
2915  * flag set we need to check that we have it
2916  */
2917 static int virtio_net_vnet_post_load(void *opaque, int version_id)
2918 {
2919     struct VirtIONetMigTmp *tmp = opaque;
2920 
2921     if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) {
2922         error_report("virtio-net: saved image requires vnet_hdr=on");
2923         return -EINVAL;
2924     }
2925 
2926     return 0;
2927 }
2928 
2929 static int virtio_net_vnet_pre_save(void *opaque)
2930 {
2931     struct VirtIONetMigTmp *tmp = opaque;
2932 
2933     tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr;
2934 
2935     return 0;
2936 }
2937 
2938 static const VMStateDescription vmstate_virtio_net_has_vnet = {
2939     .name      = "virtio-net-vnet",
2940     .post_load = virtio_net_vnet_post_load,
2941     .pre_save  = virtio_net_vnet_pre_save,
2942     .fields    = (VMStateField[]) {
2943         VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp),
2944         VMSTATE_END_OF_LIST()
2945     },
2946 };
2947 
2948 static bool virtio_net_rss_needed(void *opaque)
2949 {
2950     return VIRTIO_NET(opaque)->rss_data.enabled;
2951 }
2952 
2953 static const VMStateDescription vmstate_virtio_net_rss = {
2954     .name      = "virtio-net-device/rss",
2955     .version_id = 1,
2956     .minimum_version_id = 1,
2957     .needed = virtio_net_rss_needed,
2958     .fields = (VMStateField[]) {
2959         VMSTATE_BOOL(rss_data.enabled, VirtIONet),
2960         VMSTATE_BOOL(rss_data.redirect, VirtIONet),
2961         VMSTATE_BOOL(rss_data.populate_hash, VirtIONet),
2962         VMSTATE_UINT32(rss_data.hash_types, VirtIONet),
2963         VMSTATE_UINT16(rss_data.indirections_len, VirtIONet),
2964         VMSTATE_UINT16(rss_data.default_queue, VirtIONet),
2965         VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet,
2966                             VIRTIO_NET_RSS_MAX_KEY_SIZE),
2967         VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet,
2968                                     rss_data.indirections_len, 0,
2969                                     vmstate_info_uint16, uint16_t),
2970         VMSTATE_END_OF_LIST()
2971     },
2972 };
2973 
2974 static const VMStateDescription vmstate_virtio_net_device = {
2975     .name = "virtio-net-device",
2976     .version_id = VIRTIO_NET_VM_VERSION,
2977     .minimum_version_id = VIRTIO_NET_VM_VERSION,
2978     .post_load = virtio_net_post_load_device,
2979     .fields = (VMStateField[]) {
2980         VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN),
2981         VMSTATE_STRUCT_POINTER(vqs, VirtIONet,
2982                                vmstate_virtio_net_queue_tx_waiting,
2983                                VirtIONetQueue),
2984         VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet),
2985         VMSTATE_UINT16(status, VirtIONet),
2986         VMSTATE_UINT8(promisc, VirtIONet),
2987         VMSTATE_UINT8(allmulti, VirtIONet),
2988         VMSTATE_UINT32(mac_table.in_use, VirtIONet),
2989 
2990         /* Guarded pair: If it fits we load it, else we throw it away
2991          * - can happen if source has a larger MAC table.; post-load
2992          *  sets flags in this case.
2993          */
2994         VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet,
2995                                 0, mac_table_fits, mac_table.in_use,
2996                                  ETH_ALEN),
2997         VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0,
2998                                      mac_table.in_use, ETH_ALEN),
2999 
3000         /* Note: This is an array of uint32's that's always been saved as a
3001          * buffer; hold onto your endiannesses; it's actually used as a bitmap
3002          * but based on the uint.
