xref: /openbmc/qemu/hw/net/virtio-net.c (revision dd84028f)
1 /*
2  * Virtio Network Device
3  *
4  * Copyright IBM, Corp. 2007
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/osdep.h"
15 #include "qemu/atomic.h"
16 #include "qemu/iov.h"
17 #include "qemu/log.h"
18 #include "qemu/main-loop.h"
19 #include "qemu/module.h"
20 #include "hw/virtio/virtio.h"
21 #include "net/net.h"
22 #include "net/checksum.h"
23 #include "net/tap.h"
24 #include "qemu/error-report.h"
25 #include "qemu/timer.h"
26 #include "qemu/option.h"
27 #include "qemu/option_int.h"
28 #include "qemu/config-file.h"
29 #include "qapi/qmp/qdict.h"
30 #include "hw/virtio/virtio-net.h"
31 #include "net/vhost_net.h"
32 #include "net/announce.h"
33 #include "hw/virtio/virtio-bus.h"
34 #include "qapi/error.h"
35 #include "qapi/qapi-events-net.h"
36 #include "hw/qdev-properties.h"
37 #include "qapi/qapi-types-migration.h"
38 #include "qapi/qapi-events-migration.h"
39 #include "hw/virtio/virtio-access.h"
40 #include "migration/misc.h"
41 #include "standard-headers/linux/ethtool.h"
42 #include "sysemu/sysemu.h"
43 #include "trace.h"
44 #include "monitor/qdev.h"
45 #include "hw/pci/pci_device.h"
46 #include "net_rx_pkt.h"
47 #include "hw/virtio/vhost.h"
48 #include "sysemu/qtest.h"
49 
50 #define VIRTIO_NET_VM_VERSION    11
51 
52 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
53 
54 /* previously fixed value */
55 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
56 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
57 
58 /* for now, only allow larger queue_pairs; with virtio-1, guest can downsize */
59 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
60 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
61 
62 #define VIRTIO_NET_IP4_ADDR_SIZE   8        /* ipv4 saddr + daddr */
63 
64 #define VIRTIO_NET_TCP_FLAG         0x3F
65 #define VIRTIO_NET_TCP_HDR_LENGTH   0xF000
66 
67 /* IPv4 max payload, 16 bits in the header */
68 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
69 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
70 
71 /* header length value in ip header without option */
72 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
73 
74 #define VIRTIO_NET_IP6_ADDR_SIZE   32      /* ipv6 saddr + daddr */
75 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
76 
77 /* Purge coalesced packets timer interval, This value affects the performance
78    a lot, and should be tuned carefully, '300000'(300us) is the recommended
79    value to pass the WHQL test, '50000' can gain 2x netperf throughput with
80    tso/gso/gro 'off'. */
81 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
82 
83 #define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \
84                                          VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \
85                                          VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \
86                                          VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \
87                                          VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \
88                                          VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \
89                                          VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \
90                                          VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \
91                                          VIRTIO_NET_RSS_HASH_TYPE_UDP_EX)
92 
93 static const VirtIOFeature feature_sizes[] = {
94     {.flags = 1ULL << VIRTIO_NET_F_MAC,
95      .end = endof(struct virtio_net_config, mac)},
96     {.flags = 1ULL << VIRTIO_NET_F_STATUS,
97      .end = endof(struct virtio_net_config, status)},
98     {.flags = 1ULL << VIRTIO_NET_F_MQ,
99      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
100     {.flags = 1ULL << VIRTIO_NET_F_MTU,
101      .end = endof(struct virtio_net_config, mtu)},
102     {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX,
103      .end = endof(struct virtio_net_config, duplex)},
104     {.flags = (1ULL << VIRTIO_NET_F_RSS) | (1ULL << VIRTIO_NET_F_HASH_REPORT),
105      .end = endof(struct virtio_net_config, supported_hash_types)},
106     {}
107 };
108 
109 static const VirtIOConfigSizeParams cfg_size_params = {
110     .min_size = endof(struct virtio_net_config, mac),
111     .max_size = sizeof(struct virtio_net_config),
112     .feature_sizes = feature_sizes
113 };
114 
115 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
116 {
117     VirtIONet *n = qemu_get_nic_opaque(nc);
118 
119     return &n->vqs[nc->queue_index];
120 }
121 
122 static int vq2q(int queue_index)
123 {
124     return queue_index / 2;
125 }
126 
127 static void flush_or_purge_queued_packets(NetClientState *nc)
128 {
129     if (!nc->peer) {
130         return;
131     }
132 
133     qemu_flush_or_purge_queued_packets(nc->peer, true);
134     assert(!virtio_net_get_subqueue(nc)->async_tx.elem);
135 }
136 
137 /* TODO
138  * - we could suppress RX interrupt if we were so inclined.
