xref: /openbmc/qemu/hw/net/virtio-net.c (revision 6016b7b4)
1 /*
2  * Virtio Network Device
3  *
4  * Copyright IBM, Corp. 2007
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/osdep.h"
15 #include "qemu/atomic.h"
16 #include "qemu/iov.h"
17 #include "qemu/main-loop.h"
18 #include "qemu/module.h"
19 #include "hw/virtio/virtio.h"
20 #include "net/net.h"
21 #include "net/checksum.h"
22 #include "net/tap.h"
23 #include "qemu/error-report.h"
24 #include "qemu/timer.h"
25 #include "qemu/option.h"
26 #include "qemu/option_int.h"
27 #include "qemu/config-file.h"
28 #include "qapi/qmp/qdict.h"
29 #include "hw/virtio/virtio-net.h"
30 #include "net/vhost_net.h"
31 #include "net/announce.h"
32 #include "hw/virtio/virtio-bus.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-events-net.h"
35 #include "hw/qdev-properties.h"
36 #include "qapi/qapi-types-migration.h"
37 #include "qapi/qapi-events-migration.h"
38 #include "hw/virtio/virtio-access.h"
39 #include "migration/misc.h"
40 #include "standard-headers/linux/ethtool.h"
41 #include "sysemu/sysemu.h"
42 #include "trace.h"
43 #include "monitor/qdev.h"
44 #include "hw/pci/pci.h"
45 #include "net_rx_pkt.h"
46 #include "hw/virtio/vhost.h"
47 
48 #define VIRTIO_NET_VM_VERSION    11
49 
50 #define MAC_TABLE_ENTRIES    64
51 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
52 
53 /* previously fixed value */
54 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
55 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
56 
57 /* for now, only allow larger queue_pairs; with virtio-1, guest can downsize */
58 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
59 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
60 
61 #define VIRTIO_NET_IP4_ADDR_SIZE   8        /* ipv4 saddr + daddr */
62 
63 #define VIRTIO_NET_TCP_FLAG         0x3F
64 #define VIRTIO_NET_TCP_HDR_LENGTH   0xF000
65 
66 /* IPv4 max payload, 16 bits in the header */
67 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
68 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
69 
70 /* header length value in ip header without option */
71 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
72 
73 #define VIRTIO_NET_IP6_ADDR_SIZE   32      /* ipv6 saddr + daddr */
74 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
75 
76 /* Purge coalesced packets timer interval, This value affects the performance
77    a lot, and should be tuned carefully, '300000'(300us) is the recommended
78    value to pass the WHQL test, '50000' can gain 2x netperf throughput with
79    tso/gso/gro 'off'. */
80 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
81 
82 #define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \
83                                          VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \
84                                          VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \
85                                          VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \
86                                          VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \
87                                          VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \
88                                          VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \
89                                          VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \
90                                          VIRTIO_NET_RSS_HASH_TYPE_UDP_EX)
91 
92 static const VirtIOFeature feature_sizes[] = {
93     {.flags = 1ULL << VIRTIO_NET_F_MAC,
94      .end = endof(struct virtio_net_config, mac)},
95     {.flags = 1ULL << VIRTIO_NET_F_STATUS,
96      .end = endof(struct virtio_net_config, status)},
97     {.flags = 1ULL << VIRTIO_NET_F_MQ,
98      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
99     {.flags = 1ULL << VIRTIO_NET_F_MTU,
100      .end = endof(struct virtio_net_config, mtu)},
101     {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX,
102      .end = endof(struct virtio_net_config, duplex)},
103     {.flags = (1ULL << VIRTIO_NET_F_RSS) | (1ULL << VIRTIO_NET_F_HASH_REPORT),
104      .end = endof(struct virtio_net_config, supported_hash_types)},
105     {}
106 };
107 
108 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
109 {
110     VirtIONet *n = qemu_get_nic_opaque(nc);
111 
112     return &n->vqs[nc->queue_index];
113 }
114 
115 static int vq2q(int queue_index)
116 {
117     return queue_index / 2;
118 }
119 
120 /* TODO
121  * - we could suppress RX interrupt if we were so inclined.
122  */
123 
124 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
125 {
126     VirtIONet *n = VIRTIO_NET(vdev);
127     struct virtio_net_config netcfg;
128     NetClientState *nc = qemu_get_queue(n->nic);
129     static const MACAddr zero = { .a = { 0, 0, 0, 0, 0, 0 } };
130 
131     int ret = 0;
132     memset(&netcfg, 0 , sizeof(struct virtio_net_config));
133     virtio_stw_p(vdev, &netcfg.status, n->status);
134     virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queue_pairs);
135     virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
136     memcpy(netcfg.mac, n->mac, ETH_ALEN);
137     virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed);
138     netcfg.duplex = n->net_conf.duplex;
139     netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE;
140     virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length,
141                  virtio_host_has_feature(vdev, VIRTIO_NET_F_RSS) ?
142                  VIRTIO_NET_RSS_MAX_TABLE_LEN : 1);
143     virtio_stl_p(vdev, &netcfg.supported_hash_types,
144                  VIRTIO_NET_RSS_SUPPORTED_HASHES);
145     memcpy(config, &netcfg, n->config_size);
146 
147     /*
148      * Is this VDPA? No peer means not VDPA: there's no way to
149      * disconnect/reconnect a VDPA peer.
150      */
151     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
152         ret = vhost_net_get_config(get_vhost_net(nc->peer), (uint8_t *)&netcfg,
153                                    n->config_size);
154         if (ret != -1) {
155             /*
156              * Some NIC/kernel combinations present 0 as the mac address.  As
157              * that is not a legal address, try to proceed with the
158              * address from the QEMU command line in the hope that the
159              * address has been configured correctly elsewhere - just not
160              * reported by the device.
161              */
162             if (memcmp(&netcfg.mac, &zero, sizeof(zero)) == 0) {
163                 info_report("Zero hardware mac address detected. Ignoring.");
164                 memcpy(netcfg.mac, n->mac, ETH_ALEN);
165             }
166             memcpy(config, &netcfg, n->config_size);
167         }
168     }
169 }
170 
171 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
172 {
173     VirtIONet *n = VIRTIO_NET(vdev);
174     struct virtio_net_config netcfg = {};
175     NetClientState *nc = qemu_get_queue(n->nic);
176 
177     memcpy(&netcfg, config, n->config_size);
178 
179     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
180         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
181         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
182         memcpy(n->mac, netcfg.mac, ETH_ALEN);
183         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
184     }
185 
186     /*
187      * Is this VDPA? No peer means not VDPA: there's no way to
188      * disconnect/reconnect a VDPA peer.
189      */
190     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
191         vhost_net_set_config(get_vhost_net(nc->peer),
192                              (uint8_t *)&netcfg, 0, n->config_size,
193                              VHOST_SET_CONFIG_TYPE_MASTER);
194       }
195 }
196 
197 static bool virtio_net_started(VirtIONet *n, uint8_t status)
198 {
199     VirtIODevice *vdev = VIRTIO_DEVICE(n);
200     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
201         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
202 }
203 
204 static void virtio_net_announce_notify(VirtIONet *net)
205 {
206     VirtIODevice *vdev = VIRTIO_DEVICE(net);
207     trace_virtio_net_announce_notify();
208 
209     net->status |= VIRTIO_NET_S_ANNOUNCE;
210     virtio_notify_config(vdev);
211 }
212 
213 static void virtio_net_announce_timer(void *opaque)
214 {
215     VirtIONet *n = opaque;
216     trace_virtio_net_announce_timer(n->announce_timer.round);
217 
218     n->announce_timer.round--;
219     virtio_net_announce_notify(n);
220 }
221 
222 static void virtio_net_announce(NetClientState *nc)
223 {
224     VirtIONet *n = qemu_get_nic_opaque(nc);
225     VirtIODevice *vdev = VIRTIO_DEVICE(n);
226 
227     /*
228      * Make sure the virtio migration announcement timer isn't running
229      * If it is, let it trigger announcement so that we do not cause
230      * confusion.
231      */
232     if (n->announce_timer.round) {
233         return;
234     }
235 
236     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
237         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
238             virtio_net_announce_notify(n);
239     }
240 }
241 
242 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
243 {
244     VirtIODevice *vdev = VIRTIO_DEVICE(n);
245     NetClientState *nc = qemu_get_queue(n->nic);
246     int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
247     int cvq = n->max_ncs - n->max_queue_pairs;
248 
249     if (!get_vhost_net(nc->peer)) {
250         return;
251     }
252 
253     if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
254         !!n->vhost_started) {
255         return;
256     }
257     if (!n->vhost_started) {
258         int r, i;
259 
260         if (n->needs_vnet_hdr_swap) {
261             error_report("backend does not support %s vnet headers; "
262                          "falling back on userspace virtio",
263                          virtio_is_big_endian(vdev) ? "BE" : "LE");
264             return;
265         }
266 
267         /* Any packets outstanding? Purge them to avoid touching rings
268          * when vhost is running.
269          */
270         for (i = 0;  i < queue_pairs; i++) {
271             NetClientState *qnc = qemu_get_subqueue(n->nic, i);
272 
273             /* Purge both directions: TX and RX. */
274             qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
275             qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
276         }
277 
278         if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
279             r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
280             if (r < 0) {
281                 error_report("%uBytes MTU not supported by the backend",
282                              n->net_conf.mtu);
283 
284                 return;
285             }
286         }
287 
288         n->vhost_started = 1;
289         r = vhost_net_start(vdev, n->nic->ncs, queue_pairs, cvq);
290         if (r < 0) {
291             error_report("unable to start vhost net: %d: "
292                          "falling back on userspace virtio", -r);
293             n->vhost_started = 0;
294         }
295     } else {
296         vhost_net_stop(vdev, n->nic->ncs, queue_pairs, cvq);
297         n->vhost_started = 0;
298     }
299 }
300 
301 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
302                                           NetClientState *peer,
303                                           bool enable)
304 {
305     if (virtio_is_big_endian(vdev)) {
306         return qemu_set_vnet_be(peer, enable);
307     } else {
308         return qemu_set_vnet_le(peer, enable);
309     }
310 }
311 
312 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
313                                        int queue_pairs, bool enable)
314 {
315     int i;
316 
317     for (i = 0; i < queue_pairs; i++) {
318         if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
319             enable) {
320             while (--i >= 0) {
321                 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
322             }
323 
324             return true;
325         }
326     }
327 
328     return false;
329 }
330 
331 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
332 {
333     VirtIODevice *vdev = VIRTIO_DEVICE(n);
334     int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
335 
336     if (virtio_net_started(n, status)) {
337         /* Before using the device, we tell the network backend about the
338          * endianness to use when parsing vnet headers. If the backend
339          * can't do it, we fallback onto fixing the headers in the core
340          * virtio-net code.
341          */
342         n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
343                                                             queue_pairs, true);
344     } else if (virtio_net_started(n, vdev->status)) {
345         /* After using the device, we need to reset the network backend to
346          * the default (guest native endianness), otherwise the guest may
347          * lose network connectivity if it is rebooted into a different
348          * endianness.
349          */
350         virtio_net_set_vnet_endian(vdev, n->nic->ncs, queue_pairs, false);
351     }
352 }
353 
354 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq)
355 {
356     unsigned int dropped = virtqueue_drop_all(vq);
357     if (dropped) {
358         virtio_notify(vdev, vq);
359     }
360 }
361 
362 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
363 {
364     VirtIONet *n = VIRTIO_NET(vdev);
365     VirtIONetQueue *q;
366     int i;
367     uint8_t queue_status;
368 
369     virtio_net_vnet_endian_status(n, status);
370     virtio_net_vhost_status(n, status);
371 
372     for (i = 0; i < n->max_queue_pairs; i++) {
373         NetClientState *ncs = qemu_get_subqueue(n->nic, i);
374         bool queue_started;
375         q = &n->vqs[i];
376 
377         if ((!n->multiqueue && i != 0) || i >= n->curr_queue_pairs) {
378             queue_status = 0;
379         } else {
380             queue_status = status;
381         }
382         queue_started =
383             virtio_net_started(n, queue_status) && !n->vhost_started;
384 
385         if (queue_started) {
386             qemu_flush_queued_packets(ncs);
387         }
388 
389         if (!q->tx_waiting) {
390             continue;
391         }
392 
393         if (queue_started) {
394             if (q->tx_timer) {
395                 timer_mod(q->tx_timer,
396                                qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
397             } else {
398                 qemu_bh_schedule(q->tx_bh);
399             }
400         } else {
401             if (q->tx_timer) {
402                 timer_del(q->tx_timer);
403             } else {
404                 qemu_bh_cancel(q->tx_bh);
405             }
406             if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 &&
407                 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) &&
408                 vdev->vm_running) {
409                 /* if tx is waiting we are likely have some packets in tx queue
410                  * and disabled notification */
411                 q->tx_waiting = 0;
412                 virtio_queue_set_notification(q->tx_vq, 1);
413                 virtio_net_drop_tx_queue_data(vdev, q->tx_vq);
414             }
415         }
416     }
417 }
418 
419 static void virtio_net_set_link_status(NetClientState *nc)
420 {
421     VirtIONet *n = qemu_get_nic_opaque(nc);
422     VirtIODevice *vdev = VIRTIO_DEVICE(n);
423     uint16_t old_status = n->status;
424 
425     if (nc->link_down)
426         n->status &= ~VIRTIO_NET_S_LINK_UP;
427     else
428         n->status |= VIRTIO_NET_S_LINK_UP;
429 
430     if (n->status != old_status)
431         virtio_notify_config(vdev);
432 
433     virtio_net_set_status(vdev, vdev->status);
434 }
435 
436 static void rxfilter_notify(NetClientState *nc)
437 {
438     VirtIONet *n = qemu_get_nic_opaque(nc);
439 
440     if (nc->rxfilter_notify_enabled) {
441         char *path = object_get_canonical_path(OBJECT(n->qdev));
442         qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
443                                               n->netclient_name, path);
444         g_free(path);
445 
446         /* disable event notification to avoid events flooding */
447         nc->rxfilter_notify_enabled = 0;
448     }
449 }
450 
451 static intList *get_vlan_table(VirtIONet *n)
452 {
453     intList *list;
454     int i, j;
455 
456     list = NULL;
457     for (i = 0; i < MAX_VLAN >> 5; i++) {
458         for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
459             if (n->vlans[i] & (1U << j)) {
460                 QAPI_LIST_PREPEND(list, (i << 5) + j);
461             }
462         }
463     }
464 
465     return list;
466 }
467 
468 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
469 {
470     VirtIONet *n = qemu_get_nic_opaque(nc);
471     VirtIODevice *vdev = VIRTIO_DEVICE(n);
472     RxFilterInfo *info;
473     strList *str_list;
474     int i;
475 
476     info = g_malloc0(sizeof(*info));
477     info->name = g_strdup(nc->name);
478     info->promiscuous = n->promisc;
479 
480     if (n->nouni) {
481         info->unicast = RX_STATE_NONE;
482     } else if (n->alluni) {
483         info->unicast = RX_STATE_ALL;
484     } else {
485         info->unicast = RX_STATE_NORMAL;
486     }
487 
488     if (n->nomulti) {
489         info->multicast = RX_STATE_NONE;
490     } else if (n->allmulti) {
491         info->multicast = RX_STATE_ALL;
492     } else {
493         info->multicast = RX_STATE_NORMAL;
494     }
495 
496     info->broadcast_allowed = n->nobcast;
497     info->multicast_overflow = n->mac_table.multi_overflow;
498     info->unicast_overflow = n->mac_table.uni_overflow;
499 
500     info->main_mac = qemu_mac_strdup_printf(n->mac);
501 
502     str_list = NULL;
503     for (i = 0; i < n->mac_table.first_multi; i++) {
504         QAPI_LIST_PREPEND(str_list,
505                       qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN));
506     }
507     info->unicast_table = str_list;
508 
509     str_list = NULL;
510     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
511         QAPI_LIST_PREPEND(str_list,
512                       qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN));
513     }
514     info->multicast_table = str_list;
515     info->vlan_table = get_vlan_table(n);
516 
517     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
518         info->vlan = RX_STATE_ALL;
519     } else if (!info->vlan_table) {
520         info->vlan = RX_STATE_NONE;
521     } else {
522         info->vlan = RX_STATE_NORMAL;
523     }
524 
525     /* enable event notification after query */
526     nc->rxfilter_notify_enabled = 1;
527 
528     return info;
529 }
530 
531 static void virtio_net_reset(VirtIODevice *vdev)
532 {
533     VirtIONet *n = VIRTIO_NET(vdev);
534     int i;
535 
536     /* Reset back to compatibility mode */
537     n->promisc = 1;
538     n->allmulti = 0;
539     n->alluni = 0;
540     n->nomulti = 0;
541     n->nouni = 0;
542     n->nobcast = 0;
543     /* multiqueue is disabled by default */
544     n->curr_queue_pairs = 1;
545     timer_del(n->announce_timer.tm);
546     n->announce_timer.round = 0;
547     n->status &= ~VIRTIO_NET_S_ANNOUNCE;
548 
549     /* Flush any MAC and VLAN filter table state */
550     n->mac_table.in_use = 0;
551     n->mac_table.first_multi = 0;
552     n->mac_table.multi_overflow = 0;
553     n->mac_table.uni_overflow = 0;
554     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
555     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
556     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
557     memset(n->vlans, 0, MAX_VLAN >> 3);
558 
559     /* Flush any async TX */
560     for (i = 0;  i < n->max_queue_pairs; i++) {
561         NetClientState *nc = qemu_get_subqueue(n->nic, i);
562 
563         if (nc->peer) {
564             qemu_flush_or_purge_queued_packets(nc->peer, true);
565             assert(!virtio_net_get_subqueue(nc)->async_tx.elem);
566         }
567     }
568 }
569 
570 static void peer_test_vnet_hdr(VirtIONet *n)
571 {
572     NetClientState *nc = qemu_get_queue(n->nic);
573     if (!nc->peer) {
574         return;
575     }
576 
577     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
578 }
579 
580 static int peer_has_vnet_hdr(VirtIONet *n)
581 {
582     return n->has_vnet_hdr;
583 }
584 
585 static int peer_has_ufo(VirtIONet *n)
586 {
587     if (!peer_has_vnet_hdr(n))
588         return 0;
589 
590     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
591 
592     return n->has_ufo;
593 }
594 
595 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
596                                        int version_1, int hash_report)
597 {
598     int i;
599     NetClientState *nc;
600 
601     n->mergeable_rx_bufs = mergeable_rx_bufs;
602 
603     if (version_1) {
604         n->guest_hdr_len = hash_report ?
