xref: /openbmc/qemu/hw/net/virtio-net.c (revision 44602af8)
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/log.h"
18 #include "qemu/main-loop.h"
19 #include "qemu/module.h"
20 #include "hw/virtio/virtio.h"
21 #include "net/net.h"
22 #include "net/checksum.h"
23 #include "net/tap.h"
24 #include "qemu/error-report.h"
25 #include "qemu/timer.h"
26 #include "qemu/option.h"
27 #include "qemu/option_int.h"
28 #include "qemu/config-file.h"
29 #include "qapi/qmp/qdict.h"
30 #include "hw/virtio/virtio-net.h"
31 #include "net/vhost_net.h"
32 #include "net/announce.h"
33 #include "hw/virtio/virtio-bus.h"
34 #include "qapi/error.h"
35 #include "qapi/qapi-events-net.h"
36 #include "hw/qdev-properties.h"
37 #include "qapi/qapi-types-migration.h"
38 #include "qapi/qapi-events-migration.h"
39 #include "hw/virtio/virtio-access.h"
40 #include "migration/misc.h"
41 #include "standard-headers/linux/ethtool.h"
42 #include "sysemu/sysemu.h"
43 #include "trace.h"
44 #include "monitor/qdev.h"
45 #include "hw/pci/pci.h"
46 #include "net_rx_pkt.h"
47 #include "hw/virtio/vhost.h"
48 #include "sysemu/qtest.h"
49 
50 #define VIRTIO_NET_VM_VERSION    11
51 
52 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
53 
54 /* previously fixed value */
55 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
56 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
57 
58 /* for now, only allow larger queue_pairs; with virtio-1, guest can downsize */
59 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
60 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
61 
62 #define VIRTIO_NET_IP4_ADDR_SIZE   8        /* ipv4 saddr + daddr */
63 
64 #define VIRTIO_NET_TCP_FLAG         0x3F
65 #define VIRTIO_NET_TCP_HDR_LENGTH   0xF000
66 
67 /* IPv4 max payload, 16 bits in the header */
68 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
69 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
70 
71 /* header length value in ip header without option */
72 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
73 
74 #define VIRTIO_NET_IP6_ADDR_SIZE   32      /* ipv6 saddr + daddr */
75 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
76 
77 /* Purge coalesced packets timer interval, This value affects the performance
78    a lot, and should be tuned carefully, '300000'(300us) is the recommended
79    value to pass the WHQL test, '50000' can gain 2x netperf throughput with
80    tso/gso/gro 'off'. */
81 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
82 
83 #define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \
84                                          VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \
85                                          VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \
86                                          VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \
87                                          VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \
88                                          VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \
89                                          VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \
90                                          VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \
91                                          VIRTIO_NET_RSS_HASH_TYPE_UDP_EX)
92 
93 static const VirtIOFeature feature_sizes[] = {
94     {.flags = 1ULL << VIRTIO_NET_F_MAC,
95      .end = endof(struct virtio_net_config, mac)},
96     {.flags = 1ULL << VIRTIO_NET_F_STATUS,
97      .end = endof(struct virtio_net_config, status)},
98     {.flags = 1ULL << VIRTIO_NET_F_MQ,
99      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
100     {.flags = 1ULL << VIRTIO_NET_F_MTU,
101      .end = endof(struct virtio_net_config, mtu)},
102     {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX,
103      .end = endof(struct virtio_net_config, duplex)},
104     {.flags = (1ULL << VIRTIO_NET_F_RSS) | (1ULL << VIRTIO_NET_F_HASH_REPORT),
105      .end = endof(struct virtio_net_config, supported_hash_types)},
106     {}
107 };
108 
109 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
110 {
111     VirtIONet *n = qemu_get_nic_opaque(nc);
112 
113     return &n->vqs[nc->queue_index];
114 }
115 
116 static int vq2q(int queue_index)
117 {
118     return queue_index / 2;
119 }
120 
121 /* TODO
122  * - we could suppress RX interrupt if we were so inclined.
123  */
124 
125 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
126 {
127     VirtIONet *n = VIRTIO_NET(vdev);
128     struct virtio_net_config netcfg;
129     NetClientState *nc = qemu_get_queue(n->nic);
130     static const MACAddr zero = { .a = { 0, 0, 0, 0, 0, 0 } };
131 
132     int ret = 0;
133     memset(&netcfg, 0 , sizeof(struct virtio_net_config));
134     virtio_stw_p(vdev, &netcfg.status, n->status);
135     virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queue_pairs);
136     virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
137     memcpy(netcfg.mac, n->mac, ETH_ALEN);
138     virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed);
139     netcfg.duplex = n->net_conf.duplex;
140     netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE;
141     virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length,
142                  virtio_host_has_feature(vdev, VIRTIO_NET_F_RSS) ?
143                  VIRTIO_NET_RSS_MAX_TABLE_LEN : 1);
144     virtio_stl_p(vdev, &netcfg.supported_hash_types,
145                  VIRTIO_NET_RSS_SUPPORTED_HASHES);
146     memcpy(config, &netcfg, n->config_size);
147 
148     /*
149      * Is this VDPA? No peer means not VDPA: there's no way to
150      * disconnect/reconnect a VDPA peer.
151      */
152     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
153         ret = vhost_net_get_config(get_vhost_net(nc->peer), (uint8_t *)&netcfg,
154                                    n->config_size);
155         if (ret != -1) {
156             /*
157              * Some NIC/kernel combinations present 0 as the mac address.  As
158              * that is not a legal address, try to proceed with the
159              * address from the QEMU command line in the hope that the
160              * address has been configured correctly elsewhere - just not
161              * reported by the device.
162              */
163             if (memcmp(&netcfg.mac, &zero, sizeof(zero)) == 0) {
164                 info_report("Zero hardware mac address detected. Ignoring.");
165                 memcpy(netcfg.mac, n->mac, ETH_ALEN);
166             }
167             memcpy(config, &netcfg, n->config_size);
168         }
169     }
170 }
171 
172 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
173 {
174     VirtIONet *n = VIRTIO_NET(vdev);
175     struct virtio_net_config netcfg = {};
176     NetClientState *nc = qemu_get_queue(n->nic);
177 
178     memcpy(&netcfg, config, n->config_size);
179 
180     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
181         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
182         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
183         memcpy(n->mac, netcfg.mac, ETH_ALEN);
184         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
185     }
186 
187     /*
188      * Is this VDPA? No peer means not VDPA: there's no way to
189      * disconnect/reconnect a VDPA peer.
190      */
191     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
192         vhost_net_set_config(get_vhost_net(nc->peer),
193                              (uint8_t *)&netcfg, 0, n->config_size,
194                              VHOST_SET_CONFIG_TYPE_MASTER);
195       }
196 }
197 
198 static bool virtio_net_started(VirtIONet *n, uint8_t status)
199 {
200     VirtIODevice *vdev = VIRTIO_DEVICE(n);
201     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
202         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
203 }
204 
205 static void virtio_net_announce_notify(VirtIONet *net)
206 {
207     VirtIODevice *vdev = VIRTIO_DEVICE(net);
208     trace_virtio_net_announce_notify();
209 
210     net->status |= VIRTIO_NET_S_ANNOUNCE;
211     virtio_notify_config(vdev);
212 }
213 
214 static void virtio_net_announce_timer(void *opaque)
215 {
216     VirtIONet *n = opaque;
217     trace_virtio_net_announce_timer(n->announce_timer.round);
218 
219     n->announce_timer.round--;
220     virtio_net_announce_notify(n);
221 }
222 
223 static void virtio_net_announce(NetClientState *nc)
224 {
225     VirtIONet *n = qemu_get_nic_opaque(nc);
226     VirtIODevice *vdev = VIRTIO_DEVICE(n);
227 
228     /*
229      * Make sure the virtio migration announcement timer isn't running
230      * If it is, let it trigger announcement so that we do not cause
231      * confusion.
232      */
233     if (n->announce_timer.round) {
234         return;
235     }
236 
237     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
238         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
239             virtio_net_announce_notify(n);
240     }
241 }
242 
243 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
244 {
245     VirtIODevice *vdev = VIRTIO_DEVICE(n);
246     NetClientState *nc = qemu_get_queue(n->nic);
247     int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
248     int cvq = virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) ?
249               n->max_ncs - n->max_queue_pairs : 0;
250 
251     if (!get_vhost_net(nc->peer)) {
252         return;
253     }
254 
255     if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
256         !!n->vhost_started) {
257         return;
258     }
259     if (!n->vhost_started) {
260         int r, i;
261 
262         if (n->needs_vnet_hdr_swap) {
263             error_report("backend does not support %s vnet headers; "
264                          "falling back on userspace virtio",
265                          virtio_is_big_endian(vdev) ? "BE" : "LE");
266             return;
267         }
268 
269         /* Any packets outstanding? Purge them to avoid touching rings
270          * when vhost is running.
271          */
272         for (i = 0;  i < queue_pairs; i++) {
273             NetClientState *qnc = qemu_get_subqueue(n->nic, i);
274 
275             /* Purge both directions: TX and RX. */
276             qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
277             qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
278         }
279 
280         if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
281             r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
282             if (r < 0) {
283                 error_report("%uBytes MTU not supported by the backend",
284                              n->net_conf.mtu);
285 
286                 return;
287             }
288         }
289 
290         n->vhost_started = 1;
291         r = vhost_net_start(vdev, n->nic->ncs, queue_pairs, cvq);
292         if (r < 0) {
293             error_report("unable to start vhost net: %d: "
294                          "falling back on userspace virtio", -r);
295             n->vhost_started = 0;
296         }
297     } else {
298         vhost_net_stop(vdev, n->nic->ncs, queue_pairs, cvq);
299         n->vhost_started = 0;
300     }
301 }
302 
303 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
304                                           NetClientState *peer,
305                                           bool enable)
306 {
307     if (virtio_is_big_endian(vdev)) {
308         return qemu_set_vnet_be(peer, enable);
309     } else {
310         return qemu_set_vnet_le(peer, enable);
311     }
312 }
313 
314 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
315                                        int queue_pairs, bool enable)
316 {
317     int i;
318 
319     for (i = 0; i < queue_pairs; i++) {
320         if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
321             enable) {
322             while (--i >= 0) {
323                 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
324             }
325 
326             return true;
327         }
328     }
329 
330     return false;
331 }
332 
333 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
334 {
335     VirtIODevice *vdev = VIRTIO_DEVICE(n);
336     int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
337 
338     if (virtio_net_started(n, status)) {
339         /* Before using the device, we tell the network backend about the
340          * endianness to use when parsing vnet headers. If the backend
341          * can't do it, we fallback onto fixing the headers in the core
342          * virtio-net code.
343          */
344         n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
345                                                             queue_pairs, true);
346     } else if (virtio_net_started(n, vdev->status)) {
347         /* After using the device, we need to reset the network backend to
348          * the default (guest native endianness), otherwise the guest may
349          * lose network connectivity if it is rebooted into a different
350          * endianness.
351          */
352         virtio_net_set_vnet_endian(vdev, n->nic->ncs, queue_pairs, false);
353     }
354 }
355 
356 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq)
357 {
358     unsigned int dropped = virtqueue_drop_all(vq);
359     if (dropped) {
360         virtio_notify(vdev, vq);
361     }
362 }
363 
364 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
365 {
366     VirtIONet *n = VIRTIO_NET(vdev);
367     VirtIONetQueue *q;
368     int i;
369     uint8_t queue_status;
370 
371     virtio_net_vnet_endian_status(n, status);
372     virtio_net_vhost_status(n, status);
373 
374     for (i = 0; i < n->max_queue_pairs; i++) {
375         NetClientState *ncs = qemu_get_subqueue(n->nic, i);
376         bool queue_started;
377         q = &n->vqs[i];
378 
379         if ((!n->multiqueue && i != 0) || i >= n->curr_queue_pairs) {
380             queue_status = 0;
381         } else {
382             queue_status = status;
383         }
384         queue_started =
385             virtio_net_started(n, queue_status) && !n->vhost_started;
386 
387         if (queue_started) {
388             qemu_flush_queued_packets(ncs);
389         }
390 
391         if (!q->tx_waiting) {
392             continue;
393         }
394 
395         if (queue_started) {
396             if (q->tx_timer) {
397                 timer_mod(q->tx_timer,
398                                qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
399             } else {
400                 qemu_bh_schedule(q->tx_bh);
401             }
402         } else {
403             if (q->tx_timer) {
404                 timer_del(q->tx_timer);
405             } else {
406                 qemu_bh_cancel(q->tx_bh);
407             }
408             if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 &&
409                 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) &&
410                 vdev->vm_running) {
411                 /* if tx is waiting we are likely have some packets in tx queue
412                  * and disabled notification */
413                 q->tx_waiting = 0;
414                 virtio_queue_set_notification(q->tx_vq, 1);
415                 virtio_net_drop_tx_queue_data(vdev, q->tx_vq);
416             }
417         }
418     }
419 }
420 
421 static void virtio_net_set_link_status(NetClientState *nc)
422 {
423     VirtIONet *n = qemu_get_nic_opaque(nc);
424     VirtIODevice *vdev = VIRTIO_DEVICE(n);
425     uint16_t old_status = n->status;
426 
427     if (nc->link_down)
428         n->status &= ~VIRTIO_NET_S_LINK_UP;
429     else
430         n->status |= VIRTIO_NET_S_LINK_UP;
431 
432     if (n->status != old_status)
433         virtio_notify_config(vdev);
434 
435     virtio_net_set_status(vdev, vdev->status);
436 }
437 
438 static void rxfilter_notify(NetClientState *nc)
439 {
440     VirtIONet *n = qemu_get_nic_opaque(nc);
441 
442     if (nc->rxfilter_notify_enabled) {
443         char *path = object_get_canonical_path(OBJECT(n->qdev));
444         qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
445                                               n->netclient_name, path);
446         g_free(path);
447 
448         /* disable event notification to avoid events flooding */
449         nc->rxfilter_notify_enabled = 0;
450     }
451 }
452 
453 static intList *get_vlan_table(VirtIONet *n)
454 {
455     intList *list;
456     int i, j;
457 
458     list = NULL;
459     for (i = 0; i < MAX_VLAN >> 5; i++) {
460         for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
461             if (n->vlans[i] & (1U << j)) {
462                 QAPI_LIST_PREPEND(list, (i << 5) + j);
463             }
464         }
465     }
466 
467     return list;
468 }
469 
470 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
471 {
472     VirtIONet *n = qemu_get_nic_opaque(nc);
473     VirtIODevice *vdev = VIRTIO_DEVICE(n);
474     RxFilterInfo *info;
475     strList *str_list;
476     int i;
477 
478     info = g_malloc0(sizeof(*info));
479     info->name = g_strdup(nc->name);
480     info->promiscuous = n->promisc;
481 
482     if (n->nouni) {
483         info->unicast = RX_STATE_NONE;
484     } else if (n->alluni) {
485         info->unicast = RX_STATE_ALL;
486     } else {
487         info->unicast = RX_STATE_NORMAL;
488     }
489 
490     if (n->nomulti) {
491         info->multicast = RX_STATE_NONE;
492     } else if (n->allmulti) {
493         info->multicast = RX_STATE_ALL;
494     } else {
495         info->multicast = RX_STATE_NORMAL;
496     }
497 
498     info->broadcast_allowed = n->nobcast;
499     info->multicast_overflow = n->mac_table.multi_overflow;
500     info->unicast_overflow = n->mac_table.uni_overflow;
501 
502     info->main_mac = qemu_mac_strdup_printf(n->mac);
503 
504     str_list = NULL;
505     for (i = 0; i < n->mac_table.first_multi; i++) {
506         QAPI_LIST_PREPEND(str_list,
507                       qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN));
508     }
509     info->unicast_table = str_list;
510 
511     str_list = NULL;
512     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
513         QAPI_LIST_PREPEND(str_list,
514                       qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN));
515     }
516     info->multicast_table = str_list;
517     info->vlan_table = get_vlan_table(n);
518 
519     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
520         info->vlan = RX_STATE_ALL;
521     } else if (!info->vlan_table) {
522         info->vlan = RX_STATE_NONE;
523     } else {
524         info->vlan = RX_STATE_NORMAL;
525     }
526 
527     /* enable event notification after query */
528     nc->rxfilter_notify_enabled = 1;
529 
530     return info;
531 }
532 
533 static void virtio_net_reset(VirtIODevice *vdev)
534 {
535     VirtIONet *n = VIRTIO_NET(vdev);
536     int i;
537 
538     /* Reset back to compatibility mode */
539     n->promisc = 1;
540     n->allmulti = 0;
541     n->alluni = 0;
542     n->nomulti = 0;
543     n->nouni = 0;
544     n->nobcast = 0;
545     /* multiqueue is disabled by default */
546     n->curr_queue_pairs = 1;
547     timer_del(n->announce_timer.tm);
548     n->announce_timer.round = 0;
549     n->status &= ~VIRTIO_NET_S_ANNOUNCE;
550 
551     /* Flush any MAC and VLAN filter table state */
552     n->mac_table.in_use = 0;
553     n->mac_table.first_multi = 0;
554     n->mac_table.multi_overflow = 0;
555     n->mac_table.uni_overflow = 0;
556     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
557     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
558     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
559     memset(n->vlans, 0, MAX_VLAN >> 3);
560 
561     /* Flush any async TX */
562     for (i = 0;  i < n->max_queue_pairs; i++) {
563         NetClientState *nc = qemu_get_subqueue(n->nic, i);
564 
565         if (nc->peer) {
566             qemu_flush_or_purge_queued_packets(nc->peer, true);
567             assert(!virtio_net_get_subqueue(nc)->async_tx.elem);
568         }
569     }
570 }
571 
572 static void peer_test_vnet_hdr(VirtIONet *n)
573 {
574     NetClientState *nc = qemu_get_queue(n->nic);
575     if (!nc->peer) {
576         return;
577     }
578 
579     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
580 }
581 
582 static int peer_has_vnet_hdr(VirtIONet *n)
583 {
584     return n->has_vnet_hdr;
585 }
586 
587 static int peer_has_ufo(VirtIONet *n)
588 {
589     if (!peer_has_vnet_hdr(n))
590         return 0;
591 
592     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
593 
594     return n->has_ufo;
595 }
596 
597 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
598                                        int version_1, int hash_report)
599 {
600     int i;
601     NetClientState *nc;
602 
603     n->mergeable_rx_bufs = mergeable_rx_bufs;
604 
605     if (version_1) {
606         n->guest_hdr_len = hash_report ?