139  */
140 
141 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
142 {
143     VirtIONet *n = VIRTIO_NET(vdev);
144     struct virtio_net_config netcfg;
145     NetClientState *nc = qemu_get_queue(n->nic);
146     static const MACAddr zero = { .a = { 0, 0, 0, 0, 0, 0 } };
147 
148     int ret = 0;
149     memset(&netcfg, 0 , sizeof(struct virtio_net_config));
150     virtio_stw_p(vdev, &netcfg.status, n->status);
151     virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queue_pairs);
152     virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
153     memcpy(netcfg.mac, n->mac, ETH_ALEN);
154     virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed);
155     netcfg.duplex = n->net_conf.duplex;
156     netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE;
157     virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length,
158                  virtio_host_has_feature(vdev, VIRTIO_NET_F_RSS) ?
159                  VIRTIO_NET_RSS_MAX_TABLE_LEN : 1);
160     virtio_stl_p(vdev, &netcfg.supported_hash_types,
161                  VIRTIO_NET_RSS_SUPPORTED_HASHES);
162     memcpy(config, &netcfg, n->config_size);
163 
164     /*
165      * Is this VDPA? No peer means not VDPA: there's no way to
166      * disconnect/reconnect a VDPA peer.
167      */
168     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
169         ret = vhost_net_get_config(get_vhost_net(nc->peer), (uint8_t *)&netcfg,
170                                    n->config_size);
171         if (ret == -1) {
172             return;
173         }
174 
175         /*
176          * Some NIC/kernel combinations present 0 as the mac address.  As that
177          * is not a legal address, try to proceed with the address from the
178          * QEMU command line in the hope that the address has been configured
179          * correctly elsewhere - just not reported by the device.
180          */
181         if (memcmp(&netcfg.mac, &zero, sizeof(zero)) == 0) {
182             info_report("Zero hardware mac address detected. Ignoring.");
183             memcpy(netcfg.mac, n->mac, ETH_ALEN);
184         }
185 
186         netcfg.status |= virtio_tswap16(vdev,
187                                         n->status & VIRTIO_NET_S_ANNOUNCE);
188         memcpy(config, &netcfg, n->config_size);
189     }
190 }
191 
192 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
193 {
194     VirtIONet *n = VIRTIO_NET(vdev);
195     struct virtio_net_config netcfg = {};
196     NetClientState *nc = qemu_get_queue(n->nic);
197 
198     memcpy(&netcfg, config, n->config_size);
199 
200     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
201         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
202         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
203         memcpy(n->mac, netcfg.mac, ETH_ALEN);
204         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
205     }
206 
207     /*
208      * Is this VDPA? No peer means not VDPA: there's no way to
209      * disconnect/reconnect a VDPA peer.
210      */
211     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
212         vhost_net_set_config(get_vhost_net(nc->peer),
213                              (uint8_t *)&netcfg, 0, n->config_size,
214                              VHOST_SET_CONFIG_TYPE_MASTER);
215       }
216 }
217 
218 static bool virtio_net_started(VirtIONet *n, uint8_t status)
219 {
220     VirtIODevice *vdev = VIRTIO_DEVICE(n);
221     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
222         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
223 }
224 
225 static void virtio_net_announce_notify(VirtIONet *net)
226 {
227     VirtIODevice *vdev = VIRTIO_DEVICE(net);
228     trace_virtio_net_announce_notify();
229 
230     net->status |= VIRTIO_NET_S_ANNOUNCE;
231     virtio_notify_config(vdev);
232 }
233 
234 static void virtio_net_announce_timer(void *opaque)
235 {
236     VirtIONet *n = opaque;
237     trace_virtio_net_announce_timer(n->announce_timer.round);
238 
239     n->announce_timer.round--;
240     virtio_net_announce_notify(n);
241 }
242 
243 static void virtio_net_announce(NetClientState *nc)
244 {
245     VirtIONet *n = qemu_get_nic_opaque(nc);
246     VirtIODevice *vdev = VIRTIO_DEVICE(n);
247 
248     /*
249      * Make sure the virtio migration announcement timer isn't running
250      * If it is, let it trigger announcement so that we do not cause
251      * confusion.