605             sizeof(struct virtio_net_hdr_v1_hash) :
606             sizeof(struct virtio_net_hdr_mrg_rxbuf);
607         n->rss_data.populate_hash = !!hash_report;
608     } else {
609         n->guest_hdr_len = n->mergeable_rx_bufs ?
610             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
611             sizeof(struct virtio_net_hdr);
612     }
613 
614     for (i = 0; i < n->max_queue_pairs; i++) {
615         nc = qemu_get_subqueue(n->nic, i);
616 
617         if (peer_has_vnet_hdr(n) &&
618             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
619             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
620             n->host_hdr_len = n->guest_hdr_len;
621         }
622     }
623 }
624 
625 static int virtio_net_max_tx_queue_size(VirtIONet *n)
626 {
627     NetClientState *peer = n->nic_conf.peers.ncs[0];
628 
629     /*
630      * Backends other than vhost-user don't support max queue size.
631      */
632     if (!peer) {
633         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
634     }
635 
636     if (peer->info->type != NET_CLIENT_DRIVER_VHOST_USER) {
637         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
638     }
639 
640     return VIRTQUEUE_MAX_SIZE;
641 }
642 
643 static int peer_attach(VirtIONet *n, int index)
644 {
645     NetClientState *nc = qemu_get_subqueue(n->nic, index);
646 
647     if (!nc->peer) {
648         return 0;
649     }
650 
651     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
652         vhost_set_vring_enable(nc->peer, 1);
653     }
654 
655     if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
656         return 0;
657     }
658 
659     if (n->max_queue_pairs == 1) {
660         return 0;
661     }
662 
663     return tap_enable(nc->peer);
664 }
665 
666 static int peer_detach(VirtIONet *n, int index)
667 {
668     NetClientState *nc = qemu_get_subqueue(n->nic, index);
669 
670     if (!nc->peer) {
671         return 0;
672     }
673 
674     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
675         vhost_set_vring_enable(nc->peer, 0);
676     }
677 
678     if (nc->peer->info->type !=  NET_CLIENT_DRIVER_TAP) {
679         return 0;
680     }
681 
682     return tap_disable(nc->peer);
683 }
684 
685 static void virtio_net_set_queue_pairs(VirtIONet *n)
686 {
687     int i;
688     int r;
689 
690     if (n->nic->peer_deleted) {
691         return;
692     }
693 
694     for (i = 0; i < n->max_queue_pairs; i++) {
695         if (i < n->curr_queue_pairs) {
696             r = peer_attach(n, i);
697             assert(!r);
698         } else {
699             r = peer_detach(n, i);
700             assert(!r);
701         }
702     }
703 }
704 
705 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
706 
707 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
708                                         Error **errp)
709 {
710     VirtIONet *n = VIRTIO_NET(vdev);
711     NetClientState *nc = qemu_get_queue(n->nic);
712 
713     /* Firstly sync all virtio-net possible supported features */
714     features |= n->host_features;
715 
716     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
717 
718     if (!peer_has_vnet_hdr(n)) {
719         virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
720         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
721         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
722         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
723 
724         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
725         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
726         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
727         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
728 
729         virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT);
730     }
731 
732     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
733         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
734         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
735     }
736 
737     if (!get_vhost_net(nc->peer)) {
738         return features;
739     }
740 
741     if (!ebpf_rss_is_loaded(&n->ebpf_rss)) {
742         virtio_clear_feature(&features, VIRTIO_NET_F_RSS);
743     }
744     features = vhost_net_get_features(get_vhost_net(nc->peer), features);
745     vdev->backend_features = features;
746 
747     if (n->mtu_bypass_backend &&
748             (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) {
749         features |= (1ULL << VIRTIO_NET_F_MTU);
750     }
751 
752     return features;
753 }
754 
755 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
756 {
757     uint64_t features = 0;
758 
759     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
760      * but also these: */
761     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
762     virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
763     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
764     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
765     virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
766 
767     return features;
768 }
769 
770 static void virtio_net_apply_guest_offloads(VirtIONet *n)
771 {
772     qemu_set_offload(qemu_get_queue(n->nic)->peer,
773             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
774             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
775             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
776             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
777             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
778 }
779 
780 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
781 {
782     static const uint64_t guest_offloads_mask =
783         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
784         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
785         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
786         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
787         (1ULL << VIRTIO_NET_F_GUEST_UFO);
788 
789     return guest_offloads_mask & features;
790 }
791 
792 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
793 {
794     VirtIODevice *vdev = VIRTIO_DEVICE(n);
795     return virtio_net_guest_offloads_by_features(vdev->guest_features);
796 }
797 
798 typedef struct {
799     VirtIONet *n;
800     DeviceState *dev;
801 } FailoverDevice;
802 
803 /**
804  * Set the failover primary device
805  *
806  * @opaque: FailoverId to setup
807  * @opts: opts for device we are handling
808  * @errp: returns an error if this function fails
809  */
810 static int failover_set_primary(DeviceState *dev, void *opaque)
811 {
812     FailoverDevice *fdev = opaque;
813     PCIDevice *pci_dev = (PCIDevice *)
814         object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE);
815 
816     if (!pci_dev) {
817         return 0;
818     }
819 
820     if (!g_strcmp0(pci_dev->failover_pair_id, fdev->n->netclient_name)) {
821         fdev->dev = dev;
822         return 1;
823     }
824 
825     return 0;
826 }
827 
828 /**
829  * Find the primary device for this failover virtio-net
830  *
831  * @n: VirtIONet device
832  * @errp: returns an error if this function fails
833  */
834 static DeviceState *failover_find_primary_device(VirtIONet *n)
835 {
836     FailoverDevice fdev = {
837         .n = n,
838     };
839 
840     qbus_walk_children(sysbus_get_default(), failover_set_primary, NULL,
841                        NULL, NULL, &fdev);
842     return fdev.dev;
843 }
844 
845 static void failover_add_primary(VirtIONet *n, Error **errp)
846 {
847     Error *err = NULL;
848     DeviceState *dev = failover_find_primary_device(n);
849 
850     if (dev) {
851         return;
852     }
853 
854     if (!n->primary_opts) {
855         error_setg(errp, "Primary device not found");
856         error_append_hint(errp, "Virtio-net failover will not work. Make "
857                           "sure primary device has parameter"
858                           " failover_pair_id=%s\n", n->netclient_name);
859         return;
860     }
861 
862     dev = qdev_device_add_from_qdict(n->primary_opts,
863                                      n->primary_opts_from_json,
864                                      &err);
865     if (err) {
866         qobject_unref(n->primary_opts);
867         n->primary_opts = NULL;
868     } else {
869         object_unref(OBJECT(dev));
870     }
871     error_propagate(errp, err);
872 }
873 
874 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
875 {
876     VirtIONet *n = VIRTIO_NET(vdev);
877     Error *err = NULL;
878     int i;
879 
880     if (n->mtu_bypass_backend &&
881             !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) {
882         features &= ~(1ULL << VIRTIO_NET_F_MTU);
883     }
884 
885     virtio_net_set_multiqueue(n,
886                               virtio_has_feature(features, VIRTIO_NET_F_RSS) ||
887                               virtio_has_feature(features, VIRTIO_NET_F_MQ));
888 
889     virtio_net_set_mrg_rx_bufs(n,
890                                virtio_has_feature(features,
891                                                   VIRTIO_NET_F_MRG_RXBUF),
892                                virtio_has_feature(features,
893                                                   VIRTIO_F_VERSION_1),
894                                virtio_has_feature(features,
895                                                   VIRTIO_NET_F_HASH_REPORT));
896 
897     n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
898         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4);
899     n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
900         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6);
901     n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS);
902 
903     if (n->has_vnet_hdr) {
904         n->curr_guest_offloads =
905             virtio_net_guest_offloads_by_features(features);
906         virtio_net_apply_guest_offloads(n);
907     }
908 
909     for (i = 0;  i < n->max_queue_pairs; i++) {
910         NetClientState *nc = qemu_get_subqueue(n->nic, i);
911 
912         if (!get_vhost_net(nc->peer)) {
913             continue;
914         }
915         vhost_net_ack_features(get_vhost_net(nc->peer), features);
916     }
917 
918     if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
919         memset(n->vlans, 0, MAX_VLAN >> 3);
920     } else {
921         memset(n->vlans, 0xff, MAX_VLAN >> 3);
922     }
923 
924     if (virtio_has_feature(features, VIRTIO_NET_F_STANDBY)) {
925         qapi_event_send_failover_negotiated(n->netclient_name);
926         qatomic_set(&n->failover_primary_hidden, false);
927         failover_add_primary(n, &err);
928         if (err) {
929             warn_report_err(err);
930         }
931     }
932 }
933 
934 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
935                                      struct iovec *iov, unsigned int iov_cnt)
936 {
937     uint8_t on;
938     size_t s;
939     NetClientState *nc = qemu_get_queue(n->nic);
940 
941     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
942     if (s != sizeof(on)) {
943         return VIRTIO_NET_ERR;
944     }
945 
946     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
947         n->promisc = on;
948     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
949         n->allmulti = on;
950     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
951         n->alluni = on;
952     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
953         n->nomulti = on;
954     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
955         n->nouni = on;
956     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
957         n->nobcast = on;
958     } else {
959         return VIRTIO_NET_ERR;
960     }
961 
962     rxfilter_notify(nc);
963 
964     return VIRTIO_NET_OK;
965 }
966 
967 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
968                                      struct iovec *iov, unsigned int iov_cnt)
969 {
970     VirtIODevice *vdev = VIRTIO_DEVICE(n);
971     uint64_t offloads;
972     size_t s;
973 
974     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
975         return VIRTIO_NET_ERR;
976     }
977 
978     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
979     if (s != sizeof(offloads)) {
980         return VIRTIO_NET_ERR;
981     }
982 
983     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
984         uint64_t supported_offloads;
985 
986         offloads = virtio_ldq_p(vdev, &offloads);
987 
988         if (!n->has_vnet_hdr) {
989             return VIRTIO_NET_ERR;
990         }
991 
992         n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
993             virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4);
994         n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
995             virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6);
996         virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT);
997 
998         supported_offloads = virtio_net_supported_guest_offloads(n);
999         if (offloads & ~supported_offloads) {
1000             return VIRTIO_NET_ERR;
1001         }
1002 
1003         n->curr_guest_offloads = offloads;
1004         virtio_net_apply_guest_offloads(n);
1005 
1006         return VIRTIO_NET_OK;
1007     } else {
1008         return VIRTIO_NET_ERR;
1009     }
1010 }
1011 
1012 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
1013                                  struct iovec *iov, unsigned int iov_cnt)
1014 {
1015     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1016     struct virtio_net_ctrl_mac mac_data;
1017     size_t s;
1018     NetClientState *nc = qemu_get_queue(n->nic);
1019 
1020     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
1021         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
1022             return VIRTIO_NET_ERR;
1023         }
1024         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
1025         assert(s == sizeof(n->mac));
1026         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
1027         rxfilter_notify(nc);
1028 
1029         return VIRTIO_NET_OK;
1030     }
1031 
1032     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
1033         return VIRTIO_NET_ERR;
1034     }
1035 
1036     int in_use = 0;
1037     int first_multi = 0;
1038     uint8_t uni_overflow = 0;
1039     uint8_t multi_overflow = 0;
1040     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1041 
1042     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1043                    sizeof(mac_data.entries));
1044     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1045     if (s != sizeof(mac_data.entries)) {
1046         goto error;
1047     }
1048     iov_discard_front(&iov, &iov_cnt, s);
1049 
1050     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
1051         goto error;
1052     }
1053 
1054     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
1055         s = iov_to_buf(iov, iov_cnt, 0, macs,
1056                        mac_data.entries * ETH_ALEN);
1057         if (s != mac_data.entries * ETH_ALEN) {
1058             goto error;
1059         }
1060         in_use += mac_data.entries;
1061     } else {
1062         uni_overflow = 1;
1063     }
1064 
1065     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
1066 
1067     first_multi = in_use;
1068 
1069     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1070                    sizeof(mac_data.entries));
1071     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1072     if (s != sizeof(mac_data.entries)) {
1073         goto error;
1074     }
1075 
1076     iov_discard_front(&iov, &iov_cnt, s);
1077 
1078     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
1079         goto error;
1080     }