607             sizeof(struct virtio_net_hdr_v1_hash) :
608             sizeof(struct virtio_net_hdr_mrg_rxbuf);
609         n->rss_data.populate_hash = !!hash_report;
610     } else {
611         n->guest_hdr_len = n->mergeable_rx_bufs ?
612             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
613             sizeof(struct virtio_net_hdr);
614     }
615 
616     for (i = 0; i < n->max_queue_pairs; i++) {
617         nc = qemu_get_subqueue(n->nic, i);
618 
619         if (peer_has_vnet_hdr(n) &&
620             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
621             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
622             n->host_hdr_len = n->guest_hdr_len;
623         }
624     }
625 }
626 
627 static int virtio_net_max_tx_queue_size(VirtIONet *n)
628 {
629     NetClientState *peer = n->nic_conf.peers.ncs[0];
630 
631     /*
632      * Backends other than vhost-user or vhost-vdpa don't support max queue
633      * size.
634      */
635     if (!peer) {
636         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
637     }
638 
639     switch(peer->info->type) {
640     case NET_CLIENT_DRIVER_VHOST_USER:
641     case NET_CLIENT_DRIVER_VHOST_VDPA:
642         return VIRTQUEUE_MAX_SIZE;
643     default:
644         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
645     };
646 }
647 
648 static int peer_attach(VirtIONet *n, int index)
649 {
650     NetClientState *nc = qemu_get_subqueue(n->nic, index);
651 
652     if (!nc->peer) {
653         return 0;
654     }
655 
656     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
657         vhost_set_vring_enable(nc->peer, 1);
658     }
659 
660     if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
661         return 0;
662     }
663 
664     if (n->max_queue_pairs == 1) {
665         return 0;
666     }
667 
668     return tap_enable(nc->peer);
669 }
670 
671 static int peer_detach(VirtIONet *n, int index)
672 {
673     NetClientState *nc = qemu_get_subqueue(n->nic, index);
674 
675     if (!nc->peer) {
676         return 0;
677     }
678 
679     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
680         vhost_set_vring_enable(nc->peer, 0);
681     }
682 
683     if (nc->peer->info->type !=  NET_CLIENT_DRIVER_TAP) {
684         return 0;
685     }
686 
687     return tap_disable(nc->peer);
688 }
689 
690 static void virtio_net_set_queue_pairs(VirtIONet *n)
691 {
692     int i;
693     int r;
694 
695     if (n->nic->peer_deleted) {
696         return;
697     }
698 
699     for (i = 0; i < n->max_queue_pairs; i++) {
700         if (i < n->curr_queue_pairs) {
701             r = peer_attach(n, i);
702             assert(!r);
703         } else {
704             r = peer_detach(n, i);
705             assert(!r);
706         }
707     }
708 }
709 
710 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
711 
712 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
713                                         Error **errp)
714 {
715     VirtIONet *n = VIRTIO_NET(vdev);
716     NetClientState *nc = qemu_get_queue(n->nic);
717 
718     /* Firstly sync all virtio-net possible supported features */
719     features |= n->host_features;
720 
721     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
722 
723     if (!peer_has_vnet_hdr(n)) {
724         virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
725         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
726         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
727         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
728 
729         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
730         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
731         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
732         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
733 
734         virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT);
735     }
736 
737     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
738         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
739         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
740     }
741 
742     if (!get_vhost_net(nc->peer)) {
743         return features;
744     }
745 
746     if (!ebpf_rss_is_loaded(&n->ebpf_rss)) {
747         virtio_clear_feature(&features, VIRTIO_NET_F_RSS);
748     }
749     features = vhost_net_get_features(get_vhost_net(nc->peer), features);
750     vdev->backend_features = features;
751 
752     if (n->mtu_bypass_backend &&
753             (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) {
754         features |= (1ULL << VIRTIO_NET_F_MTU);
755     }
756 
757     return features;
758 }
759 
760 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
761 {
762     uint64_t features = 0;
763 
764     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
765      * but also these: */
766     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
767     virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
768     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
769     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
770     virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
771 
772     return features;
773 }
774 
775 static void virtio_net_apply_guest_offloads(VirtIONet *n)
776 {
777     qemu_set_offload(qemu_get_queue(n->nic)->peer,
778             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
779             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
780             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
781             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
782             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
783 }
784 
785 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
786 {
787     static const uint64_t guest_offloads_mask =
788         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
789         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
790         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
791         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
792         (1ULL << VIRTIO_NET_F_GUEST_UFO);
793 
794     return guest_offloads_mask & features;
795 }
796 
797 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
798 {
799     VirtIODevice *vdev = VIRTIO_DEVICE(n);
800     return virtio_net_guest_offloads_by_features(vdev->guest_features);
801 }
802 
803 typedef struct {
804     VirtIONet *n;
805     DeviceState *dev;
806 } FailoverDevice;
807 
808 /**
809  * Set the failover primary device
810  *
811  * @opaque: FailoverId to setup
812  * @opts: opts for device we are handling
813  * @errp: returns an error if this function fails
814  */
815 static int failover_set_primary(DeviceState *dev, void *opaque)
816 {
817     FailoverDevice *fdev = opaque;
818     PCIDevice *pci_dev = (PCIDevice *)
819         object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE);
820 
821     if (!pci_dev) {
822         return 0;
823     }
824 
825     if (!g_strcmp0(pci_dev->failover_pair_id, fdev->n->netclient_name)) {
826         fdev->dev = dev;
827         return 1;
828     }
829 
830     return 0;
831 }
832 
833 /**
834  * Find the primary device for this failover virtio-net
835  *
836  * @n: VirtIONet device
837  * @errp: returns an error if this function fails
838  */
839 static DeviceState *failover_find_primary_device(VirtIONet *n)
840 {
841     FailoverDevice fdev = {
842         .n = n,
843     };
844 
845     qbus_walk_children(sysbus_get_default(), failover_set_primary, NULL,
846                        NULL, NULL, &fdev);
847     return fdev.dev;
848 }
849 
850 static void failover_add_primary(VirtIONet *n, Error **errp)
851 {
852     Error *err = NULL;
853     DeviceState *dev = failover_find_primary_device(n);
854 
855     if (dev) {
856         return;
857     }
858 
859     if (!n->primary_opts) {
860         error_setg(errp, "Primary device not found");
861         error_append_hint(errp, "Virtio-net failover will not work. Make "
862                           "sure primary device has parameter"
863                           " failover_pair_id=%s\n", n->netclient_name);
864         return;
865     }
866 
867     dev = qdev_device_add_from_qdict(n->primary_opts,
868                                      n->primary_opts_from_json,
869                                      &err);
870     if (err) {
871         qobject_unref(n->primary_opts);
872         n->primary_opts = NULL;
873     } else {
874         object_unref(OBJECT(dev));
875     }
876     error_propagate(errp, err);
877 }
878 
879 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
880 {
881     VirtIONet *n = VIRTIO_NET(vdev);
882     Error *err = NULL;
883     int i;
884 
885     if (n->mtu_bypass_backend &&
886             !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) {
887         features &= ~(1ULL << VIRTIO_NET_F_MTU);
888     }
889 
890     virtio_net_set_multiqueue(n,
891                               virtio_has_feature(features, VIRTIO_NET_F_RSS) ||
892                               virtio_has_feature(features, VIRTIO_NET_F_MQ));
893 
894     virtio_net_set_mrg_rx_bufs(n,
895                                virtio_has_feature(features,
896                                                   VIRTIO_NET_F_MRG_RXBUF),
897                                virtio_has_feature(features,
898                                                   VIRTIO_F_VERSION_1),
899                                virtio_has_feature(features,
900                                                   VIRTIO_NET_F_HASH_REPORT));
901 
902     n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
903         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4);
904     n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
905         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6);
906     n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS);
907 
908     if (n->has_vnet_hdr) {
909         n->curr_guest_offloads =
910             virtio_net_guest_offloads_by_features(features);
911         virtio_net_apply_guest_offloads(n);
912     }
913 
914     for (i = 0;  i < n->max_queue_pairs; i++) {
915         NetClientState *nc = qemu_get_subqueue(n->nic, i);
916 
917         if (!get_vhost_net(nc->peer)) {
918             continue;
919         }
920         vhost_net_ack_features(get_vhost_net(nc->peer), features);
921     }
922 
923     if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
924         memset(n->vlans, 0, MAX_VLAN >> 3);
925     } else {
926         memset(n->vlans, 0xff, MAX_VLAN >> 3);
927     }
928 
929     if (virtio_has_feature(features, VIRTIO_NET_F_STANDBY)) {
930         qapi_event_send_failover_negotiated(n->netclient_name);
931         qatomic_set(&n->failover_primary_hidden, false);
932         failover_add_primary(n, &err);
933         if (err) {
934             if (!qtest_enabled()) {
935                 warn_report_err(err);
936             } else {
937                 error_free(err);
938             }
939         }
940     }
941 }
942 
943 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
944                                      struct iovec *iov, unsigned int iov_cnt)
945 {
946     uint8_t on;
947     size_t s;
948     NetClientState *nc = qemu_get_queue(n->nic);
949 
950     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
951     if (s != sizeof(on)) {
952         return VIRTIO_NET_ERR;
953     }
954 
955     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
956         n->promisc = on;
957     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
958         n->allmulti = on;
959     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
960         n->alluni = on;
961     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
962         n->nomulti = on;
963     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
964         n->nouni = on;
965     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
966         n->nobcast = on;
967     } else {
968         return VIRTIO_NET_ERR;
969     }
970 
971     rxfilter_notify(nc);
972 
973     return VIRTIO_NET_OK;
974 }
975 
976 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
977                                      struct iovec *iov, unsigned int iov_cnt)
978 {
979     VirtIODevice *vdev = VIRTIO_DEVICE(n);
980     uint64_t offloads;
981     size_t s;
982 
983     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
984         return VIRTIO_NET_ERR;
985     }
986 
987     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
988     if (s != sizeof(offloads)) {
989         return VIRTIO_NET_ERR;
990     }
991 
992     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
993         uint64_t supported_offloads;
994 
995         offloads = virtio_ldq_p(vdev, &offloads);
996 
997         if (!n->has_vnet_hdr) {
998             return VIRTIO_NET_ERR;
999         }
1000 
1001         n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
1002             virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4);
1003         n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
1004             virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6);
1005         virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT);
1006 
1007         supported_offloads = virtio_net_supported_guest_offloads(n);
1008         if (offloads & ~supported_offloads) {
1009             return VIRTIO_NET_ERR;
1010         }
1011 
1012         n->curr_guest_offloads = offloads;
1013         virtio_net_apply_guest_offloads(n);
1014 
1015         return VIRTIO_NET_OK;
1016     } else {
1017         return VIRTIO_NET_ERR;
1018     }
1019 }
1020 
1021 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
1022                                  struct iovec *iov, unsigned int iov_cnt)
1023 {
1024     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1025     struct virtio_net_ctrl_mac mac_data;
1026     size_t s;
1027     NetClientState *nc = qemu_get_queue(n->nic);
1028 
1029     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
1030         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
1031             return VIRTIO_NET_ERR;
1032         }
1033         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
1034         assert(s == sizeof(n->mac));
1035         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
1036         rxfilter_notify(nc);
1037 
1038         return VIRTIO_NET_OK;
1039     }
1040 
1041     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
1042         return VIRTIO_NET_ERR;
1043     }
1044 
1045     int in_use = 0;
1046     int first_multi = 0;
1047     uint8_t uni_overflow = 0;
1048     uint8_t multi_overflow = 0;
1049     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1050 
1051     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1052                    sizeof(mac_data.entries));
1053     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1054     if (s != sizeof(mac_data.entries)) {
1055         goto error;
1056     }
1057     iov_discard_front(&iov, &iov_cnt, s);
1058 
1059     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
1060         goto error;
1061     }
1062 
1063     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
1064         s = iov_to_buf(iov, iov_cnt, 0, macs,
1065                        mac_data.entries * ETH_ALEN);
1066         if (s != mac_data.entries * ETH_ALEN) {
1067             goto error;
1068         }
1069         in_use += mac_data.entries;
1070     } else {
1071         uni_overflow = 1;
1072     }
1073 
1074     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
1075 
1076     first_multi = in_use;
1077 
1078     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1079                    sizeof(mac_data.entries));
1080     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1081     if (s != sizeof(mac_data.entries)) {
1082         goto error;
1083     }
1084 
1085     iov_discard_front(&iov, &iov_cnt, s);
1086 
1087     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
1088         goto error;
1089     }
1090 
1091     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
1092         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
1093                        mac_data.entries * ETH_ALEN);
1094         if (s != mac_data.entries * ETH_ALEN) {
1095             goto error;
1096         }
1097         in_use += mac_data.entries;
1098     } else {
1099         multi_overflow = 1;
1100     }
1101 
1102     n->mac_table.in_use = in_use;