3003          */
3004         VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3),
3005         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3006                          vmstate_virtio_net_has_vnet),
3007         VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet),
3008         VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet),
3009         VMSTATE_UINT8(alluni, VirtIONet),
3010         VMSTATE_UINT8(nomulti, VirtIONet),
3011         VMSTATE_UINT8(nouni, VirtIONet),
3012         VMSTATE_UINT8(nobcast, VirtIONet),
3013         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3014                          vmstate_virtio_net_has_ufo),
3015         VMSTATE_SINGLE_TEST(max_queues, VirtIONet, max_queues_gt_1, 0,
3016                             vmstate_info_uint16_equal, uint16_t),
3017         VMSTATE_UINT16_TEST(curr_queues, VirtIONet, max_queues_gt_1),
3018         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3019                          vmstate_virtio_net_tx_waiting),
3020         VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet,
3021                             has_ctrl_guest_offloads),
3022         VMSTATE_END_OF_LIST()
3023    },
3024     .subsections = (const VMStateDescription * []) {
3025         &vmstate_virtio_net_rss,
3026         NULL
3027     }
3028 };
3029 
3030 static NetClientInfo net_virtio_info = {
3031     .type = NET_CLIENT_DRIVER_NIC,
3032     .size = sizeof(NICState),
3033     .can_receive = virtio_net_can_receive,
3034     .receive = virtio_net_receive,
3035     .link_status_changed = virtio_net_set_link_status,
3036     .query_rx_filter = virtio_net_query_rxfilter,
3037     .announce = virtio_net_announce,
3038 };
3039 
3040 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
3041 {
3042     VirtIONet *n = VIRTIO_NET(vdev);
3043     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
3044     assert(n->vhost_started);
3045     return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
3046 }
3047 
3048 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
3049                                            bool mask)
3050 {
3051     VirtIONet *n = VIRTIO_NET(vdev);
3052     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
3053     assert(n->vhost_started);
3054     vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
3055                              vdev, idx, mask);
3056 }
3057 
3058 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
3059 {
3060     virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
3061 
3062     n->config_size = virtio_feature_get_config_size(feature_sizes,
3063                                                     host_features);
3064 }
3065 
3066 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
3067                                    const char *type)
3068 {
3069     /*
3070      * The name can be NULL, the netclient name will be type.x.
3071      */
3072     assert(type != NULL);
3073 
3074     g_free(n->netclient_name);
3075     g_free(n->netclient_type);
3076     n->netclient_name = g_strdup(name);
3077     n->netclient_type = g_strdup(type);
3078 }
3079 
3080 static bool failover_unplug_primary(VirtIONet *n, DeviceState *dev)
3081 {
3082     HotplugHandler *hotplug_ctrl;
3083     PCIDevice *pci_dev;
3084     Error *err = NULL;
3085 
3086     hotplug_ctrl = qdev_get_hotplug_handler(dev);
3087     if (hotplug_ctrl) {
3088         pci_dev = PCI_DEVICE(dev);
3089         pci_dev->partially_hotplugged = true;
3090         hotplug_handler_unplug_request(hotplug_ctrl, dev, &err);
3091         if (err) {
3092             error_report_err(err);
3093             return false;
3094         }
3095     } else {
3096         return false;
3097     }
3098     return true;
3099 }
3100 
3101 static bool failover_replug_primary(VirtIONet *n, DeviceState *dev,
3102                                     Error **errp)
3103 {
3104     Error *err = NULL;
3105     HotplugHandler *hotplug_ctrl;
3106     PCIDevice *pdev = PCI_DEVICE(dev);
3107     BusState *primary_bus;
3108 
3109     if (!pdev->partially_hotplugged) {
3110         return true;
3111     }
3112     primary_bus = dev->parent_bus;
3113     if (!primary_bus) {
3114         error_setg(errp, "virtio_net: couldn't find primary bus");
3115         return false;
3116     }