252      */
253     if (n->announce_timer.round) {
254         return;
255     }
256 
257     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
258         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
259             virtio_net_announce_notify(n);
260     }
261 }
262 
263 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
264 {
265     VirtIODevice *vdev = VIRTIO_DEVICE(n);
266     NetClientState *nc = qemu_get_queue(n->nic);
267     int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
268     int cvq = virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) ?
269               n->max_ncs - n->max_queue_pairs : 0;
270 
271     if (!get_vhost_net(nc->peer)) {
272         return;
273     }
274 
275     if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
276         !!n->vhost_started) {
277         return;
278     }
279     if (!n->vhost_started) {
280         int r, i;
281 
282         if (n->needs_vnet_hdr_swap) {
283             error_report("backend does not support %s vnet headers; "
284                          "falling back on userspace virtio",
285                          virtio_is_big_endian(vdev) ? "BE" : "LE");
286             return;
287         }
288 
289         /* Any packets outstanding? Purge them to avoid touching rings
290          * when vhost is running.
291          */
292         for (i = 0;  i < queue_pairs; i++) {
293             NetClientState *qnc = qemu_get_subqueue(n->nic, i);
294 
295             /* Purge both directions: TX and RX. */
296             qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
297             qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
298         }
299 
300         if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
301             r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
302             if (r < 0) {
303                 error_report("%uBytes MTU not supported by the backend",
304                              n->net_conf.mtu);
305 
306                 return;
307             }
308         }
309 
310         n->vhost_started = 1;
311         r = vhost_net_start(vdev, n->nic->ncs, queue_pairs, cvq);
312         if (r < 0) {
313             error_report("unable to start vhost net: %d: "
314                          "falling back on userspace virtio", -r);
315             n->vhost_started = 0;
316         }
317     } else {
318         vhost_net_stop(vdev, n->nic->ncs, queue_pairs, cvq);
319         n->vhost_started = 0;
320     }
321 }
322 
323 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
324                                           NetClientState *peer,
325                                           bool enable)
326 {
327     if (virtio_is_big_endian(vdev)) {
328         return qemu_set_vnet_be(peer, enable);
329     } else {
330         return qemu_set_vnet_le(peer, enable);
331     }
332 }
333 
334 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
335                                        int queue_pairs, bool enable)
336 {
337     int i;
338 
339     for (i = 0; i < queue_pairs; i++) {
340         if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
341             enable) {
342             while (--i >= 0) {
343                 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
344             }
345 
346             return true;
347         }
348     }
349 
350     return false;
351 }
352 
353 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
354 {
355     VirtIODevice *vdev = VIRTIO_DEVICE(n);
356     int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
357 
358     if (virtio_net_started(n, status)) {
359         /* Before using the device, we tell the network backend about the
360          * endianness to use when parsing vnet headers. If the backend
361          * can't do it, we fallback onto fixing the headers in the core
362          * virtio-net code.
363          */
364         n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
365                                                             queue_pairs, true);
366     } else if (virtio_net_started(n, vdev->status)) {
367         /* After using the device, we need to reset the network backend to
368          * the default (guest native endianness), otherwise the guest may
369          * lose network connectivity if it is rebooted into a different
370          * endianness.
371          */
372         virtio_net_set_vnet_endian(vdev, n->nic->ncs, queue_pairs, false);
373     }
374 }
375 
376 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq)
377 {
378     unsigned int dropped = virtqueue_drop_all(vq);
379     if (dropped) {
380         virtio_notify(vdev, vq);
381     }
382 }
383 
384 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
385 {
386     VirtIONet *n = VIRTIO_NET(vdev);
387     VirtIONetQueue *q;
388     int i;
389     uint8_t queue_status;
390 
391     virtio_net_vnet_endian_status(n, status);
392     virtio_net_vhost_status(n, status);
393 
394     for (i = 0; i < n->max_queue_pairs; i++) {
395         NetClientState *ncs = qemu_get_subqueue(n->nic, i);
396         bool queue_started;
397         q = &n->vqs[i];
398 
399         if ((!n->multiqueue && i != 0) || i >= n->curr_queue_pairs) {