1103     n->mac_table.first_multi = first_multi;
1104     n->mac_table.uni_overflow = uni_overflow;
1105     n->mac_table.multi_overflow = multi_overflow;
1106     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
1107     g_free(macs);
1108     rxfilter_notify(nc);
1109 
1110     return VIRTIO_NET_OK;
1111 
1112 error:
1113     g_free(macs);
1114     return VIRTIO_NET_ERR;
1115 }
1116 
1117 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
1118                                         struct iovec *iov, unsigned int iov_cnt)
1119 {
1120     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1121     uint16_t vid;
1122     size_t s;
1123     NetClientState *nc = qemu_get_queue(n->nic);
1124 
1125     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
1126     vid = virtio_lduw_p(vdev, &vid);
1127     if (s != sizeof(vid)) {
1128         return VIRTIO_NET_ERR;
1129     }
1130 
1131     if (vid >= MAX_VLAN)
1132         return VIRTIO_NET_ERR;
1133 
1134     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
1135         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
1136     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
1137         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
1138     else
1139         return VIRTIO_NET_ERR;
1140 
1141     rxfilter_notify(nc);
1142 
1143     return VIRTIO_NET_OK;
1144 }
1145 
1146 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
1147                                       struct iovec *iov, unsigned int iov_cnt)
1148 {
1149     trace_virtio_net_handle_announce(n->announce_timer.round);
1150     if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
1151         n->status & VIRTIO_NET_S_ANNOUNCE) {
1152         n->status &= ~VIRTIO_NET_S_ANNOUNCE;
1153         if (n->announce_timer.round) {
1154             qemu_announce_timer_step(&n->announce_timer);
1155         }
1156         return VIRTIO_NET_OK;
1157     } else {
1158         return VIRTIO_NET_ERR;
1159     }
1160 }
1161 
1162 static void virtio_net_detach_epbf_rss(VirtIONet *n);
1163 
1164 static void virtio_net_disable_rss(VirtIONet *n)
1165 {
1166     if (n->rss_data.enabled) {
1167         trace_virtio_net_rss_disable();
1168     }
1169     n->rss_data.enabled = false;
1170 
1171     virtio_net_detach_epbf_rss(n);
1172 }
1173 
1174 static bool virtio_net_attach_ebpf_to_backend(NICState *nic, int prog_fd)
1175 {
1176     NetClientState *nc = qemu_get_peer(qemu_get_queue(nic), 0);
1177     if (nc == NULL || nc->info->set_steering_ebpf == NULL) {
1178         return false;
1179     }
1180 
1181     return nc->info->set_steering_ebpf(nc, prog_fd);
1182 }
1183 
1184 static void rss_data_to_rss_config(struct VirtioNetRssData *data,
1185                                    struct EBPFRSSConfig *config)
1186 {
1187     config->redirect = data->redirect;
1188     config->populate_hash = data->populate_hash;
1189     config->hash_types = data->hash_types;
1190     config->indirections_len = data->indirections_len;
1191     config->default_queue = data->default_queue;
1192 }
1193 
1194 static bool virtio_net_attach_epbf_rss(VirtIONet *n)
1195 {
1196     struct EBPFRSSConfig config = {};
1197 
1198     if (!ebpf_rss_is_loaded(&n->ebpf_rss)) {
1199         return false;
1200     }
1201 
1202     rss_data_to_rss_config(&n->rss_data, &config);
1203 
1204     if (!ebpf_rss_set_all(&n->ebpf_rss, &config,
1205                           n->rss_data.indirections_table, n->rss_data.key)) {
1206         return false;
1207     }
1208 
1209     if (!virtio_net_attach_ebpf_to_backend(n->nic, n->ebpf_rss.program_fd)) {
1210         return false;
1211     }
1212 
1213     return true;
1214 }
1215 
1216 static void virtio_net_detach_epbf_rss(VirtIONet *n)
1217 {
1218     virtio_net_attach_ebpf_to_backend(n->nic, -1);
1219 }
1220 
1221 static bool virtio_net_load_ebpf(VirtIONet *n)
1222 {
1223     if (!virtio_net_attach_ebpf_to_backend(n->nic, -1)) {
1224         /* backend does't support steering ebpf */
1225         return false;
1226     }
1227 
1228     return ebpf_rss_load(&n->ebpf_rss);
1229 }
1230 
1231 static void virtio_net_unload_ebpf(VirtIONet *n)
1232 {
1233     virtio_net_attach_ebpf_to_backend(n->nic, -1);
1234     ebpf_rss_unload(&n->ebpf_rss);
1235 }
1236 
1237 static uint16_t virtio_net_handle_rss(VirtIONet *n,
1238                                       struct iovec *iov,
1239                                       unsigned int iov_cnt,
1240                                       bool do_rss)
1241 {
1242     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1243     struct virtio_net_rss_config cfg;
1244     size_t s, offset = 0, size_get;
1245     uint16_t queue_pairs, i;
1246     struct {
1247         uint16_t us;
1248         uint8_t b;
1249     } QEMU_PACKED temp;
1250     const char *err_msg = "";
1251     uint32_t err_value = 0;
1252 
1253     if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) {
1254         err_msg = "RSS is not negotiated";
1255         goto error;
1256     }
1257     if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) {
1258         err_msg = "Hash report is not negotiated";
1259         goto error;
1260     }
1261     size_get = offsetof(struct virtio_net_rss_config, indirection_table);
1262     s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get);
1263     if (s != size_get) {
1264         err_msg = "Short command buffer";
1265         err_value = (uint32_t)s;
1266         goto error;
1267     }
1268     n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types);
1269     n->rss_data.indirections_len =
1270         virtio_lduw_p(vdev, &cfg.indirection_table_mask);
1271     n->rss_data.indirections_len++;
1272     if (!do_rss) {
1273         n->rss_data.indirections_len = 1;
1274     }
1275     if (!is_power_of_2(n->rss_data.indirections_len)) {
1276         err_msg = "Invalid size of indirection table";
1277         err_value = n->rss_data.indirections_len;
1278         goto error;
1279     }
1280     if (n->rss_data.indirections_len > VIRTIO_NET_RSS_MAX_TABLE_LEN) {
1281         err_msg = "Too large indirection table";
1282         err_value = n->rss_data.indirections_len;
1283         goto error;
1284     }
1285     n->rss_data.default_queue = do_rss ?
1286         virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0;
1287     if (n->rss_data.default_queue >= n->max_queue_pairs) {
1288         err_msg = "Invalid default queue";
1289         err_value = n->rss_data.default_queue;
1290         goto error;
1291     }
1292     offset += size_get;
1293     size_get = sizeof(uint16_t) * n->rss_data.indirections_len;
1294     g_free(n->rss_data.indirections_table);
1295     n->rss_data.indirections_table = g_malloc(size_get);
1296     if (!n->rss_data.indirections_table) {
1297         err_msg = "Can't allocate indirections table";
1298         err_value = n->rss_data.indirections_len;
1299         goto error;
1300     }
1301     s = iov_to_buf(iov, iov_cnt, offset,
1302                    n->rss_data.indirections_table, size_get);
1303     if (s != size_get) {
1304         err_msg = "Short indirection table buffer";
1305         err_value = (uint32_t)s;
1306         goto error;
1307     }
1308     for (i = 0; i < n->rss_data.indirections_len; ++i) {
1309         uint16_t val = n->rss_data.indirections_table[i];
1310         n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val);
1311     }
1312     offset += size_get;
1313     size_get = sizeof(temp);
1314     s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get);
1315     if (s != size_get) {
1316         err_msg = "Can't get queue_pairs";
1317         err_value = (uint32_t)s;
1318         goto error;
1319     }
1320     queue_pairs = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queue_pairs;
1321     if (queue_pairs == 0 || queue_pairs > n->max_queue_pairs) {
1322         err_msg = "Invalid number of queue_pairs";
1323         err_value = queue_pairs;
1324         goto error;
1325     }
1326     if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) {
1327         err_msg = "Invalid key size";
1328         err_value = temp.b;
1329         goto error;
1330     }
1331     if (!temp.b && n->rss_data.hash_types) {
1332         err_msg = "No key provided";
1333         err_value = 0;
1334         goto error;
1335     }
1336     if (!temp.b && !n->rss_data.hash_types) {
1337         virtio_net_disable_rss(n);
1338         return queue_pairs;
1339     }
1340     offset += size_get;
1341     size_get = temp.b;
1342     s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get);
1343     if (s != size_get) {
1344         err_msg = "Can get key buffer";
1345         err_value = (uint32_t)s;
1346         goto error;
1347     }
1348     n->rss_data.enabled = true;
1349 
1350     if (!n->rss_data.populate_hash) {
1351         if (!virtio_net_attach_epbf_rss(n)) {
1352             /* EBPF must be loaded for vhost */
1353             if (get_vhost_net(qemu_get_queue(n->nic)->peer)) {
1354                 warn_report("Can't load eBPF RSS for vhost");
1355                 goto error;
1356             }
1357             /* fallback to software RSS */
1358             warn_report("Can't load eBPF RSS - fallback to software RSS");
1359             n->rss_data.enabled_software_rss = true;
1360         }
1361     } else {
1362         /* use software RSS for hash populating */
1363         /* and detach eBPF if was loaded before */
1364         virtio_net_detach_epbf_rss(n);
1365         n->rss_data.enabled_software_rss = true;
1366     }
1367 
1368     trace_virtio_net_rss_enable(n->rss_data.hash_types,
1369                                 n->rss_data.indirections_len,
1370                                 temp.b);
1371     return queue_pairs;
1372 error:
1373     trace_virtio_net_rss_error(err_msg, err_value);
1374     virtio_net_disable_rss(n);
1375     return 0;
1376 }
1377 
1378 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
1379                                 struct iovec *iov, unsigned int iov_cnt)
1380 {
1381     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1382     uint16_t queue_pairs;
1383     NetClientState *nc = qemu_get_queue(n->nic);
1384 
1385     virtio_net_disable_rss(n);
1386     if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) {
1387         queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, false);
1388         return queue_pairs ? VIRTIO_NET_OK : VIRTIO_NET_ERR;
1389     }
1390     if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
1391         queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, true);
1392     } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
1393         struct virtio_net_ctrl_mq mq;
1394         size_t s;
1395         if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) {
1396             return VIRTIO_NET_ERR;
1397         }
1398         s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
1399         if (s != sizeof(mq)) {
1400             return VIRTIO_NET_ERR;
1401         }
1402         queue_pairs = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
1403 
1404     } else {
1405         return VIRTIO_NET_ERR;
1406     }
1407 
1408     if (queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1409         queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1410         queue_pairs > n->max_queue_pairs ||
1411         !n->multiqueue) {
1412         return VIRTIO_NET_ERR;
1413     }
1414 
1415     /* Avoid changing the number of queue_pairs for vdpa device in
1416      * userspace handler. A future fix is needed to handle the mq
1417      * change in userspace handler with vhost-vdpa. Let's disable
1418      * the mq handling from userspace for now and only allow get
1419      * done through the kernel. Ripples may be seen when falling
1420      * back to userspace, but without doing it qemu process would
1421      * crash on a recursive entry to virtio_net_set_status().
1422      */
1423     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
1424         return VIRTIO_NET_ERR;
1425     }
1426 
1427     n->curr_queue_pairs = queue_pairs;
1428     /* stop the backend before changing the number of queue_pairs to avoid handling a
1429      * disabled queue */
1430     virtio_net_set_status(vdev, vdev->status);
1431     virtio_net_set_queue_pairs(n);
1432 
1433     return VIRTIO_NET_OK;
1434 }
1435 
1436 size_t virtio_net_handle_ctrl_iov(VirtIODevice *vdev,
1437                                   const struct iovec *in_sg, unsigned in_num,
1438                                   const struct iovec *out_sg,
1439                                   unsigned out_num)
1440 {
1441     VirtIONet *n = VIRTIO_NET(vdev);
1442     struct virtio_net_ctrl_hdr ctrl;
1443     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
1444     size_t s;
1445     struct iovec *iov, *iov2;
1446 
1447     if (iov_size(in_sg, in_num) < sizeof(status) ||
1448         iov_size(out_sg, out_num) < sizeof(ctrl)) {
1449         virtio_error(vdev, "virtio-net ctrl missing headers");
1450         return 0;
1451     }
1452 
1453     iov2 = iov = g_memdup2(out_sg, sizeof(struct iovec) * out_num);
1454     s = iov_to_buf(iov, out_num, 0, &ctrl, sizeof(ctrl));
1455     iov_discard_front(&iov, &out_num, sizeof(ctrl));
1456     if (s != sizeof(ctrl)) {
1457         status = VIRTIO_NET_ERR;
1458     } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
1459         status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, out_num);
1460     } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
1461         status = virtio_net_handle_mac(n, ctrl.cmd, iov, out_num);
1462     } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
1463         status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, out_num);
1464     } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
1465         status = virtio_net_handle_announce(n, ctrl.cmd, iov, out_num);
1466     } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
1467         status = virtio_net_handle_mq(n, ctrl.cmd, iov, out_num);
1468     } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
1469         status = virtio_net_handle_offloads(n, ctrl.cmd, iov, out_num);
1470     }
1471 
1472     s = iov_from_buf(in_sg, in_num, 0, &status, sizeof(status));
1473     assert(s == sizeof(status));
1474 
1475     g_free(iov2);
1476     return sizeof(status);
1477 }
1478 
1479 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
1480 {
1481     VirtQueueElement *elem;
1482 
1483     for (;;) {
1484         size_t written;
1485         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
1486         if (!elem) {
1487             break;
1488         }
1489 
1490         written = virtio_net_handle_ctrl_iov(vdev, elem->in_sg, elem->in_num,
1491                                              elem->out_sg, elem->out_num);
1492         if (written > 0) {
1493             virtqueue_push(vq, elem, written);
1494             virtio_notify(vdev, vq);
1495             g_free(elem);
1496         } else {
1497             virtqueue_detach_element(vq, elem, 0);
1498             g_free(elem);
1499             break;
1500         }
1501     }
1502 }
1503 
1504 /* RX */
1505 
1506 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
1507 {
1508     VirtIONet *n = VIRTIO_NET(vdev);
1509     int queue_index = vq2q(virtio_get_queue_index(vq));
1510 
1511     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
1512 }
1513 
1514 static bool virtio_net_can_receive(NetClientState *nc)
1515 {
1516     VirtIONet *n = qemu_get_nic_opaque(nc);
1517     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1518     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1519 
1520     if (!vdev->vm_running) {
1521         return false;
1522     }
1523 
1524     if (nc->queue_index >= n->curr_queue_pairs) {
1525         return false;
1526     }
1527 
1528     if (!virtio_queue_ready(q->rx_vq) ||
1529         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1530         return false;
1531     }
1532 
1533     return true;
1534 }
1535 
1536 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1537 {
1538     VirtIONet *n = q->n;
1539     if (virtio_queue_empty(q->rx_vq) ||
1540         (n->mergeable_rx_bufs &&
1541          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1542         virtio_queue_set_notification(q->rx_vq, 1);
1543 
1544         /* To avoid a race condition where the guest has made some buffers
1545          * available after the above check but before notification was
1546          * enabled, check for available buffers again.