3117     qdev_set_parent_bus(dev, primary_bus, &error_abort);
3118     qatomic_set(&n->failover_primary_hidden, false);
3119     hotplug_ctrl = qdev_get_hotplug_handler(dev);
3120     if (hotplug_ctrl) {
3121         hotplug_handler_pre_plug(hotplug_ctrl, dev, &err);
3122         if (err) {
3123             goto out;
3124         }
3125         hotplug_handler_plug(hotplug_ctrl, dev, &err);
3126     }
3127 
3128 out:
3129     error_propagate(errp, err);
3130     return !err;
3131 }
3132 
3133 static void virtio_net_handle_migration_primary(VirtIONet *n, MigrationState *s)
3134 {
3135     bool should_be_hidden;
3136     Error *err = NULL;
3137     DeviceState *dev = failover_find_primary_device(n);
3138 
3139     if (!dev) {
3140         return;
3141     }
3142 
3143     should_be_hidden = qatomic_read(&n->failover_primary_hidden);
3144 
3145     if (migration_in_setup(s) && !should_be_hidden) {
3146         if (failover_unplug_primary(n, dev)) {
3147             vmstate_unregister(VMSTATE_IF(dev), qdev_get_vmsd(dev), dev);
3148             qapi_event_send_unplug_primary(dev->id);
3149             qatomic_set(&n->failover_primary_hidden, true);
3150         } else {
3151             warn_report("couldn't unplug primary device");
3152         }
3153     } else if (migration_has_failed(s)) {
3154         /* We already unplugged the device let's plug it back */
3155         if (!failover_replug_primary(n, dev, &err)) {
3156             if (err) {
3157                 error_report_err(err);
3158             }
3159         }
3160     }
3161 }
3162 
3163 static void virtio_net_migration_state_notifier(Notifier *notifier, void *data)
3164 {
3165     MigrationState *s = data;
3166     VirtIONet *n = container_of(notifier, VirtIONet, migration_state);
3167     virtio_net_handle_migration_primary(n, s);
3168 }
3169 
3170 static bool failover_hide_primary_device(DeviceListener *listener,
3171                                          QemuOpts *device_opts)
3172 {
3173     VirtIONet *n = container_of(listener, VirtIONet, primary_listener);
3174     const char *standby_id;
3175 
3176     if (!device_opts) {
3177         return false;
3178     }
3179     standby_id = qemu_opt_get(device_opts, "failover_pair_id");
3180     if (g_strcmp0(standby_id, n->netclient_name) != 0) {
3181         return false;
3182     }
3183 
3184     /* failover_primary_hidden is set during feature negotiation */
3185     return qatomic_read(&n->failover_primary_hidden);
3186 }
3187 
3188 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
3189 {
3190     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3191     VirtIONet *n = VIRTIO_NET(dev);
3192     NetClientState *nc;
3193     int i;
3194 
3195     if (n->net_conf.mtu) {
3196         n->host_features |= (1ULL << VIRTIO_NET_F_MTU);
3197     }
3198 
3199     if (n->net_conf.duplex_str) {
3200         if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) {
3201             n->net_conf.duplex = DUPLEX_HALF;
3202         } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) {
3203             n->net_conf.duplex = DUPLEX_FULL;
3204         } else {
3205             error_setg(errp, "'duplex' must be 'half' or 'full'");
3206             return;
3207         }
3208         n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3209     } else {
3210         n->net_conf.duplex = DUPLEX_UNKNOWN;
3211     }
3212 
3213     if (n->net_conf.speed < SPEED_UNKNOWN) {
3214         error_setg(errp, "'speed' must be between 0 and INT_MAX");
3215         return;
3216     }
3217     if (n->net_conf.speed >= 0) {
3218         n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3219     }
3220 
3221     if (n->failover) {
3222         n->primary_listener.hide_device = failover_hide_primary_device;
3223         qatomic_set(&n->failover_primary_hidden, true);
3224         device_listener_register(&n->primary_listener);
3225         n->migration_state.notify = virtio_net_migration_state_notifier;
3226         add_migration_state_change_notifier(&n->migration_state);
3227         n->host_features |= (1ULL << VIRTIO_NET_F_STANDBY);
3228     }
3229 
3230     virtio_net_set_config_size(n, n->host_features);
3231     virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
3232 
3233     /*
3234      * We set a lower limit on RX queue size to what it always was.