1081 
1082     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
1083         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
1084                        mac_data.entries * ETH_ALEN);
1085         if (s != mac_data.entries * ETH_ALEN) {
1086             goto error;
1087         }
1088         in_use += mac_data.entries;
1089     } else {
1090         multi_overflow = 1;
1091     }
1092 
1093     n->mac_table.in_use = in_use;
1094     n->mac_table.first_multi = first_multi;
1095     n->mac_table.uni_overflow = uni_overflow;
1096     n->mac_table.multi_overflow = multi_overflow;
1097     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
1098     g_free(macs);
1099     rxfilter_notify(nc);
1100 
1101     return VIRTIO_NET_OK;
1102 
1103 error:
1104     g_free(macs);
1105     return VIRTIO_NET_ERR;
1106 }
1107 
1108 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
1109                                         struct iovec *iov, unsigned int iov_cnt)
1110 {
1111     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1112     uint16_t vid;
1113     size_t s;
1114     NetClientState *nc = qemu_get_queue(n->nic);
1115 
1116     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
1117     vid = virtio_lduw_p(vdev, &vid);
1118     if (s != sizeof(vid)) {
1119         return VIRTIO_NET_ERR;
1120     }
1121 
1122     if (vid >= MAX_VLAN)
1123         return VIRTIO_NET_ERR;
1124 
1125     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
1126         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
1127     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
1128         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
1129     else
1130         return VIRTIO_NET_ERR;
1131 
1132     rxfilter_notify(nc);
1133 
1134     return VIRTIO_NET_OK;
1135 }
1136 
1137 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
1138                                       struct iovec *iov, unsigned int iov_cnt)
1139 {
1140     trace_virtio_net_handle_announce(n->announce_timer.round);
1141     if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
1142         n->status & VIRTIO_NET_S_ANNOUNCE) {
1143         n->status &= ~VIRTIO_NET_S_ANNOUNCE;
1144         if (n->announce_timer.round) {
1145             qemu_announce_timer_step(&n->announce_timer);
1146         }
1147         return VIRTIO_NET_OK;
1148     } else {
1149         return VIRTIO_NET_ERR;
1150     }
1151 }
1152 
1153 static void virtio_net_detach_epbf_rss(VirtIONet *n);
1154 
1155 static void virtio_net_disable_rss(VirtIONet *n)
1156 {
1157     if (n->rss_data.enabled) {
1158         trace_virtio_net_rss_disable();
1159     }
1160     n->rss_data.enabled = false;
1161 
1162     virtio_net_detach_epbf_rss(n);
1163 }
1164 
1165 static bool virtio_net_attach_ebpf_to_backend(NICState *nic, int prog_fd)
1166 {
1167     NetClientState *nc = qemu_get_peer(qemu_get_queue(nic), 0);
1168     if (nc == NULL || nc->info->set_steering_ebpf == NULL) {
1169         return false;
1170     }
1171 
1172     return nc->info->set_steering_ebpf(nc, prog_fd);
1173 }
1174 
1175 static void rss_data_to_rss_config(struct VirtioNetRssData *data,
1176                                    struct EBPFRSSConfig *config)
1177 {
1178     config->redirect = data->redirect;
1179     config->populate_hash = data->populate_hash;
1180     config->hash_types = data->hash_types;
1181     config->indirections_len = data->indirections_len;
1182     config->default_queue = data->default_queue;
1183 }
1184 
1185 static bool virtio_net_attach_epbf_rss(VirtIONet *n)
1186 {
1187     struct EBPFRSSConfig config = {};
1188 
1189     if (!ebpf_rss_is_loaded(&n->ebpf_rss)) {
1190         return false;
1191     }
1192 
1193     rss_data_to_rss_config(&n->rss_data, &config);
1194 
1195     if (!ebpf_rss_set_all(&n->ebpf_rss, &config,
1196                           n->rss_data.indirections_table, n->rss_data.key)) {
1197         return false;
1198     }
1199 
1200     if (!virtio_net_attach_ebpf_to_backend(n->nic, n->ebpf_rss.program_fd)) {
1201         return false;
1202     }
1203 
1204     return true;
1205 }
1206 
1207 static void virtio_net_detach_epbf_rss(VirtIONet *n)
1208 {
1209     virtio_net_attach_ebpf_to_backend(n->nic, -1);
1210 }
1211 
1212 static bool virtio_net_load_ebpf(VirtIONet *n)
1213 {
1214     if (!virtio_net_attach_ebpf_to_backend(n->nic, -1)) {
1215         /* backend does't support steering ebpf */
1216         return false;
1217     }
1218 
1219     return ebpf_rss_load(&n->ebpf_rss);
1220 }
1221 
1222 static void virtio_net_unload_ebpf(VirtIONet *n)
1223 {
1224     virtio_net_attach_ebpf_to_backend(n->nic, -1);
1225     ebpf_rss_unload(&n->ebpf_rss);
1226 }
1227 
1228 static uint16_t virtio_net_handle_rss(VirtIONet *n,
1229                                       struct iovec *iov,
1230                                       unsigned int iov_cnt,
1231                                       bool do_rss)
1232 {
1233     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1234     struct virtio_net_rss_config cfg;
1235     size_t s, offset = 0, size_get;
1236     uint16_t queue_pairs, i;
1237     struct {
1238         uint16_t us;
1239         uint8_t b;
1240     } QEMU_PACKED temp;
1241     const char *err_msg = "";
1242     uint32_t err_value = 0;
1243 
1244     if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) {
1245         err_msg = "RSS is not negotiated";
1246         goto error;
1247     }
1248     if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) {
1249         err_msg = "Hash report is not negotiated";
1250         goto error;
1251     }
1252     size_get = offsetof(struct virtio_net_rss_config, indirection_table);
1253     s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get);
1254     if (s != size_get) {
1255         err_msg = "Short command buffer";
1256         err_value = (uint32_t)s;
1257         goto error;
1258     }
1259     n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types);
1260     n->rss_data.indirections_len =
1261         virtio_lduw_p(vdev, &cfg.indirection_table_mask);
1262     n->rss_data.indirections_len++;
1263     if (!do_rss) {
1264         n->rss_data.indirections_len = 1;
1265     }
1266     if (!is_power_of_2(n->rss_data.indirections_len)) {
1267         err_msg = "Invalid size of indirection table";
1268         err_value = n->rss_data.indirections_len;
1269         goto error;
1270     }
1271     if (n->rss_data.indirections_len > VIRTIO_NET_RSS_MAX_TABLE_LEN) {
1272         err_msg = "Too large indirection table";
1273         err_value = n->rss_data.indirections_len;
1274         goto error;
1275     }
1276     n->rss_data.default_queue = do_rss ?
1277         virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0;
1278     if (n->rss_data.default_queue >= n->max_queue_pairs) {
1279         err_msg = "Invalid default queue";
1280         err_value = n->rss_data.default_queue;
1281         goto error;
1282     }
1283     offset += size_get;
1284     size_get = sizeof(uint16_t) * n->rss_data.indirections_len;
1285     g_free(n->rss_data.indirections_table);
1286     n->rss_data.indirections_table = g_malloc(size_get);
1287     if (!n->rss_data.indirections_table) {
1288         err_msg = "Can't allocate indirections table";
1289         err_value = n->rss_data.indirections_len;
1290         goto error;
1291     }
1292     s = iov_to_buf(iov, iov_cnt, offset,
1293                    n->rss_data.indirections_table, size_get);
1294     if (s != size_get) {
1295         err_msg = "Short indirection table buffer";
1296         err_value = (uint32_t)s;
1297         goto error;
1298     }
1299     for (i = 0; i < n->rss_data.indirections_len; ++i) {
1300         uint16_t val = n->rss_data.indirections_table[i];
1301         n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val);
1302     }
1303     offset += size_get;
1304     size_get = sizeof(temp);
1305     s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get);
1306     if (s != size_get) {
1307         err_msg = "Can't get queue_pairs";
1308         err_value = (uint32_t)s;
1309         goto error;
1310     }
1311     queue_pairs = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queue_pairs;
1312     if (queue_pairs == 0 || queue_pairs > n->max_queue_pairs) {
1313         err_msg = "Invalid number of queue_pairs";
1314         err_value = queue_pairs;
1315         goto error;
1316     }
1317     if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) {
1318         err_msg = "Invalid key size";
1319         err_value = temp.b;
1320         goto error;
1321     }
1322     if (!temp.b && n->rss_data.hash_types) {
1323         err_msg = "No key provided";
1324         err_value = 0;
1325         goto error;
1326     }
1327     if (!temp.b && !n->rss_data.hash_types) {
1328         virtio_net_disable_rss(n);
1329         return queue_pairs;
1330     }
1331     offset += size_get;
1332     size_get = temp.b;
1333     s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get);
1334     if (s != size_get) {
1335         err_msg = "Can get key buffer";
1336         err_value = (uint32_t)s;
1337         goto error;
1338     }
1339     n->rss_data.enabled = true;
1340 
1341     if (!n->rss_data.populate_hash) {
1342         if (!virtio_net_attach_epbf_rss(n)) {
1343             /* EBPF must be loaded for vhost */
1344             if (get_vhost_net(qemu_get_queue(n->nic)->peer)) {
1345                 warn_report("Can't load eBPF RSS for vhost");
1346                 goto error;
1347             }
1348             /* fallback to software RSS */
1349             warn_report("Can't load eBPF RSS - fallback to software RSS");
1350             n->rss_data.enabled_software_rss = true;
1351         }
1352     } else {
1353         /* use software RSS for hash populating */
1354         /* and detach eBPF if was loaded before */
1355         virtio_net_detach_epbf_rss(n);
1356         n->rss_data.enabled_software_rss = true;
1357     }
1358 
1359     trace_virtio_net_rss_enable(n->rss_data.hash_types,
1360                                 n->rss_data.indirections_len,
1361                                 temp.b);
1362     return queue_pairs;
1363 error:
1364     trace_virtio_net_rss_error(err_msg, err_value);
1365     virtio_net_disable_rss(n);
1366     return 0;
1367 }
1368 
1369 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
1370                                 struct iovec *iov, unsigned int iov_cnt)
1371 {
1372     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1373     uint16_t queue_pairs;
1374 
1375     virtio_net_disable_rss(n);
1376     if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) {
1377         queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, false);
1378         return queue_pairs ? VIRTIO_NET_OK : VIRTIO_NET_ERR;
1379     }
1380     if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
1381         queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, true);
1382     } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
1383         struct virtio_net_ctrl_mq mq;
1384         size_t s;
1385         if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) {
1386             return VIRTIO_NET_ERR;
1387         }
1388         s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
1389         if (s != sizeof(mq)) {
1390             return VIRTIO_NET_ERR;
1391         }
1392         queue_pairs = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
1393 
1394     } else {
1395         return VIRTIO_NET_ERR;
1396     }
1397 
1398     if (queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1399         queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1400         queue_pairs > n->max_queue_pairs ||
1401         !n->multiqueue) {
1402         return VIRTIO_NET_ERR;
1403     }
1404 
1405     n->curr_queue_pairs = queue_pairs;
1406     /* stop the backend before changing the number of queue_pairs to avoid handling a
1407      * disabled queue */
1408     virtio_net_set_status(vdev, vdev->status);
1409     virtio_net_set_queue_pairs(n);
1410 
1411     return VIRTIO_NET_OK;
1412 }
1413 
1414 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
1415 {
1416     VirtIONet *n = VIRTIO_NET(vdev);
1417     struct virtio_net_ctrl_hdr ctrl;
1418     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
1419     VirtQueueElement *elem;
1420     size_t s;
1421     struct iovec *iov, *iov2;
1422     unsigned int iov_cnt;
1423 
1424     for (;;) {
1425         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
1426         if (!elem) {
1427             break;
1428         }
1429         if (iov_size(elem->in_sg, elem->in_num) < sizeof(status) ||
1430             iov_size(elem->out_sg, elem->out_num) < sizeof(ctrl)) {
1431             virtio_error(vdev, "virtio-net ctrl missing headers");
1432             virtqueue_detach_element(vq, elem, 0);
1433             g_free(elem);
1434             break;
1435         }
1436 
1437         iov_cnt = elem->out_num;
1438         iov2 = iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num);
1439         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
1440         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
1441         if (s != sizeof(ctrl)) {
1442             status = VIRTIO_NET_ERR;
1443         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
1444             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
1445         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
1446             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
1447         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
1448             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
1449         } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
1450             status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
1451         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
1452             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
1453         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
1454             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
1455         }
1456 
1457         s = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, sizeof(status));
1458         assert(s == sizeof(status));
1459 
1460         virtqueue_push(vq, elem, sizeof(status));
1461         virtio_notify(vdev, vq);
1462         g_free(iov2);
1463         g_free(elem);
1464     }
1465 }
1466 
1467 /* RX */
1468 
1469 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
1470 {
1471     VirtIONet *n = VIRTIO_NET(vdev);
1472     int queue_index = vq2q(virtio_get_queue_index(vq));
1473 
1474     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
1475 }
1476 
1477 static bool virtio_net_can_receive(NetClientState *nc)
1478 {
1479     VirtIONet *n = qemu_get_nic_opaque(nc);
1480     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1481     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1482 
1483     if (!vdev->vm_running) {
1484         return false;
1485     }
1486 
1487     if (nc->queue_index >= n->curr_queue_pairs) {
1488         return false;
1489     }
1490 
1491     if (!virtio_queue_ready(q->rx_vq) ||
1492         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1493         return false;
1494     }
1495 
1496     return true;
1497 }
1498 
1499 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1500 {
1501     VirtIONet *n = q->n;
1502     if (virtio_queue_empty(q->rx_vq) ||
1503         (n->mergeable_rx_bufs &&
1504          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1505         virtio_queue_set_notification(q->rx_vq, 1);
1506 
1507         /* To avoid a race condition where the guest has made some buffers
1508          * available after the above check but before notification was
1509          * enabled, check for available buffers again.