1547          */
1548         if (virtio_queue_empty(q->rx_vq) ||
1549             (n->mergeable_rx_bufs &&
1550              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1551             return 0;
1552         }
1553     }
1554 
1555     virtio_queue_set_notification(q->rx_vq, 0);
1556     return 1;
1557 }
1558 
1559 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1560 {
1561     virtio_tswap16s(vdev, &hdr->hdr_len);
1562     virtio_tswap16s(vdev, &hdr->gso_size);
1563     virtio_tswap16s(vdev, &hdr->csum_start);
1564     virtio_tswap16s(vdev, &hdr->csum_offset);
1565 }
1566 
1567 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1568  * it never finds out that the packets don't have valid checksums.  This
1569  * causes dhclient to get upset.  Fedora's carried a patch for ages to
1570  * fix this with Xen but it hasn't appeared in an upstream release of
1571  * dhclient yet.
1572  *
1573  * To avoid breaking existing guests, we catch udp packets and add
1574  * checksums.  This is terrible but it's better than hacking the guest
1575  * kernels.
1576  *
1577  * N.B. if we introduce a zero-copy API, this operation is no longer free so
1578  * we should provide a mechanism to disable it to avoid polluting the host
1579  * cache.
1580  */
1581 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1582                                         uint8_t *buf, size_t size)
1583 {
1584     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1585         (size > 27 && size < 1500) && /* normal sized MTU */
1586         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1587         (buf[23] == 17) && /* ip.protocol == UDP */
1588         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1589         net_checksum_calculate(buf, size, CSUM_UDP);
1590         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1591     }
1592 }
1593 
1594 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1595                            const void *buf, size_t size)
1596 {
1597     if (n->has_vnet_hdr) {
1598         /* FIXME this cast is evil */
1599         void *wbuf = (void *)buf;
1600         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1601                                     size - n->host_hdr_len);
1602 
1603         if (n->needs_vnet_hdr_swap) {
1604             virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1605         }
1606         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1607     } else {
1608         struct virtio_net_hdr hdr = {
1609             .flags = 0,
1610             .gso_type = VIRTIO_NET_HDR_GSO_NONE
1611         };
1612         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1613     }
1614 }
1615 
1616 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1617 {
1618     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1619     static const uint8_t vlan[] = {0x81, 0x00};
1620     uint8_t *ptr = (uint8_t *)buf;
1621     int i;
1622 
1623     if (n->promisc)
1624         return 1;
1625 
1626     ptr += n->host_hdr_len;
1627 
1628     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1629         int vid = lduw_be_p(ptr + 14) & 0xfff;
1630         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1631             return 0;
1632     }
1633 
1634     if (ptr[0] & 1) { // multicast
1635         if (!memcmp(ptr, bcast, sizeof(bcast))) {
1636             return !n->nobcast;
1637         } else if (n->nomulti) {
1638             return 0;
1639         } else if (n->allmulti || n->mac_table.multi_overflow) {
1640             return 1;
1641         }
1642 
1643         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1644             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1645                 return 1;
1646             }
1647         }
1648     } else { // unicast
1649         if (n->nouni) {
1650             return 0;
1651         } else if (n->alluni || n->mac_table.uni_overflow) {
1652             return 1;
1653         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1654             return 1;
1655         }
1656 
1657         for (i = 0; i < n->mac_table.first_multi; i++) {
1658             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1659                 return 1;
1660             }
1661         }
1662     }
1663 
1664     return 0;
1665 }
1666 
1667 static uint8_t virtio_net_get_hash_type(bool isip4,
1668                                         bool isip6,
1669                                         bool isudp,
1670                                         bool istcp,
1671                                         uint32_t types)
1672 {
1673     if (isip4) {
1674         if (istcp && (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4)) {
1675             return NetPktRssIpV4Tcp;
1676         }
1677         if (isudp && (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4)) {
1678             return NetPktRssIpV4Udp;
1679         }
1680         if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
1681             return NetPktRssIpV4;
1682         }
1683     } else if (isip6) {
1684         uint32_t mask = VIRTIO_NET_RSS_HASH_TYPE_TCP_EX |
1685                         VIRTIO_NET_RSS_HASH_TYPE_TCPv6;
1686 
1687         if (istcp && (types & mask)) {
1688             return (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) ?
1689                 NetPktRssIpV6TcpEx : NetPktRssIpV6Tcp;
1690         }
1691         mask = VIRTIO_NET_RSS_HASH_TYPE_UDP_EX | VIRTIO_NET_RSS_HASH_TYPE_UDPv6;
1692         if (isudp && (types & mask)) {
1693             return (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) ?
1694                 NetPktRssIpV6UdpEx : NetPktRssIpV6Udp;
1695         }
1696         mask = VIRTIO_NET_RSS_HASH_TYPE_IP_EX | VIRTIO_NET_RSS_HASH_TYPE_IPv6;
1697         if (types & mask) {
1698             return (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) ?
1699                 NetPktRssIpV6Ex : NetPktRssIpV6;
1700         }
1701     }
1702     return 0xff;
1703 }
1704 
1705 static void virtio_set_packet_hash(const uint8_t *buf, uint8_t report,
1706                                    uint32_t hash)
1707 {
1708     struct virtio_net_hdr_v1_hash *hdr = (void *)buf;
1709     hdr->hash_value = hash;
1710     hdr->hash_report = report;
1711 }
1712 
1713 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf,
1714                                   size_t size)
1715 {
1716     VirtIONet *n = qemu_get_nic_opaque(nc);
1717     unsigned int index = nc->queue_index, new_index = index;
1718     struct NetRxPkt *pkt = n->rx_pkt;
1719     uint8_t net_hash_type;
1720     uint32_t hash;
1721     bool isip4, isip6, isudp, istcp;
1722     static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = {
1723         VIRTIO_NET_HASH_REPORT_IPv4,
1724         VIRTIO_NET_HASH_REPORT_TCPv4,
1725         VIRTIO_NET_HASH_REPORT_TCPv6,
1726         VIRTIO_NET_HASH_REPORT_IPv6,
1727         VIRTIO_NET_HASH_REPORT_IPv6_EX,
1728         VIRTIO_NET_HASH_REPORT_TCPv6_EX,
1729         VIRTIO_NET_HASH_REPORT_UDPv4,
1730         VIRTIO_NET_HASH_REPORT_UDPv6,
1731         VIRTIO_NET_HASH_REPORT_UDPv6_EX
1732     };
1733 
1734     net_rx_pkt_set_protocols(pkt, buf + n->host_hdr_len,
1735                              size - n->host_hdr_len);
1736     net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
1737     if (isip4 && (net_rx_pkt_get_ip4_info(pkt)->fragment)) {
1738         istcp = isudp = false;
1739     }
1740     if (isip6 && (net_rx_pkt_get_ip6_info(pkt)->fragment)) {
1741         istcp = isudp = false;
1742     }
1743     net_hash_type = virtio_net_get_hash_type(isip4, isip6, isudp, istcp,
1744                                              n->rss_data.hash_types);
1745     if (net_hash_type > NetPktRssIpV6UdpEx) {
1746         if (n->rss_data.populate_hash) {
1747             virtio_set_packet_hash(buf, VIRTIO_NET_HASH_REPORT_NONE, 0);
1748         }
1749         return n->rss_data.redirect ? n->rss_data.default_queue : -1;
1750     }
1751 
1752     hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key);
1753 
1754     if (n->rss_data.populate_hash) {
1755         virtio_set_packet_hash(buf, reports[net_hash_type], hash);
1756     }
1757 
1758     if (n->rss_data.redirect) {
1759         new_index = hash & (n->rss_data.indirections_len - 1);
1760         new_index = n->rss_data.indirections_table[new_index];
1761     }
1762 
1763     return (index == new_index) ? -1 : new_index;
1764 }
1765 
1766 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
1767                                       size_t size, bool no_rss)
1768 {
1769     VirtIONet *n = qemu_get_nic_opaque(nc);
1770     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1771     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1772     VirtQueueElement *elems[VIRTQUEUE_MAX_SIZE];
1773     size_t lens[VIRTQUEUE_MAX_SIZE];
1774     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1775     struct virtio_net_hdr_mrg_rxbuf mhdr;
1776     unsigned mhdr_cnt = 0;
1777     size_t offset, i, guest_offset, j;
1778     ssize_t err;
1779 
1780     if (!virtio_net_can_receive(nc)) {
1781         return -1;
1782     }
1783 
1784     if (!no_rss && n->rss_data.enabled && n->rss_data.enabled_software_rss) {
1785         int index = virtio_net_process_rss(nc, buf, size);
1786         if (index >= 0) {
1787             NetClientState *nc2 = qemu_get_subqueue(n->nic, index);
1788             return virtio_net_receive_rcu(nc2, buf, size, true);
1789         }
1790     }
1791 
1792     /* hdr_len refers to the header we supply to the guest */
1793     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1794         return 0;
1795     }
1796 
1797     if (!receive_filter(n, buf, size))
1798         return size;
1799 
1800     offset = i = 0;
1801 
1802     while (offset < size) {
1803         VirtQueueElement *elem;
1804         int len, total;
1805         const struct iovec *sg;
1806 
1807         total = 0;
1808 
1809         if (i == VIRTQUEUE_MAX_SIZE) {
1810             virtio_error(vdev, "virtio-net unexpected long buffer chain");
1811             err = size;
1812             goto err;
1813         }
1814 
1815         elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1816         if (!elem) {
1817             if (i) {
1818                 virtio_error(vdev, "virtio-net unexpected empty queue: "
1819                              "i %zd mergeable %d offset %zd, size %zd, "
1820                              "guest hdr len %zd, host hdr len %zd "
1821                              "guest features 0x%" PRIx64,
1822                              i, n->mergeable_rx_bufs, offset, size,
1823                              n->guest_hdr_len, n->host_hdr_len,
1824                              vdev->guest_features);
1825             }
1826             err = -1;
1827             goto err;
1828         }
1829 
1830         if (elem->in_num < 1) {
1831             virtio_error(vdev,
1832                          "virtio-net receive queue contains no in buffers");
1833             virtqueue_detach_element(q->rx_vq, elem, 0);
1834             g_free(elem);
1835             err = -1;
1836             goto err;
1837         }
1838 
1839         sg = elem->in_sg;
1840         if (i == 0) {
1841             assert(offset == 0);
1842             if (n->mergeable_rx_bufs) {
1843                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1844                                     sg, elem->in_num,
1845                                     offsetof(typeof(mhdr), num_buffers),
1846                                     sizeof(mhdr.num_buffers));
1847             }
1848 
1849             receive_header(n, sg, elem->in_num, buf, size);
1850             if (n->rss_data.populate_hash) {
1851                 offset = sizeof(mhdr);
1852                 iov_from_buf(sg, elem->in_num, offset,
1853                              buf + offset, n->host_hdr_len - sizeof(mhdr));
1854             }
1855             offset = n->host_hdr_len;
1856             total += n->guest_hdr_len;
1857             guest_offset = n->guest_hdr_len;
1858         } else {
1859             guest_offset = 0;
1860         }
1861 
1862         /* copy in packet.  ugh */
1863         len = iov_from_buf(sg, elem->in_num, guest_offset,
1864                            buf + offset, size - offset);
1865         total += len;
1866         offset += len;
1867         /* If buffers can't be merged, at this point we
1868          * must have consumed the complete packet.