3235      * Guests that want a smaller ring can always resize it without
3236      * help from us (using virtio 1 and up).
3237      */
3238     if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
3239         n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
3240         !is_power_of_2(n->net_conf.rx_queue_size)) {
3241         error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
3242                    "must be a power of 2 between %d and %d.",
3243                    n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
3244                    VIRTQUEUE_MAX_SIZE);
3245         virtio_cleanup(vdev);
3246         return;
3247     }
3248 
3249     if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE ||
3250         n->net_conf.tx_queue_size > VIRTQUEUE_MAX_SIZE ||
3251         !is_power_of_2(n->net_conf.tx_queue_size)) {
3252         error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), "
3253                    "must be a power of 2 between %d and %d",
3254                    n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE,
3255                    VIRTQUEUE_MAX_SIZE);
3256         virtio_cleanup(vdev);
3257         return;
3258     }
3259 
3260     n->max_queues = MAX(n->nic_conf.peers.queues, 1);
3261     if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) {
3262         error_setg(errp, "Invalid number of queues (= %" PRIu32 "), "
3263                    "must be a positive integer less than %d.",
3264                    n->max_queues, (VIRTIO_QUEUE_MAX - 1) / 2);
3265         virtio_cleanup(vdev);
3266         return;
3267     }
3268     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
3269     n->curr_queues = 1;
3270     n->tx_timeout = n->net_conf.txtimer;
3271 
3272     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
3273                        && strcmp(n->net_conf.tx, "bh")) {
3274         warn_report("virtio-net: "
3275                     "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
3276                     n->net_conf.tx);
3277         error_printf("Defaulting to \"bh\"");
3278     }
3279 
3280     n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n),
3281                                     n->net_conf.tx_queue_size);
3282 
3283     for (i = 0; i < n->max_queues; i++) {
3284         virtio_net_add_queue(n, i);
3285     }
3286 
3287     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
3288     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
3289     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
3290     n->status = VIRTIO_NET_S_LINK_UP;
3291     qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
3292                               QEMU_CLOCK_VIRTUAL,
3293                               virtio_net_announce_timer, n);
3294     n->announce_timer.round = 0;
3295 
3296     if (n->netclient_type) {
3297         /*
3298          * Happen when virtio_net_set_netclient_name has been called.
3299          */
3300         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3301                               n->netclient_type, n->netclient_name, n);
3302     } else {
3303         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3304                               object_get_typename(OBJECT(dev)), dev->id, n);
3305     }
3306 
3307     peer_test_vnet_hdr(n);
3308     if (peer_has_vnet_hdr(n)) {
3309         for (i = 0; i < n->max_queues; i++) {
3310             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
3311         }
3312         n->host_hdr_len = sizeof(struct virtio_net_hdr);
3313     } else {
3314         n->host_hdr_len = 0;
3315     }
3316 
3317     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
3318 
3319     n->vqs[0].tx_waiting = 0;
3320     n->tx_burst = n->net_conf.txburst;
3321     virtio_net_set_mrg_rx_bufs(n, 0, 0, 0);
3322     n->promisc = 1; /* for compatibility */
3323 
3324     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
3325 
3326     n->vlans = g_malloc0(MAX_VLAN >> 3);
3327 
3328     nc = qemu_get_queue(n->nic);
3329     nc->rxfilter_notify_enabled = 1;
3330 
3331    if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
3332         struct virtio_net_config netcfg = {};
3333         memcpy(&netcfg.mac, &n->nic_conf.macaddr, ETH_ALEN);
3334         vhost_net_set_config(get_vhost_net(nc->peer),
3335             (uint8_t *)&netcfg, 0, ETH_ALEN, VHOST_SET_CONFIG_TYPE_MASTER);
3336     }
3337     QTAILQ_INIT(&n->rsc_chains);
3338     n->qdev = dev;