1510          */
1511         if (virtio_queue_empty(q->rx_vq) ||
1512             (n->mergeable_rx_bufs &&
1513              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1514             return 0;
1515         }
1516     }
1517 
1518     virtio_queue_set_notification(q->rx_vq, 0);
1519     return 1;
1520 }
1521 
1522 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1523 {
1524     virtio_tswap16s(vdev, &hdr->hdr_len);
1525     virtio_tswap16s(vdev, &hdr->gso_size);
1526     virtio_tswap16s(vdev, &hdr->csum_start);
1527     virtio_tswap16s(vdev, &hdr->csum_offset);
1528 }
1529 
1530 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1531  * it never finds out that the packets don't have valid checksums.  This
1532  * causes dhclient to get upset.  Fedora's carried a patch for ages to
1533  * fix this with Xen but it hasn't appeared in an upstream release of
1534  * dhclient yet.
1535  *
1536  * To avoid breaking existing guests, we catch udp packets and add
1537  * checksums.  This is terrible but it's better than hacking the guest
1538  * kernels.
1539  *
1540  * N.B. if we introduce a zero-copy API, this operation is no longer free so
1541  * we should provide a mechanism to disable it to avoid polluting the host
1542  * cache.
1543  */
1544 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1545                                         uint8_t *buf, size_t size)
1546 {
1547     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1548         (size > 27 && size < 1500) && /* normal sized MTU */
1549         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1550         (buf[23] == 17) && /* ip.protocol == UDP */
1551         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1552         net_checksum_calculate(buf, size, CSUM_UDP);
1553         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1554     }
1555 }
1556 
1557 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1558                            const void *buf, size_t size)
1559 {
1560     if (n->has_vnet_hdr) {
1561         /* FIXME this cast is evil */
1562         void *wbuf = (void *)buf;
1563         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1564                                     size - n->host_hdr_len);
1565 
1566         if (n->needs_vnet_hdr_swap) {
1567             virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1568         }
1569         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1570     } else {
1571         struct virtio_net_hdr hdr = {
1572             .flags = 0,
1573             .gso_type = VIRTIO_NET_HDR_GSO_NONE
1574         };
1575         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1576     }
1577 }
1578 
1579 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1580 {
1581     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1582     static const uint8_t vlan[] = {0x81, 0x00};
1583     uint8_t *ptr = (uint8_t *)buf;
1584     int i;
1585 
1586     if (n->promisc)
1587         return 1;
1588 
1589     ptr += n->host_hdr_len;
1590 
1591     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1592         int vid = lduw_be_p(ptr + 14) & 0xfff;
1593         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1594             return 0;
1595     }
1596 
1597     if (ptr[0] & 1) { // multicast
1598         if (!memcmp(ptr, bcast, sizeof(bcast))) {
1599             return !n->nobcast;
1600         } else if (n->nomulti) {
1601             return 0;
1602         } else if (n->allmulti || n->mac_table.multi_overflow) {
1603             return 1;
1604         }
1605 
1606         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1607             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1608                 return 1;
1609             }
1610         }
1611     } else { // unicast
1612         if (n->nouni) {
1613             return 0;
1614         } else if (n->alluni || n->mac_table.uni_overflow) {
1615             return 1;
1616         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1617             return 1;
1618         }
1619 
1620         for (i = 0; i < n->mac_table.first_multi; i++) {
1621             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1622                 return 1;
1623             }
1624         }
1625     }
1626 
1627     return 0;
1628 }
1629 
1630 static uint8_t virtio_net_get_hash_type(bool isip4,
1631                                         bool isip6,
1632                                         bool isudp,
1633                                         bool istcp,
1634                                         uint32_t types)
1635 {
1636     if (isip4) {
1637         if (istcp && (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4)) {
1638             return NetPktRssIpV4Tcp;
1639         }
1640         if (isudp && (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4)) {
1641             return NetPktRssIpV4Udp;
1642         }
1643         if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
1644             return NetPktRssIpV4;
1645         }
1646     } else if (isip6) {
1647         uint32_t mask = VIRTIO_NET_RSS_HASH_TYPE_TCP_EX |
1648                         VIRTIO_NET_RSS_HASH_TYPE_TCPv6;
1649 
1650         if (istcp && (types & mask)) {
1651             return (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) ?
1652                 NetPktRssIpV6TcpEx : NetPktRssIpV6Tcp;
1653         }
1654         mask = VIRTIO_NET_RSS_HASH_TYPE_UDP_EX | VIRTIO_NET_RSS_HASH_TYPE_UDPv6;
1655         if (isudp && (types & mask)) {
1656             return (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) ?
1657                 NetPktRssIpV6UdpEx : NetPktRssIpV6Udp;
1658         }
1659         mask = VIRTIO_NET_RSS_HASH_TYPE_IP_EX | VIRTIO_NET_RSS_HASH_TYPE_IPv6;
1660         if (types & mask) {
1661             return (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) ?
1662                 NetPktRssIpV6Ex : NetPktRssIpV6;
1663         }
1664     }
1665     return 0xff;
1666 }
1667 
1668 static void virtio_set_packet_hash(const uint8_t *buf, uint8_t report,
1669                                    uint32_t hash)
1670 {
1671     struct virtio_net_hdr_v1_hash *hdr = (void *)buf;
1672     hdr->hash_value = hash;
1673     hdr->hash_report = report;
1674 }
1675 
1676 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf,
1677                                   size_t size)
1678 {
1679     VirtIONet *n = qemu_get_nic_opaque(nc);
1680     unsigned int index = nc->queue_index, new_index = index;
1681     struct NetRxPkt *pkt = n->rx_pkt;
1682     uint8_t net_hash_type;
1683     uint32_t hash;
1684     bool isip4, isip6, isudp, istcp;
1685     static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = {
1686         VIRTIO_NET_HASH_REPORT_IPv4,
1687         VIRTIO_NET_HASH_REPORT_TCPv4,
1688         VIRTIO_NET_HASH_REPORT_TCPv6,
1689         VIRTIO_NET_HASH_REPORT_IPv6,
1690         VIRTIO_NET_HASH_REPORT_IPv6_EX,
1691         VIRTIO_NET_HASH_REPORT_TCPv6_EX,
1692         VIRTIO_NET_HASH_REPORT_UDPv4,
1693         VIRTIO_NET_HASH_REPORT_UDPv6,
1694         VIRTIO_NET_HASH_REPORT_UDPv6_EX
1695     };
1696 
1697     net_rx_pkt_set_protocols(pkt, buf + n->host_hdr_len,
1698                              size - n->host_hdr_len);
1699     net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
1700     if (isip4 && (net_rx_pkt_get_ip4_info(pkt)->fragment)) {
1701         istcp = isudp = false;
1702     }
1703     if (isip6 && (net_rx_pkt_get_ip6_info(pkt)->fragment)) {
1704         istcp = isudp = false;
1705     }
1706     net_hash_type = virtio_net_get_hash_type(isip4, isip6, isudp, istcp,
1707                                              n->rss_data.hash_types);
1708     if (net_hash_type > NetPktRssIpV6UdpEx) {
1709         if (n->rss_data.populate_hash) {
1710             virtio_set_packet_hash(buf, VIRTIO_NET_HASH_REPORT_NONE, 0);
1711         }
1712         return n->rss_data.redirect ? n->rss_data.default_queue : -1;
1713     }
1714 
1715     hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key);
1716 
1717     if (n->rss_data.populate_hash) {
1718         virtio_set_packet_hash(buf, reports[net_hash_type], hash);
1719     }
1720 
1721     if (n->rss_data.redirect) {
1722         new_index = hash & (n->rss_data.indirections_len - 1);
1723         new_index = n->rss_data.indirections_table[new_index];
1724     }
1725 
1726     return (index == new_index) ? -1 : new_index;
1727 }
1728 
1729 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
1730                                       size_t size, bool no_rss)
1731 {
1732     VirtIONet *n = qemu_get_nic_opaque(nc);
1733     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1734     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1735     VirtQueueElement *elems[VIRTQUEUE_MAX_SIZE];
1736     size_t lens[VIRTQUEUE_MAX_SIZE];
1737     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1738     struct virtio_net_hdr_mrg_rxbuf mhdr;
1739     unsigned mhdr_cnt = 0;
1740     size_t offset, i, guest_offset, j;
1741     ssize_t err;
1742 
1743     if (!virtio_net_can_receive(nc)) {
1744         return -1;
1745     }
1746 
1747     if (!no_rss && n->rss_data.enabled && n->rss_data.enabled_software_rss) {
1748         int index = virtio_net_process_rss(nc, buf, size);
1749         if (index >= 0) {
1750             NetClientState *nc2 = qemu_get_subqueue(n->nic, index);
1751             return virtio_net_receive_rcu(nc2, buf, size, true);
1752         }
1753     }
1754 
1755     /* hdr_len refers to the header we supply to the guest */
1756     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1757         return 0;
1758     }
1759 
1760     if (!receive_filter(n, buf, size))
1761         return size;
1762 
1763     offset = i = 0;
1764 
1765     while (offset < size) {
1766         VirtQueueElement *elem;
1767         int len, total;
1768         const struct iovec *sg;
1769 
1770         total = 0;
1771 
1772         if (i == VIRTQUEUE_MAX_SIZE) {
1773             virtio_error(vdev, "virtio-net unexpected long buffer chain");
1774             err = size;
1775             goto err;
1776         }
1777 
1778         elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1779         if (!elem) {
1780             if (i) {
1781                 virtio_error(vdev, "virtio-net unexpected empty queue: "
1782                              "i %zd mergeable %d offset %zd, size %zd, "
1783                              "guest hdr len %zd, host hdr len %zd "
1784                              "guest features 0x%" PRIx64,
1785                              i, n->mergeable_rx_bufs, offset, size,
1786                              n->guest_hdr_len, n->host_hdr_len,
1787                              vdev->guest_features);
1788             }
1789             err = -1;
1790             goto err;
1791         }
1792 
1793         if (elem->in_num < 1) {
1794             virtio_error(vdev,
1795                          "virtio-net receive queue contains no in buffers");
1796             virtqueue_detach_element(q->rx_vq, elem, 0);
1797             g_free(elem);
1798             err = -1;
1799             goto err;
1800         }
1801 
1802         sg = elem->in_sg;
1803         if (i == 0) {
1804             assert(offset == 0);
1805             if (n->mergeable_rx_bufs) {
1806                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1807                                     sg, elem->in_num,
1808                                     offsetof(typeof(mhdr), num_buffers),
1809                                     sizeof(mhdr.num_buffers));
1810             }
1811 
1812             receive_header(n, sg, elem->in_num, buf, size);
1813             if (n->rss_data.populate_hash) {
1814                 offset = sizeof(mhdr);
1815                 iov_from_buf(sg, elem->in_num, offset,
1816                              buf + offset, n->host_hdr_len - sizeof(mhdr));
1817             }
1818             offset = n->host_hdr_len;
1819             total += n->guest_hdr_len;
1820             guest_offset = n->guest_hdr_len;
1821         } else {
1822             guest_offset = 0;
1823         }
1824 
1825         /* copy in packet.  ugh */
1826         len = iov_from_buf(sg, elem->in_num, guest_offset,
1827                            buf + offset, size - offset);
1828         total += len;
1829         offset += len;
1830         /* If buffers can't be merged, at this point we
1831          * must have consumed the complete packet.