1869          * Otherwise, drop it. */
1870         if (!n->mergeable_rx_bufs && offset < size) {
1871             virtqueue_unpop(q->rx_vq, elem, total);
1872             g_free(elem);
1873             err = size;
1874             goto err;
1875         }
1876 
1877         elems[i] = elem;
1878         lens[i] = total;
1879         i++;
1880     }
1881 
1882     if (mhdr_cnt) {
1883         virtio_stw_p(vdev, &mhdr.num_buffers, i);
1884         iov_from_buf(mhdr_sg, mhdr_cnt,
1885                      0,
1886                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1887     }
1888 
1889     for (j = 0; j < i; j++) {
1890         /* signal other side */
1891         virtqueue_fill(q->rx_vq, elems[j], lens[j], j);
1892         g_free(elems[j]);
1893     }
1894 
1895     virtqueue_flush(q->rx_vq, i);
1896     virtio_notify(vdev, q->rx_vq);
1897 
1898     return size;
1899 
1900 err:
1901     for (j = 0; j < i; j++) {
1902         virtqueue_detach_element(q->rx_vq, elems[j], lens[j]);
1903         g_free(elems[j]);
1904     }
1905 
1906     return err;
1907 }
1908 
1909 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf,
1910                                   size_t size)
1911 {
1912     RCU_READ_LOCK_GUARD();
1913 
1914     return virtio_net_receive_rcu(nc, buf, size, false);
1915 }
1916 
1917 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain,
1918                                          const uint8_t *buf,
1919                                          VirtioNetRscUnit *unit)
1920 {
1921     uint16_t ip_hdrlen;
1922     struct ip_header *ip;
1923 
1924     ip = (struct ip_header *)(buf + chain->n->guest_hdr_len
1925                               + sizeof(struct eth_header));
1926     unit->ip = (void *)ip;
1927     ip_hdrlen = (ip->ip_ver_len & 0xF) << 2;
1928     unit->ip_plen = &ip->ip_len;
1929     unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen);
1930     unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1931     unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen;
1932 }
1933 
1934 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain,
1935                                          const uint8_t *buf,
1936                                          VirtioNetRscUnit *unit)
1937 {
1938     struct ip6_header *ip6;
1939 
1940     ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len
1941                                  + sizeof(struct eth_header));
1942     unit->ip = ip6;
1943     unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
1944     unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip)
1945                                         + sizeof(struct ip6_header));
1946     unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1947 
1948     /* There is a difference between payload lenght in ipv4 and v6,
1949        ip header is excluded in ipv6 */
1950     unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen;
1951 }
1952 
1953 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain,
1954                                        VirtioNetRscSeg *seg)
1955 {
1956     int ret;
1957     struct virtio_net_hdr_v1 *h;
1958 
1959     h = (struct virtio_net_hdr_v1 *)seg->buf;
1960     h->flags = 0;
1961     h->gso_type = VIRTIO_NET_HDR_GSO_NONE;
1962 
1963     if (seg->is_coalesced) {
1964         h->rsc.segments = seg->packets;
1965         h->rsc.dup_acks = seg->dup_ack;
1966         h->flags = VIRTIO_NET_HDR_F_RSC_INFO;
1967         if (chain->proto == ETH_P_IP) {
1968             h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1969         } else {
1970             h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1971         }
1972     }
1973 
1974     ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size);
1975     QTAILQ_REMOVE(&chain->buffers, seg, next);
1976     g_free(seg->buf);
1977     g_free(seg);
1978 
1979     return ret;
1980 }
1981 
1982 static void virtio_net_rsc_purge(void *opq)
1983 {
1984     VirtioNetRscSeg *seg, *rn;
1985     VirtioNetRscChain *chain = (VirtioNetRscChain *)opq;
1986 
1987     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) {
1988         if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
1989             chain->stat.purge_failed++;
1990             continue;
1991         }
1992     }
1993 
1994     chain->stat.timer++;
1995     if (!QTAILQ_EMPTY(&chain->buffers)) {
1996         timer_mod(chain->drain_timer,
1997               qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
1998     }
1999 }
2000 
2001 static void virtio_net_rsc_cleanup(VirtIONet *n)
2002 {
2003     VirtioNetRscChain *chain, *rn_chain;
2004     VirtioNetRscSeg *seg, *rn_seg;
2005 
2006     QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) {
2007         QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) {
2008             QTAILQ_REMOVE(&chain->buffers, seg, next);
2009             g_free(seg->buf);
2010             g_free(seg);
2011         }
2012 
2013         timer_free(chain->drain_timer);
2014         QTAILQ_REMOVE(&n->rsc_chains, chain, next);
2015         g_free(chain);
2016     }
2017 }
2018 
2019 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain,
2020                                      NetClientState *nc,
2021                                      const uint8_t *buf, size_t size)
2022 {
2023     uint16_t hdr_len;
2024     VirtioNetRscSeg *seg;
2025 
2026     hdr_len = chain->n->guest_hdr_len;
2027     seg = g_new(VirtioNetRscSeg, 1);
2028     seg->buf = g_malloc(hdr_len + sizeof(struct eth_header)
2029         + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD);
2030     memcpy(seg->buf, buf, size);
2031     seg->size = size;
2032     seg->packets = 1;
2033     seg->dup_ack = 0;
2034     seg->is_coalesced = 0;
2035     seg->nc = nc;
2036 
2037     QTAILQ_INSERT_TAIL(&chain->buffers, seg, next);
2038     chain->stat.cache++;
2039 
2040     switch (chain->proto) {
2041     case ETH_P_IP:
2042         virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
2043         break;
2044     case ETH_P_IPV6:
2045         virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit);
2046         break;
2047     default:
2048         g_assert_not_reached();
2049     }
2050 }
2051 
2052 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain,
2053                                          VirtioNetRscSeg *seg,
2054                                          const uint8_t *buf,
2055                                          struct tcp_header *n_tcp,
2056                                          struct tcp_header *o_tcp)
2057 {
2058     uint32_t nack, oack;
2059     uint16_t nwin, owin;
2060 
2061     nack = htonl(n_tcp->th_ack);
2062     nwin = htons(n_tcp->th_win);
2063     oack = htonl(o_tcp->th_ack);
2064     owin = htons(o_tcp->th_win);
2065 
2066     if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) {
2067         chain->stat.ack_out_of_win++;
2068         return RSC_FINAL;
2069     } else if (nack == oack) {
2070         /* duplicated ack or window probe */
2071         if (nwin == owin) {
2072             /* duplicated ack, add dup ack count due to whql test up to 1 */
2073             chain->stat.dup_ack++;
2074             return RSC_FINAL;
2075         } else {
2076             /* Coalesce window update */
2077             o_tcp->th_win = n_tcp->th_win;
2078             chain->stat.win_update++;
2079             return RSC_COALESCE;
2080         }
2081     } else {
2082         /* pure ack, go to 'C', finalize*/
2083         chain->stat.pure_ack++;
2084         return RSC_FINAL;
2085     }
2086 }
2087 
2088 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain,
2089                                             VirtioNetRscSeg *seg,
2090                                             const uint8_t *buf,
2091                                             VirtioNetRscUnit *n_unit)
2092 {
2093     void *data;
2094     uint16_t o_ip_len;
2095     uint32_t nseq, oseq;
2096     VirtioNetRscUnit *o_unit;
2097 
2098     o_unit = &seg->unit;
2099     o_ip_len = htons(*o_unit->ip_plen);
2100     nseq = htonl(n_unit->tcp->th_seq);
2101     oseq = htonl(o_unit->tcp->th_seq);
2102 
2103     /* out of order or retransmitted. */
2104     if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) {
2105         chain->stat.data_out_of_win++;
2106         return RSC_FINAL;
2107     }
2108 
2109     data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen;
2110     if (nseq == oseq) {
2111         if ((o_unit->payload == 0) && n_unit->payload) {
2112             /* From no payload to payload, normal case, not a dup ack or etc */
2113             chain->stat.data_after_pure_ack++;
2114             goto coalesce;
2115         } else {
2116             return virtio_net_rsc_handle_ack(chain, seg, buf,
2117                                              n_unit->tcp, o_unit->tcp);
2118         }
2119     } else if ((nseq - oseq) != o_unit->payload) {
2120         /* Not a consistent packet, out of order */
2121         chain->stat.data_out_of_order++;
2122         return RSC_FINAL;
2123     } else {
2124 coalesce:
2125         if ((o_ip_len + n_unit->payload) > chain->max_payload) {
2126             chain->stat.over_size++;
2127             return RSC_FINAL;
2128         }
2129 
2130         /* Here comes the right data, the payload length in v4/v6 is different,
2131            so use the field value to update and record the new data len */
2132         o_unit->payload += n_unit->payload; /* update new data len */
2133 
2134         /* update field in ip header */
2135         *o_unit->ip_plen = htons(o_ip_len + n_unit->payload);
2136 
2137         /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
2138            for windows guest, while this may change the behavior for linux
2139            guest (only if it uses RSC feature). */
2140         o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags;
2141 
2142         o_unit->tcp->th_ack = n_unit->tcp->th_ack;
2143         o_unit->tcp->th_win = n_unit->tcp->th_win;
2144 
2145         memmove(seg->buf + seg->size, data, n_unit->payload);
2146         seg->size += n_unit->payload;
2147         seg->packets++;
2148         chain->stat.coalesced++;
2149         return RSC_COALESCE;
2150     }
2151 }
2152 
2153 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain,
2154                                         VirtioNetRscSeg *seg,
2155                                         const uint8_t *buf, size_t size,
2156                                         VirtioNetRscUnit *unit)
2157 {
2158     struct ip_header *ip1, *ip2;
2159 
2160     ip1 = (struct ip_header *)(unit->ip);
2161     ip2 = (struct ip_header *)(seg->unit.ip);
2162     if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst)
2163         || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2164         || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2165         chain->stat.no_match++;
2166         return RSC_NO_MATCH;
2167     }
2168 
2169     return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2170 }
2171 
2172 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain,
2173                                         VirtioNetRscSeg *seg,
2174                                         const uint8_t *buf, size_t size,
2175                                         VirtioNetRscUnit *unit)
2176 {
2177     struct ip6_header *ip1, *ip2;
2178 
2179     ip1 = (struct ip6_header *)(unit->ip);
2180     ip2 = (struct ip6_header *)(seg->unit.ip);
2181     if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address))
2182         || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address))
2183         || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2184         || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2185             chain->stat.no_match++;
2186             return RSC_NO_MATCH;
2187     }
2188 
2189     return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2190 }
2191 
2192 /* Packets with 'SYN' should bypass, other flag should be sent after drain
2193  * to prevent out of order */
2194 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain,
2195                                          struct tcp_header *tcp)
2196 {
2197     uint16_t tcp_hdr;
2198     uint16_t tcp_flag;
2199 
2200     tcp_flag = htons(tcp->th_offset_flags);
2201     tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10;
2202     tcp_flag &= VIRTIO_NET_TCP_FLAG;
2203     if (tcp_flag & TH_SYN) {
2204         chain->stat.tcp_syn++;
2205         return RSC_BYPASS;
2206     }
2207 
2208     if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) {
2209         chain->stat.tcp_ctrl_drain++;
2210         return RSC_FINAL;
2211     }
2212 
2213     if (tcp_hdr > sizeof(struct tcp_header)) {
2214         chain->stat.tcp_all_opt++;
2215         return RSC_FINAL;
2216     }
2217 
2218     return RSC_CANDIDATE;
2219 }
2220 
2221 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain,
2222                                          NetClientState *nc,
2223                                          const uint8_t *buf, size_t size,
2224                                          VirtioNetRscUnit *unit)
2225 {
2226     int ret;
2227     VirtioNetRscSeg *seg, *nseg;
2228 
2229     if (QTAILQ_EMPTY(&chain->buffers)) {
2230         chain->stat.empty_cache++;
2231         virtio_net_rsc_cache_buf(chain, nc, buf, size);
2232         timer_mod(chain->drain_timer,
2233               qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
2234         return size;
2235     }
2236 
2237     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2238         if (chain->proto == ETH_P_IP) {
2239             ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
2240         } else {
2241             ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit);
2242         }
2243 
2244         if (ret == RSC_FINAL) {
2245             if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2246                 /* Send failed */
2247                 chain->stat.final_failed++;
2248                 return 0;
2249             }
2250 
2251             /* Send current packet */
2252             return virtio_net_do_receive(nc, buf, size);
2253         } else if (ret == RSC_NO_MATCH) {
2254             continue;
2255         } else {
2256             /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
2257             seg->is_coalesced = 1;
2258             return size;
2259         }
2260     }
2261 
2262     chain->stat.no_match_cache++;
2263     virtio_net_rsc_cache_buf(chain, nc, buf, size);
2264     return size;
2265 }
2266 
2267 /* Drain a connection data, this is to avoid out of order segments */
2268 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain,
2269                                         NetClientState *nc,
2270                                         const uint8_t *buf, size_t size,
2271                                         uint16_t ip_start, uint16_t ip_size,
2272                                         uint16_t tcp_port)
2273 {
2274     VirtioNetRscSeg *seg, *nseg;
2275     uint32_t ppair1, ppair2;
2276 
2277     ppair1 = *(uint32_t *)(buf + tcp_port);
2278     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2279         ppair2 = *(uint32_t *)(seg->buf + tcp_port);
2280         if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size)
2281             || (ppair1 != ppair2)) {
2282             continue;
2283         }
2284         if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2285             chain->stat.drain_failed++;
2286         }
2287 
2288         break;
2289     }
2290 
2291     return virtio_net_do_receive(nc, buf, size);
2292 }
2293 
2294 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain,
2295                                             struct ip_header *ip,
2296                                             const uint8_t *buf, size_t size)
2297 {
2298     uint16_t ip_len;
2299 
2300     /* Not an ipv4 packet */
2301     if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) {
2302         chain->stat.ip_option++;
2303         return RSC_BYPASS;
2304     }
2305 
2306     /* Don't handle packets with ip option */
2307     if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) {
2308         chain->stat.ip_option++;
2309         return RSC_BYPASS;
2310     }
2311 
2312     if (ip->ip_p != IPPROTO_TCP) {
2313         chain->stat.bypass_not_tcp++;
2314         return RSC_BYPASS;
2315     }
2316 
2317     /* Don't handle packets with ip fragment */
2318     if (!(htons(ip->ip_off) & IP_DF)) {
2319         chain->stat.ip_frag++;
2320         return RSC_BYPASS;
2321     }
2322 
2323     /* Don't handle packets with ecn flag */
2324     if (IPTOS_ECN(ip->ip_tos)) {
2325         chain->stat.ip_ecn++;
2326         return RSC_BYPASS;
2327     }
2328 
2329     ip_len = htons(ip->ip_len);
2330     if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header))
2331         || ip_len > (size - chain->n->guest_hdr_len -
2332                      sizeof(struct eth_header))) {
2333         chain->stat.ip_hacked++;
2334         return RSC_BYPASS;
2335     }
2336 
2337     return RSC_CANDIDATE;
2338 }
2339 
2340 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain,
2341                                       NetClientState *nc,
2342                                       const uint8_t *buf, size_t size)
2343 {
2344     int32_t ret;
2345     uint16_t hdr_len;
2346     VirtioNetRscUnit unit;
2347 
2348     hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2349 
2350     if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)
2351         + sizeof(struct tcp_header))) {
2352         chain->stat.bypass_not_tcp++;
2353         return virtio_net_do_receive(nc, buf, size);
2354     }
2355 
2356     virtio_net_rsc_extract_unit4(chain, buf, &unit);
2357     if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size)
2358         != RSC_CANDIDATE) {
2359         return virtio_net_do_receive(nc, buf, size);
2360     }
2361 
2362     ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2363     if (ret == RSC_BYPASS) {
2364         return virtio_net_do_receive(nc, buf, size);
2365     } else if (ret == RSC_FINAL) {
2366         return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2367                 ((hdr_len + sizeof(struct eth_header)) + 12),
2368                 VIRTIO_NET_IP4_ADDR_SIZE,
2369                 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header));
2370     }
2371 
2372     return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2373 }
2374 
2375 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain,
2376                                             struct ip6_header *ip6,
2377                                             const uint8_t *buf, size_t size)
2378 {
2379     uint16_t ip_len;
2380 
2381     if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4)
2382         != IP_HEADER_VERSION_6) {
2383         return RSC_BYPASS;
2384     }
2385 
2386     /* Both option and protocol is checked in this */
2387     if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) {
2388         chain->stat.bypass_not_tcp++;
2389         return RSC_BYPASS;
2390     }
2391 
2392     ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
2393     if (ip_len < sizeof(struct tcp_header) ||
2394         ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header)
2395                   - sizeof(struct ip6_header))) {
2396         chain->stat.ip_hacked++;
2397         return RSC_BYPASS;
2398     }
2399 
2400     /* Don't handle packets with ecn flag */
2401     if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) {
2402         chain->stat.ip_ecn++;
2403         return RSC_BYPASS;
2404     }
2405 
2406     return RSC_CANDIDATE;
2407 }
2408 
2409 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc,
2410                                       const uint8_t *buf, size_t size)
2411 {
2412     int32_t ret;
2413     uint16_t hdr_len;
2414     VirtioNetRscChain *chain;
2415     VirtioNetRscUnit unit;
2416 
2417     chain = (VirtioNetRscChain *)opq;
2418     hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2419 
2420     if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header)
2421         + sizeof(tcp_header))) {
2422         return virtio_net_do_receive(nc, buf, size);
2423     }
2424 
2425     virtio_net_rsc_extract_unit6(chain, buf, &unit);
2426     if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain,
2427                                                  unit.ip, buf, size)) {
2428         return virtio_net_do_receive(nc, buf, size);
2429     }
2430 
2431     ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2432     if (ret == RSC_BYPASS) {
2433         return virtio_net_do_receive(nc, buf, size);
2434     } else if (ret == RSC_FINAL) {
2435         return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2436                 ((hdr_len + sizeof(struct eth_header)) + 8),
2437                 VIRTIO_NET_IP6_ADDR_SIZE,
2438                 hdr_len + sizeof(struct eth_header)
2439                 + sizeof(struct ip6_header));
2440     }
2441 
2442     return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2443 }
2444 
2445 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n,
2446                                                       NetClientState *nc,
2447                                                       uint16_t proto)
2448 {
2449     VirtioNetRscChain *chain;
2450 
2451     if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) {
2452         return NULL;
2453     }
2454 
2455     QTAILQ_FOREACH(chain, &n->rsc_chains, next) {
2456         if (chain->proto == proto) {
2457             return chain;
2458         }
2459     }
2460 
2461     chain = g_malloc(sizeof(*chain));
2462     chain->n = n;
2463     chain->proto = proto;
2464     if (proto == (uint16_t)ETH_P_IP) {
2465         chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
2466         chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2467     } else {
2468         chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD;
2469         chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2470     }
2471     chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST,
2472                                       virtio_net_rsc_purge, chain);
2473     memset(&chain->stat, 0, sizeof(chain->stat));
2474 
2475     QTAILQ_INIT(&chain->buffers);
2476     QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next);
2477 
2478     return chain;
2479 }
2480 
2481 static ssize_t virtio_net_rsc_receive(NetClientState *nc,
2482                                       const uint8_t *buf,
2483                                       size_t size)
2484 {
2485     uint16_t proto;
2486     VirtioNetRscChain *chain;
2487     struct eth_header *eth;
2488     VirtIONet *n;
2489 
2490     n = qemu_get_nic_opaque(nc);
2491     if (size < (n->host_hdr_len + sizeof(struct eth_header))) {
2492         return virtio_net_do_receive(nc, buf, size);
2493     }
2494 
2495     eth = (struct eth_header *)(buf + n->guest_hdr_len);
2496     proto = htons(eth->h_proto);
2497 
2498     chain = virtio_net_rsc_lookup_chain(n, nc, proto);
2499     if (chain) {
2500         chain->stat.received++;
2501         if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) {
2502             return virtio_net_rsc_receive4(chain, nc, buf, size);
2503         } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) {
2504             return virtio_net_rsc_receive6(chain, nc, buf, size);
2505         }
2506     }
2507     return virtio_net_do_receive(nc, buf, size);
2508 }
2509 
2510 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf,
2511                                   size_t size)
2512 {
2513     VirtIONet *n = qemu_get_nic_opaque(nc);
2514     if ((n->rsc4_enabled || n->rsc6_enabled)) {
2515         return virtio_net_rsc_receive(nc, buf, size);
2516     } else {
2517         return virtio_net_do_receive(nc, buf, size);
2518     }
2519 }
2520 
2521 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
2522 
2523 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
2524 {
2525     VirtIONet *n = qemu_get_nic_opaque(nc);
2526     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
2527     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2528 
2529     virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
2530     virtio_notify(vdev, q->tx_vq);
2531 
2532     g_free(q->async_tx.elem);
2533     q->async_tx.elem = NULL;
2534 
2535     virtio_queue_set_notification(q->tx_vq, 1);
2536     virtio_net_flush_tx(q);
2537 }
2538 
2539 /* TX */
2540 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
2541 {
2542     VirtIONet *n = q->n;
2543     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2544     VirtQueueElement *elem;
2545     int32_t num_packets = 0;
2546     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
2547     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2548         return num_packets;
2549     }
2550 
2551     if (q->async_tx.elem) {
2552         virtio_queue_set_notification(q->tx_vq, 0);
2553         return num_packets;
2554     }
2555 
2556     for (;;) {
2557         ssize_t ret;
2558         unsigned int out_num;
2559         struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
2560         struct virtio_net_hdr_mrg_rxbuf mhdr;
2561 
2562         elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
2563         if (!elem) {
2564             break;
2565         }
2566 
2567         out_num = elem->out_num;
2568         out_sg = elem->out_sg;
2569         if (out_num < 1) {
2570             virtio_error(vdev, "virtio-net header not in first element");
2571             virtqueue_detach_element(q->tx_vq, elem, 0);
2572             g_free(elem);
2573             return -EINVAL;
2574         }
2575 
2576         if (n->has_vnet_hdr) {
2577             if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
2578                 n->guest_hdr_len) {
2579                 virtio_error(vdev, "virtio-net header incorrect");
2580                 virtqueue_detach_element(q->tx_vq, elem, 0);
2581                 g_free(elem);
2582                 return -EINVAL;
2583             }
2584             if (n->needs_vnet_hdr_swap) {
2585                 virtio_net_hdr_swap(vdev, (void *) &mhdr);
2586                 sg2[0].iov_base = &mhdr;
2587                 sg2[0].iov_len = n->guest_hdr_len;
2588                 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
2589                                    out_sg, out_num,
2590                                    n->guest_hdr_len, -1);
2591                 if (out_num == VIRTQUEUE_MAX_SIZE) {
2592                     goto drop;
2593                 }
2594                 out_num += 1;
2595                 out_sg = sg2;
2596             }
2597         }
2598         /*
2599          * If host wants to see the guest header as is, we can
2600          * pass it on unchanged. Otherwise, copy just the parts
2601          * that host is interested in.