3339 
3340     net_rx_pkt_init(&n->rx_pkt, false);
3341 }
3342 
3343 static void virtio_net_device_unrealize(DeviceState *dev)
3344 {
3345     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3346     VirtIONet *n = VIRTIO_NET(dev);
3347     int i, max_queues;
3348 
3349     /* This will stop vhost backend if appropriate. */
3350     virtio_net_set_status(vdev, 0);
3351 
3352     g_free(n->netclient_name);
3353     n->netclient_name = NULL;
3354     g_free(n->netclient_type);
3355     n->netclient_type = NULL;
3356 
3357     g_free(n->mac_table.macs);
3358     g_free(n->vlans);
3359 
3360     if (n->failover) {
3361         device_listener_unregister(&n->primary_listener);
3362     }
3363 
3364     max_queues = n->multiqueue ? n->max_queues : 1;
3365     for (i = 0; i < max_queues; i++) {
3366         virtio_net_del_queue(n, i);
3367     }
3368     /* delete also control vq */
3369     virtio_del_queue(vdev, max_queues * 2);
3370     qemu_announce_timer_del(&n->announce_timer, false);
3371     g_free(n->vqs);
3372     qemu_del_nic(n->nic);
3373     virtio_net_rsc_cleanup(n);
3374     g_free(n->rss_data.indirections_table);
3375     net_rx_pkt_uninit(n->rx_pkt);
3376     virtio_cleanup(vdev);
3377 }
3378 
3379 static void virtio_net_instance_init(Object *obj)
3380 {
3381     VirtIONet *n = VIRTIO_NET(obj);
3382 
3383     /*
3384      * The default config_size is sizeof(struct virtio_net_config).
3385      * Can be overriden with virtio_net_set_config_size.
3386      */
3387     n->config_size = sizeof(struct virtio_net_config);
3388     device_add_bootindex_property(obj, &n->nic_conf.bootindex,
3389                                   "bootindex", "/ethernet-phy@0",
3390                                   DEVICE(n));
3391 }
3392 
3393 static int virtio_net_pre_save(void *opaque)
3394 {
3395     VirtIONet *n = opaque;
3396 
3397     /* At this point, backend must be stopped, otherwise
3398      * it might keep writing to memory. */
3399     assert(!n->vhost_started);
3400 
3401     return 0;
3402 }
3403 
3404 static bool primary_unplug_pending(void *opaque)
3405 {
3406     DeviceState *dev = opaque;
3407     DeviceState *primary;
3408     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3409     VirtIONet *n = VIRTIO_NET(vdev);
3410 
3411     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
3412         return false;
3413     }
3414     primary = failover_find_primary_device(n);
3415     return primary ? primary->pending_deleted_event : false;
3416 }
3417 
3418 static bool dev_unplug_pending(void *opaque)
3419 {
3420     DeviceState *dev = opaque;
3421     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3422 
3423     return vdc->primary_unplug_pending(dev);
3424 }
3425 
3426 static const VMStateDescription vmstate_virtio_net = {
3427     .name = "virtio-net",
3428     .minimum_version_id = VIRTIO_NET_VM_VERSION,
3429     .version_id = VIRTIO_NET_VM_VERSION,
3430     .fields = (VMStateField[]) {
3431         VMSTATE_VIRTIO_DEVICE,
3432         VMSTATE_END_OF_LIST()
3433     },
3434     .pre_save = virtio_net_pre_save,
3435     .dev_unplug_pending = dev_unplug_pending,
3436 };
3437 
3438 static Property virtio_net_properties[] = {
3439     DEFINE_PROP_BIT64("csum", VirtIONet, host_features,
3440                     VIRTIO_NET_F_CSUM, true),
3441     DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features,
3442                     VIRTIO_NET_F_GUEST_CSUM, true),
3443     DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
3444     DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features,
3445                     VIRTIO_NET_F_GUEST_TSO4, true),
3446     DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features,
3447                     VIRTIO_NET_F_GUEST_TSO6, true),
3448     DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features,
3449                     VIRTIO_NET_F_GUEST_ECN, true),
3450     DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features,
3451                     VIRTIO_NET_F_GUEST_UFO, true),
3452     DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features,
3453                     VIRTIO_NET_F_GUEST_ANNOUNCE, true),
3454     DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features,
3455                     VIRTIO_NET_F_HOST_TSO4, true),
3456     DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features,
3457                     VIRTIO_NET_F_HOST_TSO6, true),