1832          * Otherwise, drop it. */
1833         if (!n->mergeable_rx_bufs && offset < size) {
1834             virtqueue_unpop(q->rx_vq, elem, total);
1835             g_free(elem);
1836             err = size;
1837             goto err;
1838         }
1839 
1840         elems[i] = elem;
1841         lens[i] = total;
1842         i++;
1843     }
1844 
1845     if (mhdr_cnt) {
1846         virtio_stw_p(vdev, &mhdr.num_buffers, i);
1847         iov_from_buf(mhdr_sg, mhdr_cnt,
1848                      0,
1849                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1850     }
1851 
1852     for (j = 0; j < i; j++) {
1853         /* signal other side */
1854         virtqueue_fill(q->rx_vq, elems[j], lens[j], j);
1855         g_free(elems[j]);
1856     }
1857 
1858     virtqueue_flush(q->rx_vq, i);
1859     virtio_notify(vdev, q->rx_vq);
1860 
1861     return size;
1862 
1863 err:
1864     for (j = 0; j < i; j++) {
1865         g_free(elems[j]);
1866     }
1867 
1868     return err;
1869 }
1870 
1871 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf,
1872                                   size_t size)
1873 {
1874     RCU_READ_LOCK_GUARD();
1875 
1876     return virtio_net_receive_rcu(nc, buf, size, false);
1877 }
1878 
1879 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain,
1880                                          const uint8_t *buf,
1881                                          VirtioNetRscUnit *unit)
1882 {
1883     uint16_t ip_hdrlen;
1884     struct ip_header *ip;
1885 
1886     ip = (struct ip_header *)(buf + chain->n->guest_hdr_len
1887                               + sizeof(struct eth_header));
1888     unit->ip = (void *)ip;
1889     ip_hdrlen = (ip->ip_ver_len & 0xF) << 2;
1890     unit->ip_plen = &ip->ip_len;
1891     unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen);
1892     unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1893     unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen;
1894 }
1895 
1896 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain,
1897                                          const uint8_t *buf,
1898                                          VirtioNetRscUnit *unit)
1899 {
1900     struct ip6_header *ip6;
1901 
1902     ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len
1903                                  + sizeof(struct eth_header));
1904     unit->ip = ip6;
1905     unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
1906     unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip)
1907                                         + sizeof(struct ip6_header));
1908     unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1909 
1910     /* There is a difference between payload lenght in ipv4 and v6,
1911        ip header is excluded in ipv6 */
1912     unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen;
1913 }
1914 
1915 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain,
1916                                        VirtioNetRscSeg *seg)
1917 {
1918     int ret;
1919     struct virtio_net_hdr_v1 *h;
1920 
1921     h = (struct virtio_net_hdr_v1 *)seg->buf;
1922     h->flags = 0;
1923     h->gso_type = VIRTIO_NET_HDR_GSO_NONE;
1924 
1925     if (seg->is_coalesced) {
1926         h->rsc.segments = seg->packets;
1927         h->rsc.dup_acks = seg->dup_ack;
1928         h->flags = VIRTIO_NET_HDR_F_RSC_INFO;
1929         if (chain->proto == ETH_P_IP) {
1930             h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1931         } else {
1932             h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1933         }
1934     }
1935 
1936     ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size);
1937     QTAILQ_REMOVE(&chain->buffers, seg, next);
1938     g_free(seg->buf);
1939     g_free(seg);
1940 
1941     return ret;
1942 }
1943 
1944 static void virtio_net_rsc_purge(void *opq)
1945 {
1946     VirtioNetRscSeg *seg, *rn;
1947     VirtioNetRscChain *chain = (VirtioNetRscChain *)opq;
1948 
1949     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) {
1950         if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
1951             chain->stat.purge_failed++;
1952             continue;
1953         }
1954     }
1955 
1956     chain->stat.timer++;
1957     if (!QTAILQ_EMPTY(&chain->buffers)) {
1958         timer_mod(chain->drain_timer,
1959               qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
1960     }
1961 }
1962 
1963 static void virtio_net_rsc_cleanup(VirtIONet *n)
1964 {
1965     VirtioNetRscChain *chain, *rn_chain;
1966     VirtioNetRscSeg *seg, *rn_seg;
1967 
1968     QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) {
1969         QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) {
1970             QTAILQ_REMOVE(&chain->buffers, seg, next);
1971             g_free(seg->buf);
1972             g_free(seg);
1973         }
1974 
1975         timer_free(chain->drain_timer);
1976         QTAILQ_REMOVE(&n->rsc_chains, chain, next);
1977         g_free(chain);
1978     }
1979 }
1980 
1981 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain,
1982                                      NetClientState *nc,
1983                                      const uint8_t *buf, size_t size)
1984 {
1985     uint16_t hdr_len;
1986     VirtioNetRscSeg *seg;
1987 
1988     hdr_len = chain->n->guest_hdr_len;
1989     seg = g_malloc(sizeof(VirtioNetRscSeg));
1990     seg->buf = g_malloc(hdr_len + sizeof(struct eth_header)
1991         + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD);
1992     memcpy(seg->buf, buf, size);
1993     seg->size = size;
1994     seg->packets = 1;
1995     seg->dup_ack = 0;
1996     seg->is_coalesced = 0;
1997     seg->nc = nc;
1998 
1999     QTAILQ_INSERT_TAIL(&chain->buffers, seg, next);
2000     chain->stat.cache++;
2001 
2002     switch (chain->proto) {
2003     case ETH_P_IP:
2004         virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
2005         break;
2006     case ETH_P_IPV6:
2007         virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit);
2008         break;
2009     default:
2010         g_assert_not_reached();
2011     }
2012 }
2013 
2014 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain,
2015                                          VirtioNetRscSeg *seg,
2016                                          const uint8_t *buf,
2017                                          struct tcp_header *n_tcp,
2018                                          struct tcp_header *o_tcp)
2019 {
2020     uint32_t nack, oack;
2021     uint16_t nwin, owin;
2022 
2023     nack = htonl(n_tcp->th_ack);
2024     nwin = htons(n_tcp->th_win);
2025     oack = htonl(o_tcp->th_ack);
2026     owin = htons(o_tcp->th_win);
2027 
2028     if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) {
2029         chain->stat.ack_out_of_win++;
2030         return RSC_FINAL;
2031     } else if (nack == oack) {
2032         /* duplicated ack or window probe */
2033         if (nwin == owin) {
2034             /* duplicated ack, add dup ack count due to whql test up to 1 */
2035             chain->stat.dup_ack++;
2036             return RSC_FINAL;
2037         } else {
2038             /* Coalesce window update */
2039             o_tcp->th_win = n_tcp->th_win;
2040             chain->stat.win_update++;
2041             return RSC_COALESCE;
2042         }
2043     } else {
2044         /* pure ack, go to 'C', finalize*/
2045         chain->stat.pure_ack++;
2046         return RSC_FINAL;
2047     }
2048 }
2049 
2050 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain,
2051                                             VirtioNetRscSeg *seg,
2052                                             const uint8_t *buf,
2053                                             VirtioNetRscUnit *n_unit)
2054 {
2055     void *data;
2056     uint16_t o_ip_len;
2057     uint32_t nseq, oseq;
2058     VirtioNetRscUnit *o_unit;
2059 
2060     o_unit = &seg->unit;
2061     o_ip_len = htons(*o_unit->ip_plen);
2062     nseq = htonl(n_unit->tcp->th_seq);
2063     oseq = htonl(o_unit->tcp->th_seq);
2064 
2065     /* out of order or retransmitted. */
2066     if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) {
2067         chain->stat.data_out_of_win++;
2068         return RSC_FINAL;
2069     }
2070 
2071     data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen;
2072     if (nseq == oseq) {
2073         if ((o_unit->payload == 0) && n_unit->payload) {
2074             /* From no payload to payload, normal case, not a dup ack or etc */
2075             chain->stat.data_after_pure_ack++;
2076             goto coalesce;
2077         } else {
2078             return virtio_net_rsc_handle_ack(chain, seg, buf,
2079                                              n_unit->tcp, o_unit->tcp);
2080         }
2081     } else if ((nseq - oseq) != o_unit->payload) {
2082         /* Not a consistent packet, out of order */
2083         chain->stat.data_out_of_order++;
2084         return RSC_FINAL;
2085     } else {
2086 coalesce:
2087         if ((o_ip_len + n_unit->payload) > chain->max_payload) {
2088             chain->stat.over_size++;
2089             return RSC_FINAL;
2090         }
2091 
2092         /* Here comes the right data, the payload length in v4/v6 is different,
2093            so use the field value to update and record the new data len */
2094         o_unit->payload += n_unit->payload; /* update new data len */
2095 
2096         /* update field in ip header */
2097         *o_unit->ip_plen = htons(o_ip_len + n_unit->payload);
2098 
2099         /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
2100            for windows guest, while this may change the behavior for linux
2101            guest (only if it uses RSC feature). */
2102         o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags;
2103 
2104         o_unit->tcp->th_ack = n_unit->tcp->th_ack;
2105         o_unit->tcp->th_win = n_unit->tcp->th_win;
2106 
2107         memmove(seg->buf + seg->size, data, n_unit->payload);
2108         seg->size += n_unit->payload;
2109         seg->packets++;
2110         chain->stat.coalesced++;
2111         return RSC_COALESCE;
2112     }
2113 }
2114 
2115 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain,
2116                                         VirtioNetRscSeg *seg,
2117                                         const uint8_t *buf, size_t size,
2118                                         VirtioNetRscUnit *unit)
2119 {
2120     struct ip_header *ip1, *ip2;
2121 
2122     ip1 = (struct ip_header *)(unit->ip);
2123     ip2 = (struct ip_header *)(seg->unit.ip);
2124     if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst)
2125         || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2126         || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2127         chain->stat.no_match++;
2128         return RSC_NO_MATCH;
2129     }
2130 
2131     return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2132 }
2133 
2134 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain,
2135                                         VirtioNetRscSeg *seg,
2136                                         const uint8_t *buf, size_t size,
2137                                         VirtioNetRscUnit *unit)
2138 {
2139     struct ip6_header *ip1, *ip2;
2140 
2141     ip1 = (struct ip6_header *)(unit->ip);
2142     ip2 = (struct ip6_header *)(seg->unit.ip);
2143     if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address))
2144         || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address))
2145         || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2146         || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2147             chain->stat.no_match++;
2148             return RSC_NO_MATCH;
2149     }
2150 
2151     return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2152 }
2153 
2154 /* Packets with 'SYN' should bypass, other flag should be sent after drain
2155  * to prevent out of order */
2156 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain,
2157                                          struct tcp_header *tcp)
2158 {
2159     uint16_t tcp_hdr;
2160     uint16_t tcp_flag;
2161 
2162     tcp_flag = htons(tcp->th_offset_flags);
2163     tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10;
2164     tcp_flag &= VIRTIO_NET_TCP_FLAG;
2165     if (tcp_flag & TH_SYN) {
2166         chain->stat.tcp_syn++;
2167         return RSC_BYPASS;
2168     }
2169 
2170     if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) {
2171         chain->stat.tcp_ctrl_drain++;
2172         return RSC_FINAL;
2173     }
2174 
2175     if (tcp_hdr > sizeof(struct tcp_header)) {
2176         chain->stat.tcp_all_opt++;
2177         return RSC_FINAL;
2178     }
2179 
2180     return RSC_CANDIDATE;
2181 }
2182 
2183 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain,
2184                                          NetClientState *nc,
2185                                          const uint8_t *buf, size_t size,
2186                                          VirtioNetRscUnit *unit)
2187 {
2188     int ret;
2189     VirtioNetRscSeg *seg, *nseg;
2190 
2191     if (QTAILQ_EMPTY(&chain->buffers)) {
2192         chain->stat.empty_cache++;
2193         virtio_net_rsc_cache_buf(chain, nc, buf, size);
2194         timer_mod(chain->drain_timer,
2195               qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
2196         return size;
2197     }
2198 
2199     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2200         if (chain->proto == ETH_P_IP) {
2201             ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
2202         } else {
2203             ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit);
2204         }
2205 
2206         if (ret == RSC_FINAL) {
2207             if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2208                 /* Send failed */
2209                 chain->stat.final_failed++;
2210                 return 0;
2211             }
2212 
2213             /* Send current packet */
2214             return virtio_net_do_receive(nc, buf, size);
2215         } else if (ret == RSC_NO_MATCH) {
2216             continue;
2217         } else {
2218             /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
2219             seg->is_coalesced = 1;
2220             return size;
2221         }
2222     }
2223 
2224     chain->stat.no_match_cache++;
2225     virtio_net_rsc_cache_buf(chain, nc, buf, size);
2226     return size;
2227 }
2228 
2229 /* Drain a connection data, this is to avoid out of order segments */
2230 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain,
2231                                         NetClientState *nc,
2232                                         const uint8_t *buf, size_t size,
2233                                         uint16_t ip_start, uint16_t ip_size,
2234                                         uint16_t tcp_port)
2235 {
2236     VirtioNetRscSeg *seg, *nseg;
2237     uint32_t ppair1, ppair2;
2238 
2239     ppair1 = *(uint32_t *)(buf + tcp_port);
2240     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2241         ppair2 = *(uint32_t *)(seg->buf + tcp_port);
2242         if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size)
2243             || (ppair1 != ppair2)) {
2244             continue;
2245         }
2246         if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2247             chain->stat.drain_failed++;
2248         }
2249 
2250         break;
2251     }
2252 
2253     return virtio_net_do_receive(nc, buf, size);
2254 }
2255 
2256 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain,
2257                                             struct ip_header *ip,
2258                                             const uint8_t *buf, size_t size)
2259 {
2260     uint16_t ip_len;
2261 
2262     /* Not an ipv4 packet */
2263     if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) {
2264         chain->stat.ip_option++;
2265         return RSC_BYPASS;
2266     }
2267 
2268     /* Don't handle packets with ip option */
2269     if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) {
2270         chain->stat.ip_option++;
2271         return RSC_BYPASS;
2272     }
2273 
2274     if (ip->ip_p != IPPROTO_TCP) {
2275         chain->stat.bypass_not_tcp++;
2276         return RSC_BYPASS;
2277     }
2278 
2279     /* Don't handle packets with ip fragment */
2280     if (!(htons(ip->ip_off) & IP_DF)) {
2281         chain->stat.ip_frag++;
2282         return RSC_BYPASS;
2283     }
2284 
2285     /* Don't handle packets with ecn flag */
2286     if (IPTOS_ECN(ip->ip_tos)) {
2287         chain->stat.ip_ecn++;
2288         return RSC_BYPASS;
2289     }
2290 
2291     ip_len = htons(ip->ip_len);
2292     if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header))
2293         || ip_len > (size - chain->n->guest_hdr_len -
2294                      sizeof(struct eth_header))) {
2295         chain->stat.ip_hacked++;
2296         return RSC_BYPASS;
2297     }
2298 
2299     return RSC_CANDIDATE;
2300 }
2301 
2302 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain,
2303                                       NetClientState *nc,
2304                                       const uint8_t *buf, size_t size)
2305 {
2306     int32_t ret;
2307     uint16_t hdr_len;
2308     VirtioNetRscUnit unit;
2309 
2310     hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2311 
2312     if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)
2313         + sizeof(struct tcp_header))) {
2314         chain->stat.bypass_not_tcp++;
2315         return virtio_net_do_receive(nc, buf, size);
2316     }
2317 
2318     virtio_net_rsc_extract_unit4(chain, buf, &unit);
2319     if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size)
2320         != RSC_CANDIDATE) {
2321         return virtio_net_do_receive(nc, buf, size);
2322     }
2323 
2324     ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2325     if (ret == RSC_BYPASS) {
2326         return virtio_net_do_receive(nc, buf, size);
2327     } else if (ret == RSC_FINAL) {
2328         return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2329                 ((hdr_len + sizeof(struct eth_header)) + 12),
2330                 VIRTIO_NET_IP4_ADDR_SIZE,
2331                 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header));
2332     }
2333 
2334     return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2335 }
2336 
2337 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain,
2338                                             struct ip6_header *ip6,
2339                                             const uint8_t *buf, size_t size)
2340 {
2341     uint16_t ip_len;
2342 
2343     if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4)
2344         != IP_HEADER_VERSION_6) {
2345         return RSC_BYPASS;
2346     }
2347 
2348     /* Both option and protocol is checked in this */
2349     if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) {
2350         chain->stat.bypass_not_tcp++;
2351         return RSC_BYPASS;
2352     }
2353 
2354     ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
2355     if (ip_len < sizeof(struct tcp_header) ||
2356         ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header)
2357                   - sizeof(struct ip6_header))) {
2358         chain->stat.ip_hacked++;
2359         return RSC_BYPASS;
2360     }
2361 
2362     /* Don't handle packets with ecn flag */
2363     if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) {
2364         chain->stat.ip_ecn++;
2365         return RSC_BYPASS;
2366     }
2367 
2368     return RSC_CANDIDATE;
2369 }
2370 
2371 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc,
2372                                       const uint8_t *buf, size_t size)
2373 {
2374     int32_t ret;
2375     uint16_t hdr_len;
2376     VirtioNetRscChain *chain;
2377     VirtioNetRscUnit unit;
2378 
2379     chain = (VirtioNetRscChain *)opq;
2380     hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2381 
2382     if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header)
2383         + sizeof(tcp_header))) {
2384         return virtio_net_do_receive(nc, buf, size);
2385     }
2386 
2387     virtio_net_rsc_extract_unit6(chain, buf, &unit);
2388     if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain,
2389                                                  unit.ip, buf, size)) {
2390         return virtio_net_do_receive(nc, buf, size);
2391     }
2392 
2393     ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2394     if (ret == RSC_BYPASS) {
2395         return virtio_net_do_receive(nc, buf, size);
2396     } else if (ret == RSC_FINAL) {
2397         return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2398                 ((hdr_len + sizeof(struct eth_header)) + 8),
2399                 VIRTIO_NET_IP6_ADDR_SIZE,
2400                 hdr_len + sizeof(struct eth_header)
2401                 + sizeof(struct ip6_header));
2402     }
2403 
2404     return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2405 }
2406 
2407 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n,
2408                                                       NetClientState *nc,
2409                                                       uint16_t proto)
2410 {
2411     VirtioNetRscChain *chain;
2412 
2413     if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) {
2414         return NULL;
2415     }
2416 
2417     QTAILQ_FOREACH(chain, &n->rsc_chains, next) {
2418         if (chain->proto == proto) {
2419             return chain;
2420         }
2421     }
2422 
2423     chain = g_malloc(sizeof(*chain));
2424     chain->n = n;
2425     chain->proto = proto;
2426     if (proto == (uint16_t)ETH_P_IP) {
2427         chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
2428         chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2429     } else {
2430         chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD;
2431         chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2432     }
2433     chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST,
2434                                       virtio_net_rsc_purge, chain);
2435     memset(&chain->stat, 0, sizeof(chain->stat));
2436 
2437     QTAILQ_INIT(&chain->buffers);
2438     QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next);
2439 
2440     return chain;
2441 }
2442 
2443 static ssize_t virtio_net_rsc_receive(NetClientState *nc,
2444                                       const uint8_t *buf,
2445                                       size_t size)
2446 {
2447     uint16_t proto;
2448     VirtioNetRscChain *chain;
2449     struct eth_header *eth;
2450     VirtIONet *n;
2451 
2452     n = qemu_get_nic_opaque(nc);
2453     if (size < (n->host_hdr_len + sizeof(struct eth_header))) {
2454         return virtio_net_do_receive(nc, buf, size);
2455     }
2456 
2457     eth = (struct eth_header *)(buf + n->guest_hdr_len);
2458     proto = htons(eth->h_proto);
2459 
2460     chain = virtio_net_rsc_lookup_chain(n, nc, proto);
2461     if (chain) {
2462         chain->stat.received++;
2463         if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) {
2464             return virtio_net_rsc_receive4(chain, nc, buf, size);
2465         } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) {
2466             return virtio_net_rsc_receive6(chain, nc, buf, size);
2467         }
2468     }
2469     return virtio_net_do_receive(nc, buf, size);
2470 }
2471 
2472 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf,
2473                                   size_t size)
2474 {
2475     VirtIONet *n = qemu_get_nic_opaque(nc);
2476     if ((n->rsc4_enabled || n->rsc6_enabled)) {
2477         return virtio_net_rsc_receive(nc, buf, size);
2478     } else {
2479         return virtio_net_do_receive(nc, buf, size);
2480     }
2481 }
2482 
2483 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
2484 
2485 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
2486 {
2487     VirtIONet *n = qemu_get_nic_opaque(nc);
2488     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
2489     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2490 
2491     virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
2492     virtio_notify(vdev, q->tx_vq);
2493 
2494     g_free(q->async_tx.elem);
2495     q->async_tx.elem = NULL;
2496 
2497     virtio_queue_set_notification(q->tx_vq, 1);
2498     virtio_net_flush_tx(q);
2499 }
2500 
2501 /* TX */
2502 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
2503 {
2504     VirtIONet *n = q->n;
2505     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2506     VirtQueueElement *elem;
2507     int32_t num_packets = 0;
2508     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
2509     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2510         return num_packets;
2511     }
2512 
2513     if (q->async_tx.elem) {
2514         virtio_queue_set_notification(q->tx_vq, 0);
2515         return num_packets;
2516     }
2517 
2518     for (;;) {
2519         ssize_t ret;
2520         unsigned int out_num;
2521         struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
2522         struct virtio_net_hdr_mrg_rxbuf mhdr;
2523 
2524         elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
2525         if (!elem) {
2526             break;
2527         }
2528 
2529         out_num = elem->out_num;
2530         out_sg = elem->out_sg;
2531         if (out_num < 1) {
2532             virtio_error(vdev, "virtio-net header not in first element");
2533             virtqueue_detach_element(q->tx_vq, elem, 0);
2534             g_free(elem);
2535             return -EINVAL;
2536         }
2537 
2538         if (n->has_vnet_hdr) {
2539             if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
2540                 n->guest_hdr_len) {
2541                 virtio_error(vdev, "virtio-net header incorrect");
2542                 virtqueue_detach_element(q->tx_vq, elem, 0);
2543                 g_free(elem);
2544                 return -EINVAL;
2545             }
2546             if (n->needs_vnet_hdr_swap) {
2547                 virtio_net_hdr_swap(vdev, (void *) &mhdr);
2548                 sg2[0].iov_base = &mhdr;
2549                 sg2[0].iov_len = n->guest_hdr_len;
2550                 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
2551                                    out_sg, out_num,
2552                                    n->guest_hdr_len, -1);
2553                 if (out_num == VIRTQUEUE_MAX_SIZE) {
2554                     goto drop;
2555                 }
2556                 out_num += 1;
2557                 out_sg = sg2;
2558             }
2559         }
2560         /*
2561          * If host wants to see the guest header as is, we can
2562          * pass it on unchanged. Otherwise, copy just the parts
2563          * that host is interested in.