2602          */
2603         assert(n->host_hdr_len <= n->guest_hdr_len);
2604         if (n->host_hdr_len != n->guest_hdr_len) {
2605             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
2606                                        out_sg, out_num,
2607                                        0, n->host_hdr_len);
2608             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
2609                              out_sg, out_num,
2610                              n->guest_hdr_len, -1);
2611             out_num = sg_num;
2612             out_sg = sg;
2613         }
2614 
2615         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
2616                                       out_sg, out_num, virtio_net_tx_complete);
2617         if (ret == 0) {
2618             virtio_queue_set_notification(q->tx_vq, 0);
2619             q->async_tx.elem = elem;
2620             return -EBUSY;
2621         }
2622 
2623 drop:
2624         virtqueue_push(q->tx_vq, elem, 0);
2625         virtio_notify(vdev, q->tx_vq);
2626         g_free(elem);
2627 
2628         if (++num_packets >= n->tx_burst) {
2629             break;
2630         }
2631     }
2632     return num_packets;
2633 }
2634 
2635 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
2636 {
2637     VirtIONet *n = VIRTIO_NET(vdev);
2638     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2639 
2640     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2641         virtio_net_drop_tx_queue_data(vdev, vq);
2642         return;
2643     }
2644 
2645     /* This happens when device was stopped but VCPU wasn't. */
2646     if (!vdev->vm_running) {
2647         q->tx_waiting = 1;
2648         return;
2649     }
2650 
2651     if (q->tx_waiting) {
2652         virtio_queue_set_notification(vq, 1);
2653         timer_del(q->tx_timer);
2654         q->tx_waiting = 0;
2655         if (virtio_net_flush_tx(q) == -EINVAL) {
2656             return;
2657         }
2658     } else {
2659         timer_mod(q->tx_timer,
2660                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2661         q->tx_waiting = 1;
2662         virtio_queue_set_notification(vq, 0);
2663     }
2664 }
2665 
2666 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
2667 {
2668     VirtIONet *n = VIRTIO_NET(vdev);
2669     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2670 
2671     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2672         virtio_net_drop_tx_queue_data(vdev, vq);
2673         return;
2674     }
2675 
2676     if (unlikely(q->tx_waiting)) {
2677         return;
2678     }
2679     q->tx_waiting = 1;
2680     /* This happens when device was stopped but VCPU wasn't. */
2681     if (!vdev->vm_running) {
2682         return;
2683     }
2684     virtio_queue_set_notification(vq, 0);
2685     qemu_bh_schedule(q->tx_bh);
2686 }
2687 
2688 static void virtio_net_tx_timer(void *opaque)
2689 {
2690     VirtIONetQueue *q = opaque;
2691     VirtIONet *n = q->n;
2692     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2693     /* This happens when device was stopped but BH wasn't. */
2694     if (!vdev->vm_running) {
2695         /* Make sure tx waiting is set, so we'll run when restarted. */
2696         assert(q->tx_waiting);
2697         return;
2698     }
2699 
2700     q->tx_waiting = 0;
2701 
2702     /* Just in case the driver is not ready on more */
2703     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2704         return;
2705     }
2706 
2707     virtio_queue_set_notification(q->tx_vq, 1);
2708     virtio_net_flush_tx(q);
2709 }
2710 
2711 static void virtio_net_tx_bh(void *opaque)
2712 {
2713     VirtIONetQueue *q = opaque;
2714     VirtIONet *n = q->n;
2715     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2716     int32_t ret;
2717 
2718     /* This happens when device was stopped but BH wasn't. */
2719     if (!vdev->vm_running) {
2720         /* Make sure tx waiting is set, so we'll run when restarted. */
2721         assert(q->tx_waiting);
2722         return;
2723     }
2724 
2725     q->tx_waiting = 0;
2726 
2727     /* Just in case the driver is not ready on more */
2728     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
2729         return;
2730     }
2731 
2732     ret = virtio_net_flush_tx(q);
2733     if (ret == -EBUSY || ret == -EINVAL) {
2734         return; /* Notification re-enable handled by tx_complete or device
2735                  * broken */
2736     }
2737 
2738     /* If we flush a full burst of packets, assume there are
2739      * more coming and immediately reschedule */
2740     if (ret >= n->tx_burst) {
2741         qemu_bh_schedule(q->tx_bh);
2742         q->tx_waiting = 1;
2743         return;
2744     }
2745 
2746     /* If less than a full burst, re-enable notification and flush
2747      * anything that may have come in while we weren't looking.  If
2748      * we find something, assume the guest is still active and reschedule */
2749     virtio_queue_set_notification(q->tx_vq, 1);
2750     ret = virtio_net_flush_tx(q);
2751     if (ret == -EINVAL) {
2752         return;
2753     } else if (ret > 0) {
2754         virtio_queue_set_notification(q->tx_vq, 0);
2755         qemu_bh_schedule(q->tx_bh);
2756         q->tx_waiting = 1;
2757     }
2758 }
2759 
2760 static void virtio_net_add_queue(VirtIONet *n, int index)
2761 {
2762     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2763 
2764     n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
2765                                            virtio_net_handle_rx);
2766 
2767     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
2768         n->vqs[index].tx_vq =
2769             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2770                              virtio_net_handle_tx_timer);
2771         n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
2772                                               virtio_net_tx_timer,
2773                                               &n->vqs[index]);
2774     } else {
2775         n->vqs[index].tx_vq =
2776             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2777                              virtio_net_handle_tx_bh);
2778         n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
2779     }
2780 
2781     n->vqs[index].tx_waiting = 0;
2782     n->vqs[index].n = n;
2783 }
2784 
2785 static void virtio_net_del_queue(VirtIONet *n, int index)
2786 {
2787     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2788     VirtIONetQueue *q = &n->vqs[index];
2789     NetClientState *nc = qemu_get_subqueue(n->nic, index);
2790 
2791     qemu_purge_queued_packets(nc);
2792 
2793     virtio_del_queue(vdev, index * 2);
2794     if (q->tx_timer) {
2795         timer_free(q->tx_timer);
2796         q->tx_timer = NULL;
2797     } else {
2798         qemu_bh_delete(q->tx_bh);
2799         q->tx_bh = NULL;
2800     }
2801     q->tx_waiting = 0;
2802     virtio_del_queue(vdev, index * 2 + 1);
2803 }
2804 
2805 static void virtio_net_change_num_queue_pairs(VirtIONet *n, int new_max_queue_pairs)
2806 {
2807     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2808     int old_num_queues = virtio_get_num_queues(vdev);
2809     int new_num_queues = new_max_queue_pairs * 2 + 1;
2810     int i;
2811 
2812     assert(old_num_queues >= 3);
2813     assert(old_num_queues % 2 == 1);
2814 
2815     if (old_num_queues == new_num_queues) {
2816         return;
2817     }
2818 
2819     /*
2820      * We always need to remove and add ctrl vq if
2821      * old_num_queues != new_num_queues. Remove ctrl_vq first,
2822      * and then we only enter one of the following two loops.
2823      */
2824     virtio_del_queue(vdev, old_num_queues - 1);
2825 
2826     for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
2827         /* new_num_queues < old_num_queues */
2828         virtio_net_del_queue(n, i / 2);
2829     }
2830 
2831     for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
2832         /* new_num_queues > old_num_queues */
2833         virtio_net_add_queue(n, i / 2);
2834     }
2835 
2836     /* add ctrl_vq last */
2837     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
2838 }
2839 
2840 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
2841 {
2842     int max = multiqueue ? n->max_queue_pairs : 1;
2843 
2844     n->multiqueue = multiqueue;
2845     virtio_net_change_num_queue_pairs(n, max);
2846 
2847     virtio_net_set_queue_pairs(n);
2848 }
2849 
2850 static int virtio_net_post_load_device(void *opaque, int version_id)
2851 {
2852     VirtIONet *n = opaque;
2853     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2854     int i, link_down;
2855 
2856     trace_virtio_net_post_load_device();
2857     virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
2858                                virtio_vdev_has_feature(vdev,
2859                                                        VIRTIO_F_VERSION_1),
2860                                virtio_vdev_has_feature(vdev,
2861                                                        VIRTIO_NET_F_HASH_REPORT));
2862 
2863     /* MAC_TABLE_ENTRIES may be different from the saved image */
2864     if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
2865         n->mac_table.in_use = 0;
2866     }
2867 
2868     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
2869         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
2870     }
2871 
2872     /*
2873      * curr_guest_offloads will be later overwritten by the
2874      * virtio_set_features_nocheck call done from the virtio_load.
2875      * Here we make sure it is preserved and restored accordingly
2876      * in the virtio_net_post_load_virtio callback.
2877      */
2878     n->saved_guest_offloads = n->curr_guest_offloads;
2879 
2880     virtio_net_set_queue_pairs(n);
2881 
2882     /* Find the first multicast entry in the saved MAC filter */
2883     for (i = 0; i < n->mac_table.in_use; i++) {
2884         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
2885             break;
2886         }
2887     }
2888     n->mac_table.first_multi = i;
2889 
2890     /* nc.link_down can't be migrated, so infer link_down according
2891      * to link status bit in n->status */
2892     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
2893     for (i = 0; i < n->max_queue_pairs; i++) {
2894         qemu_get_subqueue(n->nic, i)->link_down = link_down;
2895     }
2896 
2897     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
2898         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
2899         qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
2900                                   QEMU_CLOCK_VIRTUAL,
2901                                   virtio_net_announce_timer, n);
2902         if (n->announce_timer.round) {
2903             timer_mod(n->announce_timer.tm,
2904                       qemu_clock_get_ms(n->announce_timer.type));
2905         } else {
2906             qemu_announce_timer_del(&n->announce_timer, false);
2907         }
2908     }
2909 
2910     if (n->rss_data.enabled) {
2911         n->rss_data.enabled_software_rss = n->rss_data.populate_hash;
2912         if (!n->rss_data.populate_hash) {
2913             if (!virtio_net_attach_epbf_rss(n)) {
2914                 if (get_vhost_net(qemu_get_queue(n->nic)->peer)) {
2915                     warn_report("Can't post-load eBPF RSS for vhost");
2916                 } else {
2917                     warn_report("Can't post-load eBPF RSS - "
2918                                 "fallback to software RSS");
2919                     n->rss_data.enabled_software_rss = true;
2920                 }
2921             }
2922         }
2923 
2924         trace_virtio_net_rss_enable(n->rss_data.hash_types,
2925                                     n->rss_data.indirections_len,
2926                                     sizeof(n->rss_data.key));
2927     } else {
2928         trace_virtio_net_rss_disable();
2929     }
2930     return 0;
2931 }
2932 
2933 static int virtio_net_post_load_virtio(VirtIODevice *vdev)
2934 {
2935     VirtIONet *n = VIRTIO_NET(vdev);
2936     /*
2937      * The actual needed state is now in saved_guest_offloads,
2938      * see virtio_net_post_load_device for detail.
2939      * Restore it back and apply the desired offloads.
2940      */
2941     n->curr_guest_offloads = n->saved_guest_offloads;
2942     if (peer_has_vnet_hdr(n)) {
2943         virtio_net_apply_guest_offloads(n);
2944     }
2945 
2946     return 0;
2947 }
2948 
2949 /* tx_waiting field of a VirtIONetQueue */
2950 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = {
2951     .name = "virtio-net-queue-tx_waiting",
2952     .fields = (VMStateField[]) {
2953         VMSTATE_UINT32(tx_waiting, VirtIONetQueue),
2954         VMSTATE_END_OF_LIST()
2955    },
2956 };
2957 
2958 static bool max_queue_pairs_gt_1(void *opaque, int version_id)
2959 {
2960     return VIRTIO_NET(opaque)->max_queue_pairs > 1;
2961 }
2962 
2963 static bool has_ctrl_guest_offloads(void *opaque, int version_id)
2964 {
2965     return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque),
2966                                    VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
2967 }
2968 
2969 static bool mac_table_fits(void *opaque, int version_id)
2970 {
2971     return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES;
2972 }
2973 
2974 static bool mac_table_doesnt_fit(void *opaque, int version_id)
2975 {
2976     return !mac_table_fits(opaque, version_id);
2977 }
2978 
2979 /* This temporary type is shared by all the WITH_TMP methods
2980  * although only some fields are used by each.