3458     DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features,
3459                     VIRTIO_NET_F_HOST_ECN, true),
3460     DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features,
3461                     VIRTIO_NET_F_HOST_UFO, true),
3462     DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features,
3463                     VIRTIO_NET_F_MRG_RXBUF, true),
3464     DEFINE_PROP_BIT64("status", VirtIONet, host_features,
3465                     VIRTIO_NET_F_STATUS, true),
3466     DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features,
3467                     VIRTIO_NET_F_CTRL_VQ, true),
3468     DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features,
3469                     VIRTIO_NET_F_CTRL_RX, true),
3470     DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features,
3471                     VIRTIO_NET_F_CTRL_VLAN, true),
3472     DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features,
3473                     VIRTIO_NET_F_CTRL_RX_EXTRA, true),
3474     DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features,
3475                     VIRTIO_NET_F_CTRL_MAC_ADDR, true),
3476     DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features,
3477                     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
3478     DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
3479     DEFINE_PROP_BIT64("rss", VirtIONet, host_features,
3480                     VIRTIO_NET_F_RSS, false),
3481     DEFINE_PROP_BIT64("hash", VirtIONet, host_features,
3482                     VIRTIO_NET_F_HASH_REPORT, false),
3483     DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features,
3484                     VIRTIO_NET_F_RSC_EXT, false),
3485     DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout,
3486                        VIRTIO_NET_RSC_DEFAULT_INTERVAL),
3487     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
3488     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
3489                        TX_TIMER_INTERVAL),
3490     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
3491     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
3492     DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
3493                        VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
3494     DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size,
3495                        VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE),
3496     DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
3497     DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend,
3498                      true),
3499     DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN),
3500     DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str),
3501     DEFINE_PROP_BOOL("failover", VirtIONet, failover, false),
3502     DEFINE_PROP_END_OF_LIST(),
3503 };
3504 
3505 static void virtio_net_class_init(ObjectClass *klass, void *data)
3506 {
3507     DeviceClass *dc = DEVICE_CLASS(klass);
3508     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3509 
3510     device_class_set_props(dc, virtio_net_properties);
3511     dc->vmsd = &vmstate_virtio_net;
3512     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
3513     vdc->realize = virtio_net_device_realize;
3514     vdc->unrealize = virtio_net_device_unrealize;
3515     vdc->get_config = virtio_net_get_config;
3516     vdc->set_config = virtio_net_set_config;
3517     vdc->get_features = virtio_net_get_features;
3518     vdc->set_features = virtio_net_set_features;
3519     vdc->bad_features = virtio_net_bad_features;
3520     vdc->reset = virtio_net_reset;
3521     vdc->set_status = virtio_net_set_status;
3522     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
3523     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
3524     vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
3525     vdc->post_load = virtio_net_post_load_virtio;
3526     vdc->vmsd = &vmstate_virtio_net_device;
3527     vdc->primary_unplug_pending = primary_unplug_pending;
3528 }
3529 
3530 static const TypeInfo virtio_net_info = {
3531     .name = TYPE_VIRTIO_NET,
3532     .parent = TYPE_VIRTIO_DEVICE,
3533     .instance_size = sizeof(VirtIONet),
3534     .instance_init = virtio_net_instance_init,
3535     .class_init = virtio_net_class_init,
3536 };
3537 
3538 static void virtio_register_types(void)
3539 {
3540     type_register_static(&virtio_net_info);
3541 }
3542 
3543 type_init(virtio_register_types)
3544