2564          */
2565         assert(n->host_hdr_len <= n->guest_hdr_len);
2566         if (n->host_hdr_len != n->guest_hdr_len) {
2567             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
2568                                        out_sg, out_num,
2569                                        0, n->host_hdr_len);
2570             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
2571                              out_sg, out_num,
2572                              n->guest_hdr_len, -1);
2573             out_num = sg_num;
2574             out_sg = sg;
2575         }
2576 
2577         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
2578                                       out_sg, out_num, virtio_net_tx_complete);
2579         if (ret == 0) {
2580             virtio_queue_set_notification(q->tx_vq, 0);
2581             q->async_tx.elem = elem;
2582             return -EBUSY;
2583         }
2584 
2585 drop:
2586         virtqueue_push(q->tx_vq, elem, 0);
2587         virtio_notify(vdev, q->tx_vq);
2588         g_free(elem);
2589 
2590         if (++num_packets >= n->tx_burst) {
2591             break;
2592         }
2593     }
2594     return num_packets;
2595 }
2596 
2597 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
2598 {
2599     VirtIONet *n = VIRTIO_NET(vdev);
2600     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2601 
2602     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2603         virtio_net_drop_tx_queue_data(vdev, vq);
2604         return;
2605     }
2606 
2607     /* This happens when device was stopped but VCPU wasn't. */
2608     if (!vdev->vm_running) {
2609         q->tx_waiting = 1;
2610         return;
2611     }
2612 
2613     if (q->tx_waiting) {
2614         virtio_queue_set_notification(vq, 1);
2615         timer_del(q->tx_timer);
2616         q->tx_waiting = 0;
2617         if (virtio_net_flush_tx(q) == -EINVAL) {
2618             return;
2619         }
2620     } else {
2621         timer_mod(q->tx_timer,
2622                        qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2623         q->tx_waiting = 1;
2624         virtio_queue_set_notification(vq, 0);
2625     }
2626 }
2627 
2628 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
2629 {
2630     VirtIONet *n = VIRTIO_NET(vdev);
2631     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2632 
2633     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2634         virtio_net_drop_tx_queue_data(vdev, vq);
2635         return;
2636     }
2637 
2638     if (unlikely(q->tx_waiting)) {
2639         return;
2640     }
2641     q->tx_waiting = 1;
2642     /* This happens when device was stopped but VCPU wasn't. */
2643     if (!vdev->vm_running) {
2644         return;
2645     }
2646     virtio_queue_set_notification(vq, 0);
2647     qemu_bh_schedule(q->tx_bh);
2648 }
2649 
2650 static void virtio_net_tx_timer(void *opaque)
2651 {
2652     VirtIONetQueue *q = opaque;
2653     VirtIONet *n = q->n;
2654     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2655     /* This happens when device was stopped but BH wasn't. */
2656     if (!vdev->vm_running) {
2657         /* Make sure tx waiting is set, so we'll run when restarted. */
2658         assert(q->tx_waiting);
2659         return;
2660     }
2661 
2662     q->tx_waiting = 0;
2663 
2664     /* Just in case the driver is not ready on more */
2665     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2666         return;
2667     }
2668 
2669     virtio_queue_set_notification(q->tx_vq, 1);
2670     virtio_net_flush_tx(q);
2671 }
2672 
2673 static void virtio_net_tx_bh(void *opaque)
2674 {
2675     VirtIONetQueue *q = opaque;
2676     VirtIONet *n = q->n;
2677     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2678     int32_t ret;
2679 
2680     /* This happens when device was stopped but BH wasn't. */
2681     if (!vdev->vm_running) {
2682         /* Make sure tx waiting is set, so we'll run when restarted. */
2683         assert(q->tx_waiting);
2684         return;
2685     }
2686 
2687     q->tx_waiting = 0;
2688 
2689     /* Just in case the driver is not ready on more */
2690     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
2691         return;
2692     }
2693 
2694     ret = virtio_net_flush_tx(q);
2695     if (ret == -EBUSY || ret == -EINVAL) {
2696         return; /* Notification re-enable handled by tx_complete or device
2697                  * broken */
2698     }
2699 
2700     /* If we flush a full burst of packets, assume there are
2701      * more coming and immediately reschedule */
2702     if (ret >= n->tx_burst) {
2703         qemu_bh_schedule(q->tx_bh);
2704         q->tx_waiting = 1;
2705         return;
2706     }
2707 
2708     /* If less than a full burst, re-enable notification and flush
2709      * anything that may have come in while we weren't looking.  If
2710      * we find something, assume the guest is still active and reschedule */
2711     virtio_queue_set_notification(q->tx_vq, 1);
2712     ret = virtio_net_flush_tx(q);
2713     if (ret == -EINVAL) {
2714         return;
2715     } else if (ret > 0) {
2716         virtio_queue_set_notification(q->tx_vq, 0);
2717         qemu_bh_schedule(q->tx_bh);
2718         q->tx_waiting = 1;
2719     }
2720 }
2721 
2722 static void virtio_net_add_queue(VirtIONet *n, int index)
2723 {
2724     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2725 
2726     n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
2727                                            virtio_net_handle_rx);
2728 
2729     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
2730         n->vqs[index].tx_vq =
2731             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2732                              virtio_net_handle_tx_timer);
2733         n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
2734                                               virtio_net_tx_timer,
2735                                               &n->vqs[index]);
2736     } else {
2737         n->vqs[index].tx_vq =
2738             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2739                              virtio_net_handle_tx_bh);
2740         n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
2741     }
2742 
2743     n->vqs[index].tx_waiting = 0;
2744     n->vqs[index].n = n;
2745 }
2746 
2747 static void virtio_net_del_queue(VirtIONet *n, int index)
2748 {
2749     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2750     VirtIONetQueue *q = &n->vqs[index];
2751     NetClientState *nc = qemu_get_subqueue(n->nic, index);
2752 
2753     qemu_purge_queued_packets(nc);
2754 
2755     virtio_del_queue(vdev, index * 2);
2756     if (q->tx_timer) {
2757         timer_free(q->tx_timer);
2758         q->tx_timer = NULL;
2759     } else {
2760         qemu_bh_delete(q->tx_bh);
2761         q->tx_bh = NULL;
2762     }
2763     q->tx_waiting = 0;
2764     virtio_del_queue(vdev, index * 2 + 1);
2765 }
2766 
2767 static void virtio_net_change_num_queue_pairs(VirtIONet *n, int new_max_queue_pairs)
2768 {
2769     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2770     int old_num_queues = virtio_get_num_queues(vdev);
2771     int new_num_queues = new_max_queue_pairs * 2 + 1;
2772     int i;
2773 
2774     assert(old_num_queues >= 3);
2775     assert(old_num_queues % 2 == 1);
2776 
2777     if (old_num_queues == new_num_queues) {
2778         return;
2779     }
2780 
2781     /*
2782      * We always need to remove and add ctrl vq if
2783      * old_num_queues != new_num_queues. Remove ctrl_vq first,
2784      * and then we only enter one of the following two loops.
2785      */
2786     virtio_del_queue(vdev, old_num_queues - 1);
2787 
2788     for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
2789         /* new_num_queues < old_num_queues */
2790         virtio_net_del_queue(n, i / 2);
2791     }
2792 
2793     for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
2794         /* new_num_queues > old_num_queues */
2795         virtio_net_add_queue(n, i / 2);
2796     }
2797 
2798     /* add ctrl_vq last */
2799     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
2800 }
2801 
2802 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
2803 {
2804     int max = multiqueue ? n->max_queue_pairs : 1;
2805 
2806     n->multiqueue = multiqueue;
2807     virtio_net_change_num_queue_pairs(n, max);
2808 
2809     virtio_net_set_queue_pairs(n);
2810 }
2811 
2812 static int virtio_net_post_load_device(void *opaque, int version_id)
2813 {
2814     VirtIONet *n = opaque;
2815     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2816     int i, link_down;
2817 
2818     trace_virtio_net_post_load_device();
2819     virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
2820                                virtio_vdev_has_feature(vdev,
2821                                                        VIRTIO_F_VERSION_1),
2822                                virtio_vdev_has_feature(vdev,
2823                                                        VIRTIO_NET_F_HASH_REPORT));
2824 
2825     /* MAC_TABLE_ENTRIES may be different from the saved image */
2826     if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
2827         n->mac_table.in_use = 0;
2828     }
2829 
2830     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
2831         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
2832     }
2833 
2834     /*
2835      * curr_guest_offloads will be later overwritten by the
2836      * virtio_set_features_nocheck call done from the virtio_load.
2837      * Here we make sure it is preserved and restored accordingly
2838      * in the virtio_net_post_load_virtio callback.
2839      */
2840     n->saved_guest_offloads = n->curr_guest_offloads;
2841 
2842     virtio_net_set_queue_pairs(n);
2843 
2844     /* Find the first multicast entry in the saved MAC filter */
2845     for (i = 0; i < n->mac_table.in_use; i++) {
2846         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
2847             break;
2848         }
2849     }
2850     n->mac_table.first_multi = i;
2851 
2852     /* nc.link_down can't be migrated, so infer link_down according
2853      * to link status bit in n->status */
2854     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
2855     for (i = 0; i < n->max_queue_pairs; i++) {
2856         qemu_get_subqueue(n->nic, i)->link_down = link_down;
2857     }
2858 
2859     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
2860         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
2861         qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
2862                                   QEMU_CLOCK_VIRTUAL,
2863                                   virtio_net_announce_timer, n);
2864         if (n->announce_timer.round) {
2865             timer_mod(n->announce_timer.tm,
2866                       qemu_clock_get_ms(n->announce_timer.type));
2867         } else {
2868             qemu_announce_timer_del(&n->announce_timer, false);
2869         }
2870     }
2871 
2872     if (n->rss_data.enabled) {
2873         n->rss_data.enabled_software_rss = n->rss_data.populate_hash;
2874         if (!n->rss_data.populate_hash) {
2875             if (!virtio_net_attach_epbf_rss(n)) {
2876                 if (get_vhost_net(qemu_get_queue(n->nic)->peer)) {
2877                     warn_report("Can't post-load eBPF RSS for vhost");
2878                 } else {
2879                     warn_report("Can't post-load eBPF RSS - "
2880                                 "fallback to software RSS");
2881                     n->rss_data.enabled_software_rss = true;
2882                 }
2883             }
2884         }
2885 
2886         trace_virtio_net_rss_enable(n->rss_data.hash_types,
2887                                     n->rss_data.indirections_len,
2888                                     sizeof(n->rss_data.key));
2889     } else {
2890         trace_virtio_net_rss_disable();
2891     }
2892     return 0;
2893 }
2894 
2895 static int virtio_net_post_load_virtio(VirtIODevice *vdev)
2896 {
2897     VirtIONet *n = VIRTIO_NET(vdev);
2898     /*
2899      * The actual needed state is now in saved_guest_offloads,
2900      * see virtio_net_post_load_device for detail.
2901      * Restore it back and apply the desired offloads.