2981  */
2982 struct VirtIONetMigTmp {
2983     VirtIONet      *parent;
2984     VirtIONetQueue *vqs_1;
2985     uint16_t        curr_queue_pairs_1;
2986     uint8_t         has_ufo;
2987     uint32_t        has_vnet_hdr;
2988 };
2989 
2990 /* The 2nd and subsequent tx_waiting flags are loaded later than
2991  * the 1st entry in the queue_pairs and only if there's more than one
2992  * entry.  We use the tmp mechanism to calculate a temporary
2993  * pointer and count and also validate the count.
2994  */
2995 
2996 static int virtio_net_tx_waiting_pre_save(void *opaque)
2997 {
2998     struct VirtIONetMigTmp *tmp = opaque;
2999 
3000     tmp->vqs_1 = tmp->parent->vqs + 1;
3001     tmp->curr_queue_pairs_1 = tmp->parent->curr_queue_pairs - 1;
3002     if (tmp->parent->curr_queue_pairs == 0) {
3003         tmp->curr_queue_pairs_1 = 0;
3004     }
3005 
3006     return 0;
3007 }
3008 
3009 static int virtio_net_tx_waiting_pre_load(void *opaque)
3010 {
3011     struct VirtIONetMigTmp *tmp = opaque;
3012 
3013     /* Reuse the pointer setup from save */
3014     virtio_net_tx_waiting_pre_save(opaque);
3015 
3016     if (tmp->parent->curr_queue_pairs > tmp->parent->max_queue_pairs) {
3017         error_report("virtio-net: curr_queue_pairs %x > max_queue_pairs %x",
3018             tmp->parent->curr_queue_pairs, tmp->parent->max_queue_pairs);
3019 
3020         return -EINVAL;
3021     }
3022 
3023     return 0; /* all good */
3024 }
3025 
3026 static const VMStateDescription vmstate_virtio_net_tx_waiting = {
3027     .name      = "virtio-net-tx_waiting",
3028     .pre_load  = virtio_net_tx_waiting_pre_load,
3029     .pre_save  = virtio_net_tx_waiting_pre_save,
3030     .fields    = (VMStateField[]) {
3031         VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp,
3032                                      curr_queue_pairs_1,
3033                                      vmstate_virtio_net_queue_tx_waiting,
3034                                      struct VirtIONetQueue),
3035         VMSTATE_END_OF_LIST()
3036     },
3037 };
3038 
3039 /* the 'has_ufo' flag is just tested; if the incoming stream has the
3040  * flag set we need to check that we have it
3041  */
3042 static int virtio_net_ufo_post_load(void *opaque, int version_id)
3043 {
3044     struct VirtIONetMigTmp *tmp = opaque;
3045 
3046     if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) {
3047         error_report("virtio-net: saved image requires TUN_F_UFO support");
3048         return -EINVAL;
3049     }
3050 
3051     return 0;
3052 }
3053 
3054 static int virtio_net_ufo_pre_save(void *opaque)
3055 {
3056     struct VirtIONetMigTmp *tmp = opaque;
3057 
3058     tmp->has_ufo = tmp->parent->has_ufo;
3059 
3060     return 0;
3061 }
3062 
3063 static const VMStateDescription vmstate_virtio_net_has_ufo = {
3064     .name      = "virtio-net-ufo",
3065     .post_load = virtio_net_ufo_post_load,
3066     .pre_save  = virtio_net_ufo_pre_save,
3067     .fields    = (VMStateField[]) {
3068         VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp),
3069         VMSTATE_END_OF_LIST()
3070     },
3071 };
3072 
3073 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
3074  * flag set we need to check that we have it
3075  */
3076 static int virtio_net_vnet_post_load(void *opaque, int version_id)
3077 {
3078     struct VirtIONetMigTmp *tmp = opaque;
3079 
3080     if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) {
3081         error_report("virtio-net: saved image requires vnet_hdr=on");
3082         return -EINVAL;
3083     }
3084 
3085     return 0;
3086 }
3087 
3088 static int virtio_net_vnet_pre_save(void *opaque)
3089 {
3090     struct VirtIONetMigTmp *tmp = opaque;
3091 
3092     tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr;
3093 
3094     return 0;
3095 }
3096 
3097 static const VMStateDescription vmstate_virtio_net_has_vnet = {
3098     .name      = "virtio-net-vnet",
3099     .post_load = virtio_net_vnet_post_load,
3100     .pre_save  = virtio_net_vnet_pre_save,
3101     .fields    = (VMStateField[]) {
3102         VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp),
3103         VMSTATE_END_OF_LIST()
3104     },
3105 };
3106 
3107 static bool virtio_net_rss_needed(void *opaque)
3108 {
3109     return VIRTIO_NET(opaque)->rss_data.enabled;
3110 }
3111 
3112 static const VMStateDescription vmstate_virtio_net_rss = {
3113     .name      = "virtio-net-device/rss",
3114     .version_id = 1,
3115     .minimum_version_id = 1,
3116     .needed = virtio_net_rss_needed,
3117     .fields = (VMStateField[]) {
3118         VMSTATE_BOOL(rss_data.enabled, VirtIONet),
3119         VMSTATE_BOOL(rss_data.redirect, VirtIONet),
3120         VMSTATE_BOOL(rss_data.populate_hash, VirtIONet),
3121         VMSTATE_UINT32(rss_data.hash_types, VirtIONet),
3122         VMSTATE_UINT16(rss_data.indirections_len, VirtIONet),
3123         VMSTATE_UINT16(rss_data.default_queue, VirtIONet),
3124         VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet,
3125                             VIRTIO_NET_RSS_MAX_KEY_SIZE),
3126         VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet,
3127                                     rss_data.indirections_len, 0,
3128                                     vmstate_info_uint16, uint16_t),
3129         VMSTATE_END_OF_LIST()
3130     },
3131 };
3132 
3133 static const VMStateDescription vmstate_virtio_net_device = {
3134     .name = "virtio-net-device",
3135     .version_id = VIRTIO_NET_VM_VERSION,
3136     .minimum_version_id = VIRTIO_NET_VM_VERSION,
3137     .post_load = virtio_net_post_load_device,
3138     .fields = (VMStateField[]) {
3139         VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN),
3140         VMSTATE_STRUCT_POINTER(vqs, VirtIONet,
3141                                vmstate_virtio_net_queue_tx_waiting,
3142                                VirtIONetQueue),
3143         VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet),
3144         VMSTATE_UINT16(status, VirtIONet),
3145         VMSTATE_UINT8(promisc, VirtIONet),
3146         VMSTATE_UINT8(allmulti, VirtIONet),
3147         VMSTATE_UINT32(mac_table.in_use, VirtIONet),
3148 
3149         /* Guarded pair: If it fits we load it, else we throw it away
3150          * - can happen if source has a larger MAC table.; post-load
3151          *  sets flags in this case.
3152          */
3153         VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet,
3154                                 0, mac_table_fits, mac_table.in_use,
3155                                  ETH_ALEN),
3156         VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0,
3157                                      mac_table.in_use, ETH_ALEN),
3158 
3159         /* Note: This is an array of uint32's that's always been saved as a
3160          * buffer; hold onto your endiannesses; it's actually used as a bitmap
3161          * but based on the uint.
3162          */
3163         VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3),
3164         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3165                          vmstate_virtio_net_has_vnet),
3166         VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet),
3167         VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet),
3168         VMSTATE_UINT8(alluni, VirtIONet),
3169         VMSTATE_UINT8(nomulti, VirtIONet),
3170         VMSTATE_UINT8(nouni, VirtIONet),
3171         VMSTATE_UINT8(nobcast, VirtIONet),
3172         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3173                          vmstate_virtio_net_has_ufo),
3174         VMSTATE_SINGLE_TEST(max_queue_pairs, VirtIONet, max_queue_pairs_gt_1, 0,
3175                             vmstate_info_uint16_equal, uint16_t),
3176         VMSTATE_UINT16_TEST(curr_queue_pairs, VirtIONet, max_queue_pairs_gt_1),
3177         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3178                          vmstate_virtio_net_tx_waiting),
3179         VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet,
3180                             has_ctrl_guest_offloads),
3181         VMSTATE_END_OF_LIST()
3182    },
3183     .subsections = (const VMStateDescription * []) {
3184         &vmstate_virtio_net_rss,
3185         NULL
3186     }
3187 };
3188 
3189 static NetClientInfo net_virtio_info = {
3190     .type = NET_CLIENT_DRIVER_NIC,
3191     .size = sizeof(NICState),
3192     .can_receive = virtio_net_can_receive,
3193     .receive = virtio_net_receive,
3194     .link_status_changed = virtio_net_set_link_status,
3195     .query_rx_filter = virtio_net_query_rxfilter,
3196     .announce = virtio_net_announce,
3197 };
3198 
3199 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
3200 {
3201     VirtIONet *n = VIRTIO_NET(vdev);
3202     NetClientState *nc;
3203     assert(n->vhost_started);
3204     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ) && idx == 2) {
3205         /* Must guard against invalid features and bogus queue index
3206          * from being set by malicious guest, or penetrated through
3207          * buggy migration stream.
3208          */
3209         if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
3210             qemu_log_mask(LOG_GUEST_ERROR,
3211                           "%s: bogus vq index ignored\n", __func__);
3212             return false;
3213         }
3214         nc = qemu_get_subqueue(n->nic, n->max_queue_pairs);
3215     } else {
3216         nc = qemu_get_subqueue(n->nic, vq2q(idx));
3217     }
3218     return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
3219 }
3220 
3221 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
3222                                            bool mask)
3223 {
3224     VirtIONet *n = VIRTIO_NET(vdev);
3225     NetClientState *nc;
3226     assert(n->vhost_started);
3227     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ) && idx == 2) {
3228         /* Must guard against invalid features and bogus queue index
3229          * from being set by malicious guest, or penetrated through
3230          * buggy migration stream.
3231          */
3232         if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
3233             qemu_log_mask(LOG_GUEST_ERROR,
3234                           "%s: bogus vq index ignored\n", __func__);
3235             return;
3236         }
3237         nc = qemu_get_subqueue(n->nic, n->max_queue_pairs);
3238     } else {
3239         nc = qemu_get_subqueue(n->nic, vq2q(idx));
3240     }
3241     vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
3242                              vdev, idx, mask);
3243 }
3244 
3245 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
3246 {
3247     virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
3248 
3249     n->config_size = virtio_feature_get_config_size(feature_sizes,
3250                                                     host_features);
3251 }
3252 
3253 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
3254                                    const char *type)
3255 {
3256     /*
3257      * The name can be NULL, the netclient name will be type.x.
3258      */
3259     assert(type != NULL);
3260 
3261     g_free(n->netclient_name);
3262     g_free(n->netclient_type);
3263     n->netclient_name = g_strdup(name);
3264     n->netclient_type = g_strdup(type);
3265 }
3266 
3267 static bool failover_unplug_primary(VirtIONet *n, DeviceState *dev)
3268 {
3269     HotplugHandler *hotplug_ctrl;
3270     PCIDevice *pci_dev;
3271     Error *err = NULL;
3272 
3273     hotplug_ctrl = qdev_get_hotplug_handler(dev);
3274     if (hotplug_ctrl) {
3275         pci_dev = PCI_DEVICE(dev);
3276         pci_dev->partially_hotplugged = true;
3277         hotplug_handler_unplug_request(hotplug_ctrl, dev, &err);
3278         if (err) {
3279             error_report_err(err);
3280             return false;
3281         }
3282     } else {
3283         return false;
3284     }
3285     return true;
3286 }
3287 
3288 static bool failover_replug_primary(VirtIONet *n, DeviceState *dev,
3289                                     Error **errp)
3290 {
3291     Error *err = NULL;
3292     HotplugHandler *hotplug_ctrl;
3293     PCIDevice *pdev = PCI_DEVICE(dev);
3294     BusState *primary_bus;
3295 
3296     if (!pdev->partially_hotplugged) {
3297         return true;
3298     }
3299     primary_bus = dev->parent_bus;
3300     if (!primary_bus) {
3301         error_setg(errp, "virtio_net: couldn't find primary bus");
3302         return false;
3303     }
3304     qdev_set_parent_bus(dev, primary_bus, &error_abort);
3305     qatomic_set(&n->failover_primary_hidden, false);
3306     hotplug_ctrl = qdev_get_hotplug_handler(dev);
3307     if (hotplug_ctrl) {
3308         hotplug_handler_pre_plug(hotplug_ctrl, dev, &err);
3309         if (err) {
3310             goto out;
3311         }
3312         hotplug_handler_plug(hotplug_ctrl, dev, &err);
3313     }
3314     pdev->partially_hotplugged = false;
3315 
3316 out:
3317     error_propagate(errp, err);
3318     return !err;
3319 }
3320 
3321 static void virtio_net_handle_migration_primary(VirtIONet *n, MigrationState *s)
3322 {
3323     bool should_be_hidden;
3324     Error *err = NULL;
3325     DeviceState *dev = failover_find_primary_device(n);
3326 
3327     if (!dev) {
3328         return;
3329     }
3330 
3331     should_be_hidden = qatomic_read(&n->failover_primary_hidden);
3332 
3333     if (migration_in_setup(s) && !should_be_hidden) {
3334         if (failover_unplug_primary(n, dev)) {
3335             vmstate_unregister(VMSTATE_IF(dev), qdev_get_vmsd(dev), dev);
3336             qapi_event_send_unplug_primary(dev->id);
3337             qatomic_set(&n->failover_primary_hidden, true);
3338         } else {
3339             warn_report("couldn't unplug primary device");
3340         }
3341     } else if (migration_has_failed(s)) {
3342         /* We already unplugged the device let's plug it back */
3343         if (!failover_replug_primary(n, dev, &err)) {
3344             if (err) {
3345                 error_report_err(err);
3346             }
3347         }
3348     }
3349 }
3350 
3351 static void virtio_net_migration_state_notifier(Notifier *notifier, void *data)
3352 {
3353     MigrationState *s = data;
3354     VirtIONet *n = container_of(notifier, VirtIONet, migration_state);
3355     virtio_net_handle_migration_primary(n, s);
3356 }
3357 
3358 static bool failover_hide_primary_device(DeviceListener *listener,
3359                                          const QDict *device_opts,
3360                                          bool from_json,
3361                                          Error **errp)
3362 {
3363     VirtIONet *n = container_of(listener, VirtIONet, primary_listener);
3364     const char *standby_id;
3365 
3366     if (!device_opts) {
3367         return false;
3368     }
3369 
3370     if (!qdict_haskey(device_opts, "failover_pair_id")) {
3371         return false;
3372     }
3373 
3374     if (!qdict_haskey(device_opts, "id")) {
3375         error_setg(errp, "Device with failover_pair_id needs to have id");
3376         return false;
3377     }
3378 
3379     standby_id = qdict_get_str(device_opts, "failover_pair_id");
3380     if (g_strcmp0(standby_id, n->netclient_name) != 0) {
3381         return false;
3382     }
3383 
3384     /*
3385      * The hide helper can be called several times for a given device.
3386      * Check there is only one primary for a virtio-net device but
3387      * don't duplicate the qdict several times if it's called for the same
3388      * device.