2902      */
2903     n->curr_guest_offloads = n->saved_guest_offloads;
2904     if (peer_has_vnet_hdr(n)) {
2905         virtio_net_apply_guest_offloads(n);
2906     }
2907 
2908     return 0;
2909 }
2910 
2911 /* tx_waiting field of a VirtIONetQueue */
2912 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = {
2913     .name = "virtio-net-queue-tx_waiting",
2914     .fields = (VMStateField[]) {
2915         VMSTATE_UINT32(tx_waiting, VirtIONetQueue),
2916         VMSTATE_END_OF_LIST()
2917    },
2918 };
2919 
2920 static bool max_queue_pairs_gt_1(void *opaque, int version_id)
2921 {
2922     return VIRTIO_NET(opaque)->max_queue_pairs > 1;
2923 }
2924 
2925 static bool has_ctrl_guest_offloads(void *opaque, int version_id)
2926 {
2927     return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque),
2928                                    VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
2929 }
2930 
2931 static bool mac_table_fits(void *opaque, int version_id)
2932 {
2933     return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES;
2934 }
2935 
2936 static bool mac_table_doesnt_fit(void *opaque, int version_id)
2937 {
2938     return !mac_table_fits(opaque, version_id);
2939 }
2940 
2941 /* This temporary type is shared by all the WITH_TMP methods
2942  * although only some fields are used by each.
2943  */
2944 struct VirtIONetMigTmp {
2945     VirtIONet      *parent;
2946     VirtIONetQueue *vqs_1;
2947     uint16_t        curr_queue_pairs_1;
2948     uint8_t         has_ufo;
2949     uint32_t        has_vnet_hdr;
2950 };
2951 
2952 /* The 2nd and subsequent tx_waiting flags are loaded later than
2953  * the 1st entry in the queue_pairs and only if there's more than one
2954  * entry.  We use the tmp mechanism to calculate a temporary
2955  * pointer and count and also validate the count.
2956  */
2957 
2958 static int virtio_net_tx_waiting_pre_save(void *opaque)
2959 {
2960     struct VirtIONetMigTmp *tmp = opaque;
2961 
2962     tmp->vqs_1 = tmp->parent->vqs + 1;
2963     tmp->curr_queue_pairs_1 = tmp->parent->curr_queue_pairs - 1;
2964     if (tmp->parent->curr_queue_pairs == 0) {
2965         tmp->curr_queue_pairs_1 = 0;
2966     }
2967 
2968     return 0;
2969 }
2970 
2971 static int virtio_net_tx_waiting_pre_load(void *opaque)
2972 {
2973     struct VirtIONetMigTmp *tmp = opaque;
2974 
2975     /* Reuse the pointer setup from save */
2976     virtio_net_tx_waiting_pre_save(opaque);
2977 
2978     if (tmp->parent->curr_queue_pairs > tmp->parent->max_queue_pairs) {
2979         error_report("virtio-net: curr_queue_pairs %x > max_queue_pairs %x",
2980             tmp->parent->curr_queue_pairs, tmp->parent->max_queue_pairs);
2981 
2982         return -EINVAL;
2983     }
2984 
2985     return 0; /* all good */
2986 }
2987 
2988 static const VMStateDescription vmstate_virtio_net_tx_waiting = {
2989     .name      = "virtio-net-tx_waiting",
2990     .pre_load  = virtio_net_tx_waiting_pre_load,
2991     .pre_save  = virtio_net_tx_waiting_pre_save,
2992     .fields    = (VMStateField[]) {
2993         VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp,
2994                                      curr_queue_pairs_1,
2995                                      vmstate_virtio_net_queue_tx_waiting,
2996                                      struct VirtIONetQueue),
2997         VMSTATE_END_OF_LIST()
2998     },
2999 };
3000 
3001 /* the 'has_ufo' flag is just tested; if the incoming stream has the
3002  * flag set we need to check that we have it
3003  */
3004 static int virtio_net_ufo_post_load(void *opaque, int version_id)
3005 {
3006     struct VirtIONetMigTmp *tmp = opaque;
3007 
3008     if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) {
3009         error_report("virtio-net: saved image requires TUN_F_UFO support");
3010         return -EINVAL;
3011     }
3012 
3013     return 0;
3014 }
3015 
3016 static int virtio_net_ufo_pre_save(void *opaque)
3017 {
3018     struct VirtIONetMigTmp *tmp = opaque;
3019 
3020     tmp->has_ufo = tmp->parent->has_ufo;
3021 
3022     return 0;
3023 }
3024 
3025 static const VMStateDescription vmstate_virtio_net_has_ufo = {
3026     .name      = "virtio-net-ufo",
3027     .post_load = virtio_net_ufo_post_load,
3028     .pre_save  = virtio_net_ufo_pre_save,
3029     .fields    = (VMStateField[]) {
3030         VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp),
3031         VMSTATE_END_OF_LIST()
3032     },
3033 };
3034 
3035 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
3036  * flag set we need to check that we have it
3037  */
3038 static int virtio_net_vnet_post_load(void *opaque, int version_id)
3039 {
3040     struct VirtIONetMigTmp *tmp = opaque;
3041 
3042     if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) {
3043         error_report("virtio-net: saved image requires vnet_hdr=on");
3044         return -EINVAL;
3045     }
3046 
3047     return 0;
3048 }
3049 
3050 static int virtio_net_vnet_pre_save(void *opaque)
3051 {
3052     struct VirtIONetMigTmp *tmp = opaque;
3053 
3054     tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr;
3055 
3056     return 0;
3057 }
3058 
3059 static const VMStateDescription vmstate_virtio_net_has_vnet = {
3060     .name      = "virtio-net-vnet",
3061     .post_load = virtio_net_vnet_post_load,
3062     .pre_save  = virtio_net_vnet_pre_save,
3063     .fields    = (VMStateField[]) {
3064         VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp),
3065         VMSTATE_END_OF_LIST()
3066     },
3067 };
3068 
3069 static bool virtio_net_rss_needed(void *opaque)
3070 {
3071     return VIRTIO_NET(opaque)->rss_data.enabled;
3072 }
3073 
3074 static const VMStateDescription vmstate_virtio_net_rss = {
3075     .name      = "virtio-net-device/rss",
3076     .version_id = 1,
3077     .minimum_version_id = 1,
3078     .needed = virtio_net_rss_needed,
3079     .fields = (VMStateField[]) {
3080         VMSTATE_BOOL(rss_data.enabled, VirtIONet),
3081         VMSTATE_BOOL(rss_data.redirect, VirtIONet),
3082         VMSTATE_BOOL(rss_data.populate_hash, VirtIONet),
3083         VMSTATE_UINT32(rss_data.hash_types, VirtIONet),
3084         VMSTATE_UINT16(rss_data.indirections_len, VirtIONet),
3085         VMSTATE_UINT16(rss_data.default_queue, VirtIONet),
3086         VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet,
3087                             VIRTIO_NET_RSS_MAX_KEY_SIZE),
3088         VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet,
3089                                     rss_data.indirections_len, 0,
3090                                     vmstate_info_uint16, uint16_t),
3091         VMSTATE_END_OF_LIST()
3092     },
3093 };
3094 
3095 static const VMStateDescription vmstate_virtio_net_device = {
3096     .name = "virtio-net-device",
3097     .version_id = VIRTIO_NET_VM_VERSION,
3098     .minimum_version_id = VIRTIO_NET_VM_VERSION,
3099     .post_load = virtio_net_post_load_device,
3100     .fields = (VMStateField[]) {
3101         VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN),
3102         VMSTATE_STRUCT_POINTER(vqs, VirtIONet,
3103                                vmstate_virtio_net_queue_tx_waiting,
3104                                VirtIONetQueue),
3105         VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet),
3106         VMSTATE_UINT16(status, VirtIONet),
3107         VMSTATE_UINT8(promisc, VirtIONet),
3108         VMSTATE_UINT8(allmulti, VirtIONet),
3109         VMSTATE_UINT32(mac_table.in_use, VirtIONet),
3110 
3111         /* Guarded pair: If it fits we load it, else we throw it away
3112          * - can happen if source has a larger MAC table.; post-load
3113          *  sets flags in this case.
3114          */
3115         VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet,
3116                                 0, mac_table_fits, mac_table.in_use,
3117                                  ETH_ALEN),
3118         VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0,
3119                                      mac_table.in_use, ETH_ALEN),
3120 
3121         /* Note: This is an array of uint32's that's always been saved as a
3122          * buffer; hold onto your endiannesses; it's actually used as a bitmap
3123          * but based on the uint.
3124          */
3125         VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3),
3126         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3127                          vmstate_virtio_net_has_vnet),
3128         VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet),
3129         VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet),
3130         VMSTATE_UINT8(alluni, VirtIONet),
3131         VMSTATE_UINT8(nomulti, VirtIONet),
3132         VMSTATE_UINT8(nouni, VirtIONet),
3133         VMSTATE_UINT8(nobcast, VirtIONet),
3134         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3135                          vmstate_virtio_net_has_ufo),
3136         VMSTATE_SINGLE_TEST(max_queue_pairs, VirtIONet, max_queue_pairs_gt_1, 0,
3137                             vmstate_info_uint16_equal, uint16_t),
3138         VMSTATE_UINT16_TEST(curr_queue_pairs, VirtIONet, max_queue_pairs_gt_1),
3139         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3140                          vmstate_virtio_net_tx_waiting),
3141         VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet,
3142                             has_ctrl_guest_offloads),
3143         VMSTATE_END_OF_LIST()
3144    },
3145     .subsections = (const VMStateDescription * []) {
3146         &vmstate_virtio_net_rss,
3147         NULL
3148     }
3149 };
3150 
3151 static NetClientInfo net_virtio_info = {
3152     .type = NET_CLIENT_DRIVER_NIC,
3153     .size = sizeof(NICState),
3154     .can_receive = virtio_net_can_receive,
3155     .receive = virtio_net_receive,
3156     .link_status_changed = virtio_net_set_link_status,
3157     .query_rx_filter = virtio_net_query_rxfilter,
3158     .announce = virtio_net_announce,
3159 };
3160 
3161 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
3162 {
3163     VirtIONet *n = VIRTIO_NET(vdev);
3164     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
3165     assert(n->vhost_started);
3166     return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
3167 }
3168 
3169 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
3170                                            bool mask)
3171 {
3172     VirtIONet *n = VIRTIO_NET(vdev);
3173     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
3174     assert(n->vhost_started);
3175     vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
3176                              vdev, idx, mask);
3177 }
3178 
3179 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
3180 {
3181     virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
3182 
3183     n->config_size = virtio_feature_get_config_size(feature_sizes,
3184                                                     host_features);
3185 }
3186 
3187 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
3188                                    const char *type)
3189 {
3190     /*
3191      * The name can be NULL, the netclient name will be type.x.
3192      */
3193     assert(type != NULL);
3194 
3195     g_free(n->netclient_name);
3196     g_free(n->netclient_type);
3197     n->netclient_name = g_strdup(name);
3198     n->netclient_type = g_strdup(type);
3199 }
3200 
3201 static bool failover_unplug_primary(VirtIONet *n, DeviceState *dev)
3202 {
3203     HotplugHandler *hotplug_ctrl;
3204     PCIDevice *pci_dev;
3205     Error *err = NULL;
3206 
3207     hotplug_ctrl = qdev_get_hotplug_handler(dev);
3208     if (hotplug_ctrl) {
3209         pci_dev = PCI_DEVICE(dev);
3210         pci_dev->partially_hotplugged = true;
3211         hotplug_handler_unplug_request(hotplug_ctrl, dev, &err);
3212         if (err) {
3213             error_report_err(err);
3214             return false;
3215         }
3216     } else {
3217         return false;
3218     }
3219     return true;
3220 }
3221 
3222 static bool failover_replug_primary(VirtIONet *n, DeviceState *dev,
3223                                     Error **errp)
3224 {
3225     Error *err = NULL;
3226     HotplugHandler *hotplug_ctrl;
3227     PCIDevice *pdev = PCI_DEVICE(dev);
3228     BusState *primary_bus;
3229 
3230     if (!pdev->partially_hotplugged) {
3231         return true;
3232     }
3233     primary_bus = dev->parent_bus;
3234     if (!primary_bus) {
3235         error_setg(errp, "virtio_net: couldn't find primary bus");
3236         return false;
3237     }
3238     qdev_set_parent_bus(dev, primary_bus, &error_abort);
3239     qatomic_set(&n->failover_primary_hidden, false);
3240     hotplug_ctrl = qdev_get_hotplug_handler(dev);
3241     if (hotplug_ctrl) {
3242         hotplug_handler_pre_plug(hotplug_ctrl, dev, &err);
3243         if (err) {
3244             goto out;
3245         }
3246         hotplug_handler_plug(hotplug_ctrl, dev, &err);
3247     }
3248     pdev->partially_hotplugged = false;
3249 
3250 out:
3251     error_propagate(errp, err);
3252     return !err;
3253 }
3254 
3255 static void virtio_net_handle_migration_primary(VirtIONet *n, MigrationState *s)
3256 {
3257     bool should_be_hidden;
3258     Error *err = NULL;
3259     DeviceState *dev = failover_find_primary_device(n);
3260 
3261     if (!dev) {
3262         return;
3263     }
3264 
3265     should_be_hidden = qatomic_read(&n->failover_primary_hidden);
3266 
3267     if (migration_in_setup(s) && !should_be_hidden) {
3268         if (failover_unplug_primary(n, dev)) {
3269             vmstate_unregister(VMSTATE_IF(dev), qdev_get_vmsd(dev), dev);
3270             qapi_event_send_unplug_primary(dev->id);
3271             qatomic_set(&n->failover_primary_hidden, true);
3272         } else {
3273             warn_report("couldn't unplug primary device");
3274         }
3275     } else if (migration_has_failed(s)) {
3276         /* We already unplugged the device let's plug it back */
3277         if (!failover_replug_primary(n, dev, &err)) {
3278             if (err) {
3279                 error_report_err(err);
3280             }
3281         }
3282     }
3283 }
3284 
3285 static void virtio_net_migration_state_notifier(Notifier *notifier, void *data)
3286 {
3287     MigrationState *s = data;
3288     VirtIONet *n = container_of(notifier, VirtIONet, migration_state);
3289     virtio_net_handle_migration_primary(n, s);
3290 }
3291 
3292 static bool failover_hide_primary_device(DeviceListener *listener,
3293                                          const QDict *device_opts,
3294                                          bool from_json,
3295                                          Error **errp)
3296 {
3297     VirtIONet *n = container_of(listener, VirtIONet, primary_listener);
3298     const char *standby_id;
3299 
3300     if (!device_opts) {
3301         return false;
3302     }
3303 
3304     if (!qdict_haskey(device_opts, "failover_pair_id")) {
3305         return false;
3306     }
3307 
3308     if (!qdict_haskey(device_opts, "id")) {
3309         error_setg(errp, "Device with failover_pair_id needs to have id");
3310         return false;
3311     }
3312 
3313     standby_id = qdict_get_str(device_opts, "failover_pair_id");
3314     if (g_strcmp0(standby_id, n->netclient_name) != 0) {
3315         return false;
3316     }
3317 
3318     /*
3319      * The hide helper can be called several times for a given device.
3320      * Check there is only one primary for a virtio-net device but
3321      * don't duplicate the qdict several times if it's called for the same
3322      * device.