3389      */
3390     if (n->primary_opts) {
3391         const char *old, *new;
3392         /* devices with failover_pair_id always have an id */
3393         old = qdict_get_str(n->primary_opts, "id");
3394         new = qdict_get_str(device_opts, "id");
3395         if (strcmp(old, new) != 0) {
3396             error_setg(errp, "Cannot attach more than one primary device to "
3397                        "'%s': '%s' and '%s'", n->netclient_name, old, new);
3398             return false;
3399         }
3400     } else {
3401         n->primary_opts = qdict_clone_shallow(device_opts);
3402         n->primary_opts_from_json = from_json;
3403     }
3404 
3405     /* failover_primary_hidden is set during feature negotiation */
3406     return qatomic_read(&n->failover_primary_hidden);
3407 }
3408 
3409 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
3410 {
3411     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3412     VirtIONet *n = VIRTIO_NET(dev);
3413     NetClientState *nc;
3414     int i;
3415 
3416     if (n->net_conf.mtu) {
3417         n->host_features |= (1ULL << VIRTIO_NET_F_MTU);
3418     }
3419 
3420     if (n->net_conf.duplex_str) {
3421         if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) {
3422             n->net_conf.duplex = DUPLEX_HALF;
3423         } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) {
3424             n->net_conf.duplex = DUPLEX_FULL;
3425         } else {
3426             error_setg(errp, "'duplex' must be 'half' or 'full'");
3427             return;
3428         }
3429         n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3430     } else {
3431         n->net_conf.duplex = DUPLEX_UNKNOWN;
3432     }
3433 
3434     if (n->net_conf.speed < SPEED_UNKNOWN) {
3435         error_setg(errp, "'speed' must be between 0 and INT_MAX");
3436         return;
3437     }
3438     if (n->net_conf.speed >= 0) {
3439         n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3440     }
3441 
3442     if (n->failover) {
3443         n->primary_listener.hide_device = failover_hide_primary_device;
3444         qatomic_set(&n->failover_primary_hidden, true);
3445         device_listener_register(&n->primary_listener);
3446         n->migration_state.notify = virtio_net_migration_state_notifier;
3447         add_migration_state_change_notifier(&n->migration_state);
3448         n->host_features |= (1ULL << VIRTIO_NET_F_STANDBY);
3449     }
3450 
3451     virtio_net_set_config_size(n, n->host_features);
3452     virtio_init(vdev, VIRTIO_ID_NET, n->config_size);
3453 
3454     /*
3455      * We set a lower limit on RX queue size to what it always was.
3456      * Guests that want a smaller ring can always resize it without
3457      * help from us (using virtio 1 and up).
3458      */
3459     if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
3460         n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
3461         !is_power_of_2(n->net_conf.rx_queue_size)) {
3462         error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
3463                    "must be a power of 2 between %d and %d.",
3464                    n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
3465                    VIRTQUEUE_MAX_SIZE);
3466         virtio_cleanup(vdev);
3467         return;
3468     }
3469 
3470     if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE ||
3471         n->net_conf.tx_queue_size > VIRTQUEUE_MAX_SIZE ||
3472         !is_power_of_2(n->net_conf.tx_queue_size)) {
3473         error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), "
3474                    "must be a power of 2 between %d and %d",
3475                    n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE,
3476                    VIRTQUEUE_MAX_SIZE);
3477         virtio_cleanup(vdev);
3478         return;
3479     }
3480 
3481     n->max_ncs = MAX(n->nic_conf.peers.queues, 1);
3482 
3483     /*
3484      * Figure out the datapath queue pairs since the backend could
3485      * provide control queue via peers as well.
3486      */
3487     if (n->nic_conf.peers.queues) {
3488         for (i = 0; i < n->max_ncs; i++) {
3489             if (n->nic_conf.peers.ncs[i]->is_datapath) {
3490                 ++n->max_queue_pairs;
3491             }
3492         }
3493     }
3494     n->max_queue_pairs = MAX(n->max_queue_pairs, 1);
3495 
3496     if (n->max_queue_pairs * 2 + 1 > VIRTIO_QUEUE_MAX) {
3497         error_setg(errp, "Invalid number of queue pairs (= %" PRIu32 "), "
3498                    "must be a positive integer less than %d.",
3499                    n->max_queue_pairs, (VIRTIO_QUEUE_MAX - 1) / 2);
3500         virtio_cleanup(vdev);
3501         return;
3502     }
3503     n->vqs = g_new0(VirtIONetQueue, n->max_queue_pairs);
3504     n->curr_queue_pairs = 1;
3505     n->tx_timeout = n->net_conf.txtimer;
3506 
3507     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
3508                        && strcmp(n->net_conf.tx, "bh")) {
3509         warn_report("virtio-net: "
3510                     "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
3511                     n->net_conf.tx);
3512         error_printf("Defaulting to \"bh\"");
3513     }
3514 
3515     n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n),
3516                                     n->net_conf.tx_queue_size);
3517 
3518     for (i = 0; i < n->max_queue_pairs; i++) {
3519         virtio_net_add_queue(n, i);
3520     }
3521 
3522     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
3523     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
3524     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
3525     n->status = VIRTIO_NET_S_LINK_UP;
3526     qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
3527                               QEMU_CLOCK_VIRTUAL,
3528                               virtio_net_announce_timer, n);
3529     n->announce_timer.round = 0;
3530 
3531     if (n->netclient_type) {
3532         /*
3533          * Happen when virtio_net_set_netclient_name has been called.
3534          */
3535         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3536                               n->netclient_type, n->netclient_name, n);
3537     } else {
3538         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3539                               object_get_typename(OBJECT(dev)), dev->id, n);
3540     }
3541 
3542     for (i = 0; i < n->max_queue_pairs; i++) {
3543         n->nic->ncs[i].do_not_pad = true;
3544     }
3545 
3546     peer_test_vnet_hdr(n);
3547     if (peer_has_vnet_hdr(n)) {
3548         for (i = 0; i < n->max_queue_pairs; i++) {
3549             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
3550         }
3551         n->host_hdr_len = sizeof(struct virtio_net_hdr);
3552     } else {
3553         n->host_hdr_len = 0;
3554     }
3555 
3556     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
3557 
3558     n->vqs[0].tx_waiting = 0;
3559     n->tx_burst = n->net_conf.txburst;
3560     virtio_net_set_mrg_rx_bufs(n, 0, 0, 0);
3561     n->promisc = 1; /* for compatibility */
3562 
3563     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
3564 
3565     n->vlans = g_malloc0(MAX_VLAN >> 3);
3566 
3567     nc = qemu_get_queue(n->nic);
3568     nc->rxfilter_notify_enabled = 1;
3569 
3570    if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
3571         struct virtio_net_config netcfg = {};
3572         memcpy(&netcfg.mac, &n->nic_conf.macaddr, ETH_ALEN);
3573         vhost_net_set_config(get_vhost_net(nc->peer),
3574             (uint8_t *)&netcfg, 0, ETH_ALEN, VHOST_SET_CONFIG_TYPE_MASTER);
3575     }
3576     QTAILQ_INIT(&n->rsc_chains);
3577     n->qdev = dev;
3578 
3579     net_rx_pkt_init(&n->rx_pkt, false);
3580 
3581     if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) {
3582         virtio_net_load_ebpf(n);
3583     }
3584 }
3585 
3586 static void virtio_net_device_unrealize(DeviceState *dev)
3587 {
3588     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3589     VirtIONet *n = VIRTIO_NET(dev);
3590     int i, max_queue_pairs;
3591 
3592     if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) {
3593         virtio_net_unload_ebpf(n);
3594     }
3595 
3596     /* This will stop vhost backend if appropriate. */
3597     virtio_net_set_status(vdev, 0);
3598 
3599     g_free(n->netclient_name);
3600     n->netclient_name = NULL;
3601     g_free(n->netclient_type);
3602     n->netclient_type = NULL;
3603 
3604     g_free(n->mac_table.macs);
3605     g_free(n->vlans);
3606 
3607     if (n->failover) {
3608         qobject_unref(n->primary_opts);
3609         device_listener_unregister(&n->primary_listener);
3610         remove_migration_state_change_notifier(&n->migration_state);
3611     } else {
3612         assert(n->primary_opts == NULL);
3613     }
3614 
3615     max_queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
3616     for (i = 0; i < max_queue_pairs; i++) {
3617         virtio_net_del_queue(n, i);
3618     }
3619     /* delete also control vq */
3620     virtio_del_queue(vdev, max_queue_pairs * 2);
3621     qemu_announce_timer_del(&n->announce_timer, false);
3622     g_free(n->vqs);
3623     qemu_del_nic(n->nic);
3624     virtio_net_rsc_cleanup(n);
3625     g_free(n->rss_data.indirections_table);
3626     net_rx_pkt_uninit(n->rx_pkt);
3627     virtio_cleanup(vdev);
3628 }
3629 
3630 static void virtio_net_instance_init(Object *obj)
3631 {
3632     VirtIONet *n = VIRTIO_NET(obj);
3633 
3634     /*
3635      * The default config_size is sizeof(struct virtio_net_config).
3636      * Can be overriden with virtio_net_set_config_size.
3637      */
3638     n->config_size = sizeof(struct virtio_net_config);
3639     device_add_bootindex_property(obj, &n->nic_conf.bootindex,
3640                                   "bootindex", "/ethernet-phy@0",
3641                                   DEVICE(n));
3642 
3643     ebpf_rss_init(&n->ebpf_rss);
3644 }
3645 
3646 static int virtio_net_pre_save(void *opaque)
3647 {
3648     VirtIONet *n = opaque;
3649 
3650     /* At this point, backend must be stopped, otherwise
3651      * it might keep writing to memory. */
3652     assert(!n->vhost_started);
3653 
3654     return 0;
3655 }
3656 
3657 static bool primary_unplug_pending(void *opaque)
3658 {
3659     DeviceState *dev = opaque;
3660     DeviceState *primary;
3661     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3662     VirtIONet *n = VIRTIO_NET(vdev);
3663 
3664     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
3665         return false;
3666     }
3667     primary = failover_find_primary_device(n);
3668     return primary ? primary->pending_deleted_event : false;
3669 }
3670 
3671 static bool dev_unplug_pending(void *opaque)
3672 {
3673     DeviceState *dev = opaque;
3674     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3675 
3676     return vdc->primary_unplug_pending(dev);
3677 }
3678 
3679 static struct vhost_dev *virtio_net_get_vhost(VirtIODevice *vdev)
3680 {
3681     VirtIONet *n = VIRTIO_NET(vdev);
3682     NetClientState *nc = qemu_get_queue(n->nic);
3683     struct vhost_net *net = get_vhost_net(nc->peer);
3684     return &net->dev;
3685 }
3686 
3687 static const VMStateDescription vmstate_virtio_net = {
3688     .name = "virtio-net",
3689     .minimum_version_id = VIRTIO_NET_VM_VERSION,
3690     .version_id = VIRTIO_NET_VM_VERSION,
3691     .fields = (VMStateField[]) {
3692         VMSTATE_VIRTIO_DEVICE,
3693         VMSTATE_END_OF_LIST()
3694     },
3695     .pre_save = virtio_net_pre_save,
3696     .dev_unplug_pending = dev_unplug_pending,
3697 };
3698 
3699 static Property virtio_net_properties[] = {
3700     DEFINE_PROP_BIT64("csum", VirtIONet, host_features,
3701                     VIRTIO_NET_F_CSUM, true),
3702     DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features,
3703                     VIRTIO_NET_F_GUEST_CSUM, true),
3704     DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
3705     DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features,
3706                     VIRTIO_NET_F_GUEST_TSO4, true),
3707     DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features,
3708                     VIRTIO_NET_F_GUEST_TSO6, true),
3709     DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features,
3710                     VIRTIO_NET_F_GUEST_ECN, true),
3711     DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features,
3712                     VIRTIO_NET_F_GUEST_UFO, true),
3713     DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features,
3714                     VIRTIO_NET_F_GUEST_ANNOUNCE, true),
3715     DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features,
3716                     VIRTIO_NET_F_HOST_TSO4, true),
3717     DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features,
3718                     VIRTIO_NET_F_HOST_TSO6, true),
3719     DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features,
3720                     VIRTIO_NET_F_HOST_ECN, true),
3721     DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features,
3722                     VIRTIO_NET_F_HOST_UFO, true),
3723     DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features,
3724                     VIRTIO_NET_F_MRG_RXBUF, true),
3725     DEFINE_PROP_BIT64("status", VirtIONet, host_features,
3726                     VIRTIO_NET_F_STATUS, true),
3727     DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features,
3728                     VIRTIO_NET_F_CTRL_VQ, true),
3729     DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features,
3730                     VIRTIO_NET_F_CTRL_RX, true),
3731     DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features,
3732                     VIRTIO_NET_F_CTRL_VLAN, true),
3733     DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features,
3734                     VIRTIO_NET_F_CTRL_RX_EXTRA, true),
3735     DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features,
3736                     VIRTIO_NET_F_CTRL_MAC_ADDR, true),
3737     DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features,
3738                     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
3739     DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
3740     DEFINE_PROP_BIT64("rss", VirtIONet, host_features,
3741                     VIRTIO_NET_F_RSS, false),
3742     DEFINE_PROP_BIT64("hash", VirtIONet, host_features,
3743                     VIRTIO_NET_F_HASH_REPORT, false),
3744     DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features,
3745                     VIRTIO_NET_F_RSC_EXT, false),
3746     DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout,
3747                        VIRTIO_NET_RSC_DEFAULT_INTERVAL),
3748     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
3749     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
3750                        TX_TIMER_INTERVAL),
3751     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
3752     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
3753     DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
3754                        VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
3755     DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size,
3756                        VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE),
3757     DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
3758     DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend,
3759                      true),
3760     DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN),
3761     DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str),
3762     DEFINE_PROP_BOOL("failover", VirtIONet, failover, false),
3763     DEFINE_PROP_END_OF_LIST(),
3764 };
3765 
3766 static void virtio_net_class_init(ObjectClass *klass, void *data)
3767 {
3768     DeviceClass *dc = DEVICE_CLASS(klass);
3769     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3770 
3771     device_class_set_props(dc, virtio_net_properties);
3772     dc->vmsd = &vmstate_virtio_net;
3773     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
3774     vdc->realize = virtio_net_device_realize;
3775     vdc->unrealize = virtio_net_device_unrealize;
3776     vdc->get_config = virtio_net_get_config;
3777     vdc->set_config = virtio_net_set_config;
3778     vdc->get_features = virtio_net_get_features;
3779     vdc->set_features = virtio_net_set_features;
3780     vdc->bad_features = virtio_net_bad_features;
3781     vdc->reset = virtio_net_reset;
3782     vdc->set_status = virtio_net_set_status;
3783     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
3784     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
3785     vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
3786     vdc->post_load = virtio_net_post_load_virtio;
3787     vdc->vmsd = &vmstate_virtio_net_device;
3788     vdc->primary_unplug_pending = primary_unplug_pending;
3789     vdc->get_vhost = virtio_net_get_vhost;
3790 }
3791 
3792 static const TypeInfo virtio_net_info = {
3793     .name = TYPE_VIRTIO_NET,
3794     .parent = TYPE_VIRTIO_DEVICE,
3795     .instance_size = sizeof(VirtIONet),
3796     .instance_init = virtio_net_instance_init,
3797     .class_init = virtio_net_class_init,
3798 };
3799 
3800 static void virtio_register_types(void)
3801 {
3802     type_register_static(&virtio_net_info);
3803 }
3804 
3805 type_init(virtio_register_types)
3806