3323      */
3324     if (n->primary_opts) {
3325         const char *old, *new;
3326         /* devices with failover_pair_id always have an id */
3327         old = qdict_get_str(n->primary_opts, "id");
3328         new = qdict_get_str(device_opts, "id");
3329         if (strcmp(old, new) != 0) {
3330             error_setg(errp, "Cannot attach more than one primary device to "
3331                        "'%s': '%s' and '%s'", n->netclient_name, old, new);
3332             return false;
3333         }
3334     } else {
3335         n->primary_opts = qdict_clone_shallow(device_opts);
3336         n->primary_opts_from_json = from_json;
3337     }
3338 
3339     /* failover_primary_hidden is set during feature negotiation */
3340     return qatomic_read(&n->failover_primary_hidden);
3341 }
3342 
3343 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
3344 {
3345     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3346     VirtIONet *n = VIRTIO_NET(dev);
3347     NetClientState *nc;
3348     int i;
3349 
3350     if (n->net_conf.mtu) {
3351         n->host_features |= (1ULL << VIRTIO_NET_F_MTU);
3352     }
3353 
3354     if (n->net_conf.duplex_str) {
3355         if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) {
3356             n->net_conf.duplex = DUPLEX_HALF;
3357         } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) {
3358             n->net_conf.duplex = DUPLEX_FULL;
3359         } else {
3360             error_setg(errp, "'duplex' must be 'half' or 'full'");
3361             return;
3362         }
3363         n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3364     } else {
3365         n->net_conf.duplex = DUPLEX_UNKNOWN;
3366     }
3367 
3368     if (n->net_conf.speed < SPEED_UNKNOWN) {
3369         error_setg(errp, "'speed' must be between 0 and INT_MAX");
3370         return;
3371     }
3372     if (n->net_conf.speed >= 0) {
3373         n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3374     }
3375 
3376     if (n->failover) {
3377         n->primary_listener.hide_device = failover_hide_primary_device;
3378         qatomic_set(&n->failover_primary_hidden, true);
3379         device_listener_register(&n->primary_listener);
3380         n->migration_state.notify = virtio_net_migration_state_notifier;
3381         add_migration_state_change_notifier(&n->migration_state);
3382         n->host_features |= (1ULL << VIRTIO_NET_F_STANDBY);
3383     }
3384 
3385     virtio_net_set_config_size(n, n->host_features);
3386     virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
3387 
3388     /*
3389      * We set a lower limit on RX queue size to what it always was.
3390      * Guests that want a smaller ring can always resize it without
3391      * help from us (using virtio 1 and up).
3392      */
3393     if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
3394         n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
3395         !is_power_of_2(n->net_conf.rx_queue_size)) {
3396         error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
3397                    "must be a power of 2 between %d and %d.",
3398                    n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
3399                    VIRTQUEUE_MAX_SIZE);
3400         virtio_cleanup(vdev);
3401         return;
3402     }
3403 
3404     if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE ||
3405         n->net_conf.tx_queue_size > VIRTQUEUE_MAX_SIZE ||
3406         !is_power_of_2(n->net_conf.tx_queue_size)) {
3407         error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), "
3408                    "must be a power of 2 between %d and %d",
3409                    n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE,
3410                    VIRTQUEUE_MAX_SIZE);
3411         virtio_cleanup(vdev);
3412         return;
3413     }
3414 
3415     n->max_ncs = MAX(n->nic_conf.peers.queues, 1);
3416 
3417     /*
3418      * Figure out the datapath queue pairs since the backend could
3419      * provide control queue via peers as well.
3420      */
3421     if (n->nic_conf.peers.queues) {
3422         for (i = 0; i < n->max_ncs; i++) {
3423             if (n->nic_conf.peers.ncs[i]->is_datapath) {
3424                 ++n->max_queue_pairs;
3425             }
3426         }
3427     }
3428     n->max_queue_pairs = MAX(n->max_queue_pairs, 1);
3429 
3430     if (n->max_queue_pairs * 2 + 1 > VIRTIO_QUEUE_MAX) {
3431         error_setg(errp, "Invalid number of queue pairs (= %" PRIu32 "), "
3432                    "must be a positive integer less than %d.",
3433                    n->max_queue_pairs, (VIRTIO_QUEUE_MAX - 1) / 2);
3434         virtio_cleanup(vdev);
3435         return;
3436     }
3437     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queue_pairs);
3438     n->curr_queue_pairs = 1;
3439     n->tx_timeout = n->net_conf.txtimer;
3440 
3441     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
3442                        && strcmp(n->net_conf.tx, "bh")) {
3443         warn_report("virtio-net: "
3444                     "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
3445                     n->net_conf.tx);
3446         error_printf("Defaulting to \"bh\"");
3447     }
3448 
3449     n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n),
3450                                     n->net_conf.tx_queue_size);
3451 
3452     for (i = 0; i < n->max_queue_pairs; i++) {
3453         virtio_net_add_queue(n, i);
3454     }
3455 
3456     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
3457     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
3458     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
3459     n->status = VIRTIO_NET_S_LINK_UP;
3460     qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
3461                               QEMU_CLOCK_VIRTUAL,
3462                               virtio_net_announce_timer, n);
3463     n->announce_timer.round = 0;
3464 
3465     if (n->netclient_type) {
3466         /*
3467          * Happen when virtio_net_set_netclient_name has been called.
3468          */
3469         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3470                               n->netclient_type, n->netclient_name, n);
3471     } else {
3472         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3473                               object_get_typename(OBJECT(dev)), dev->id, n);
3474     }
3475 
3476     for (i = 0; i < n->max_queue_pairs; i++) {
3477         n->nic->ncs[i].do_not_pad = true;
3478     }
3479 
3480     peer_test_vnet_hdr(n);
3481     if (peer_has_vnet_hdr(n)) {
3482         for (i = 0; i < n->max_queue_pairs; i++) {
3483             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
3484         }
3485         n->host_hdr_len = sizeof(struct virtio_net_hdr);
3486     } else {
3487         n->host_hdr_len = 0;
3488     }
3489 
3490     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
3491 
3492     n->vqs[0].tx_waiting = 0;
3493     n->tx_burst = n->net_conf.txburst;
3494     virtio_net_set_mrg_rx_bufs(n, 0, 0, 0);
3495     n->promisc = 1; /* for compatibility */
3496 
3497     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
3498 
3499     n->vlans = g_malloc0(MAX_VLAN >> 3);
3500 
3501     nc = qemu_get_queue(n->nic);
3502     nc->rxfilter_notify_enabled = 1;
3503 
3504    if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
3505         struct virtio_net_config netcfg = {};
3506         memcpy(&netcfg.mac, &n->nic_conf.macaddr, ETH_ALEN);
3507         vhost_net_set_config(get_vhost_net(nc->peer),
3508             (uint8_t *)&netcfg, 0, ETH_ALEN, VHOST_SET_CONFIG_TYPE_MASTER);
3509     }
3510     QTAILQ_INIT(&n->rsc_chains);
3511     n->qdev = dev;
3512 
3513     net_rx_pkt_init(&n->rx_pkt, false);
3514 
3515     if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) {
3516         virtio_net_load_ebpf(n);
3517     }
3518 }
3519 
3520 static void virtio_net_device_unrealize(DeviceState *dev)
3521 {
3522     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3523     VirtIONet *n = VIRTIO_NET(dev);
3524     int i, max_queue_pairs;
3525 
3526     if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) {
3527         virtio_net_unload_ebpf(n);
3528     }
3529 
3530     /* This will stop vhost backend if appropriate. */
3531     virtio_net_set_status(vdev, 0);
3532 
3533     g_free(n->netclient_name);
3534     n->netclient_name = NULL;
3535     g_free(n->netclient_type);
3536     n->netclient_type = NULL;
3537 
3538     g_free(n->mac_table.macs);
3539     g_free(n->vlans);
3540 
3541     if (n->failover) {
3542         qobject_unref(n->primary_opts);
3543         device_listener_unregister(&n->primary_listener);
3544         remove_migration_state_change_notifier(&n->migration_state);
3545     } else {
3546         assert(n->primary_opts == NULL);
3547     }
3548 
3549     max_queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
3550     for (i = 0; i < max_queue_pairs; i++) {
3551         virtio_net_del_queue(n, i);
3552     }
3553     /* delete also control vq */
3554     virtio_del_queue(vdev, max_queue_pairs * 2);
3555     qemu_announce_timer_del(&n->announce_timer, false);
3556     g_free(n->vqs);
3557     qemu_del_nic(n->nic);
3558     virtio_net_rsc_cleanup(n);
3559     g_free(n->rss_data.indirections_table);
3560     net_rx_pkt_uninit(n->rx_pkt);
3561     virtio_cleanup(vdev);
3562 }
3563 
3564 static void virtio_net_instance_init(Object *obj)
3565 {
3566     VirtIONet *n = VIRTIO_NET(obj);
3567 
3568     /*
3569      * The default config_size is sizeof(struct virtio_net_config).
3570      * Can be overriden with virtio_net_set_config_size.
3571      */
3572     n->config_size = sizeof(struct virtio_net_config);
3573     device_add_bootindex_property(obj, &n->nic_conf.bootindex,
3574                                   "bootindex", "/ethernet-phy@0",
3575                                   DEVICE(n));
3576 
3577     ebpf_rss_init(&n->ebpf_rss);
3578 }
3579 
3580 static int virtio_net_pre_save(void *opaque)
3581 {
3582     VirtIONet *n = opaque;
3583 
3584     /* At this point, backend must be stopped, otherwise
3585      * it might keep writing to memory. */
3586     assert(!n->vhost_started);
3587 
3588     return 0;
3589 }
3590 
3591 static bool primary_unplug_pending(void *opaque)
3592 {
3593     DeviceState *dev = opaque;
3594     DeviceState *primary;
3595     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3596     VirtIONet *n = VIRTIO_NET(vdev);
3597 
3598     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
3599         return false;
3600     }
3601     primary = failover_find_primary_device(n);
3602     return primary ? primary->pending_deleted_event : false;
3603 }
3604 
3605 static bool dev_unplug_pending(void *opaque)
3606 {
3607     DeviceState *dev = opaque;
3608     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3609 
3610     return vdc->primary_unplug_pending(dev);
3611 }
3612 
3613 static const VMStateDescription vmstate_virtio_net = {
3614     .name = "virtio-net",
3615     .minimum_version_id = VIRTIO_NET_VM_VERSION,
3616     .version_id = VIRTIO_NET_VM_VERSION,
3617     .fields = (VMStateField[]) {
3618         VMSTATE_VIRTIO_DEVICE,
3619         VMSTATE_END_OF_LIST()
3620     },
3621     .pre_save = virtio_net_pre_save,
3622     .dev_unplug_pending = dev_unplug_pending,
3623 };
3624 
3625 static Property virtio_net_properties[] = {
3626     DEFINE_PROP_BIT64("csum", VirtIONet, host_features,
3627                     VIRTIO_NET_F_CSUM, true),
3628     DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features,
3629                     VIRTIO_NET_F_GUEST_CSUM, true),
3630     DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
3631     DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features,
3632                     VIRTIO_NET_F_GUEST_TSO4, true),
3633     DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features,
3634                     VIRTIO_NET_F_GUEST_TSO6, true),
3635     DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features,
3636                     VIRTIO_NET_F_GUEST_ECN, true),
3637     DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features,
3638                     VIRTIO_NET_F_GUEST_UFO, true),
3639     DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features,
3640                     VIRTIO_NET_F_GUEST_ANNOUNCE, true),
3641     DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features,
3642                     VIRTIO_NET_F_HOST_TSO4, true),
3643     DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features,
3644                     VIRTIO_NET_F_HOST_TSO6, true),
3645     DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features,
3646                     VIRTIO_NET_F_HOST_ECN, true),
3647     DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features,
3648                     VIRTIO_NET_F_HOST_UFO, true),
3649     DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features,
3650                     VIRTIO_NET_F_MRG_RXBUF, true),
3651     DEFINE_PROP_BIT64("status", VirtIONet, host_features,
3652                     VIRTIO_NET_F_STATUS, true),
3653     DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features,
3654                     VIRTIO_NET_F_CTRL_VQ, true),
3655     DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features,
3656                     VIRTIO_NET_F_CTRL_RX, true),
3657     DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features,
3658                     VIRTIO_NET_F_CTRL_VLAN, true),
3659     DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features,
3660                     VIRTIO_NET_F_CTRL_RX_EXTRA, true),
3661     DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features,
3662                     VIRTIO_NET_F_CTRL_MAC_ADDR, true),
3663     DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features,
3664                     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
3665     DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
3666     DEFINE_PROP_BIT64("rss", VirtIONet, host_features,
3667                     VIRTIO_NET_F_RSS, false),
3668     DEFINE_PROP_BIT64("hash", VirtIONet, host_features,
3669                     VIRTIO_NET_F_HASH_REPORT, false),
3670     DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features,
3671                     VIRTIO_NET_F_RSC_EXT, false),
3672     DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout,
3673                        VIRTIO_NET_RSC_DEFAULT_INTERVAL),
3674     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
3675     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
3676                        TX_TIMER_INTERVAL),
3677     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
3678     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
3679     DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
3680                        VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
3681     DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size,
3682                        VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE),
3683     DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
3684     DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend,
3685                      true),
3686     DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN),
3687     DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str),
3688     DEFINE_PROP_BOOL("failover", VirtIONet, failover, false),
3689     DEFINE_PROP_END_OF_LIST(),
3690 };
3691 
3692 static void virtio_net_class_init(ObjectClass *klass, void *data)
3693 {
3694     DeviceClass *dc = DEVICE_CLASS(klass);
3695     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3696 
3697     device_class_set_props(dc, virtio_net_properties);
3698     dc->vmsd = &vmstate_virtio_net;
3699     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
3700     vdc->realize = virtio_net_device_realize;
3701     vdc->unrealize = virtio_net_device_unrealize;
3702     vdc->get_config = virtio_net_get_config;
3703     vdc->set_config = virtio_net_set_config;
3704     vdc->get_features = virtio_net_get_features;
3705     vdc->set_features = virtio_net_set_features;
3706     vdc->bad_features = virtio_net_bad_features;
3707     vdc->reset = virtio_net_reset;
3708     vdc->set_status = virtio_net_set_status;
3709     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
3710     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
3711     vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
3712     vdc->post_load = virtio_net_post_load_virtio;
3713     vdc->vmsd = &vmstate_virtio_net_device;
3714     vdc->primary_unplug_pending = primary_unplug_pending;
3715 }
3716 
3717 static const TypeInfo virtio_net_info = {
3718     .name = TYPE_VIRTIO_NET,
3719     .parent = TYPE_VIRTIO_DEVICE,
3720     .instance_size = sizeof(VirtIONet),
3721     .instance_init = virtio_net_instance_init,
3722     .class_init = virtio_net_class_init,
3723 };
3724 
3725 static void virtio_register_types(void)
3726 {
3727     type_register_static(&virtio_net_info);
3728 }
3729 
3730 type_init(virtio_register_types)
3731