400             queue_status = 0;
401         } else {
402             queue_status = status;
403         }
404         queue_started =
405             virtio_net_started(n, queue_status) && !n->vhost_started;
406 
407         if (queue_started) {
408             qemu_flush_queued_packets(ncs);
409         }
410 
411         if (!q->tx_waiting) {
412             continue;
413         }
414 
415         if (queue_started) {
416             if (q->tx_timer) {
417                 timer_mod(q->tx_timer,
418                                qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
419             } else {
420                 qemu_bh_schedule(q->tx_bh);
421             }
422         } else {
423             if (q->tx_timer) {
424                 timer_del(q->tx_timer);
425             } else {
426                 qemu_bh_cancel(q->tx_bh);
427             }
428             if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 &&
429                 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) &&
430                 vdev->vm_running) {
431                 /* if tx is waiting we are likely have some packets in tx queue
432                  * and disabled notification */
433                 q->tx_waiting = 0;
434                 virtio_queue_set_notification(q->tx_vq, 1);
435                 virtio_net_drop_tx_queue_data(vdev, q->tx_vq);
436             }
437         }
438     }
439 }
440 
441 static void virtio_net_set_link_status(NetClientState *nc)
442 {
443     VirtIONet *n = qemu_get_nic_opaque(nc);
444     VirtIODevice *vdev = VIRTIO_DEVICE(n);
445     uint16_t old_status = n->status;
446 
447     if (nc->link_down)
448         n->status &= ~VIRTIO_NET_S_LINK_UP;
449     else
450         n->status |= VIRTIO_NET_S_LINK_UP;
451 
452     if (n->status != old_status)
453         virtio_notify_config(vdev);
454 
455     virtio_net_set_status(vdev, vdev->status);
456 }
457 
458 static void rxfilter_notify(NetClientState *nc)
459 {
460     VirtIONet *n = qemu_get_nic_opaque(nc);
461 
462     if (nc->rxfilter_notify_enabled) {
463         char *path = object_get_canonical_path(OBJECT(n->qdev));
464         qapi_event_send_nic_rx_filter_changed(n->netclient_name, path);
465         g_free(path);
466 
467         /* disable event notification to avoid events flooding */
468         nc->rxfilter_notify_enabled = 0;
469     }
470 }
471 
472 static intList *get_vlan_table(VirtIONet *n)
473 {
474     intList *list;
475     int i, j;
476 
477     list = NULL;
478     for (i = 0; i < MAX_VLAN >> 5; i++) {
479         for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
480             if (n->vlans[i] & (1U << j)) {
481                 QAPI_LIST_PREPEND(list, (i << 5) + j);
482             }
483         }
484     }
485 
486     return list;
487 }
488 
489 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
490 {
491     VirtIONet *n = qemu_get_nic_opaque(nc);
492     VirtIODevice *vdev = VIRTIO_DEVICE(n);
493     RxFilterInfo *info;
494     strList *str_list;
495     int i;
496 
497     info = g_malloc0(sizeof(*info));
498     info->name = g_strdup(nc->name);
499     info->promiscuous = n->promisc;
500 
501     if (n->nouni) {
502         info->unicast = RX_STATE_NONE;
503     } else if (n->alluni) {
504         info->unicast = RX_STATE_ALL;
505     } else {
506         info->unicast = RX_STATE_NORMAL;
507     }
508 
509     if (n->nomulti) {
510         info->multicast = RX_STATE_NONE;
511     } else if (n->allmulti) {
512         info->multicast = RX_STATE_ALL;
513     } else {
514         info->multicast = RX_STATE_NORMAL;
515     }
516 
517     info->broadcast_allowed = n->nobcast;
518     info->multicast_overflow = n->mac_table.multi_overflow;
519     info->unicast_overflow = n->mac_table.uni_overflow;
520 
521     info->main_mac = qemu_mac_strdup_printf(n->mac);
522 
523     str_list = NULL;
524     for (i = 0; i < n->mac_table.first_multi; i++) {
525         QAPI_LIST_PREPEND(str_list,
526                       qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN));
527     }
528     info->unicast_table = str_list;
529 
530     str_list = NULL;
531     for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
532         QAPI_LIST_PREPEND(str_list,
533                       qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN));
534     }
535     info->multicast_table = str_list;
536     info->vlan_table = get_vlan_table(n);
537 
538     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
539         info->vlan = RX_STATE_ALL;
540     } else if (!info->vlan_table) {
541         info->vlan = RX_STATE_NONE;
542     } else {
543         info->vlan = RX_STATE_NORMAL;
544     }
545 
546     /* enable event notification after query */
547     nc->rxfilter_notify_enabled = 1;
548 
549     return info;
550 }
551 
552 static void virtio_net_queue_reset(VirtIODevice *vdev, uint32_t queue_index)
553 {
554     VirtIONet *n = VIRTIO_NET(vdev);
555     NetClientState *nc;
556 
557     /* validate queue_index and skip for cvq */
558     if (queue_index >= n->max_queue_pairs * 2) {
559         return;
560     }
561 
562     nc = qemu_get_subqueue(n->nic, vq2q(queue_index));
563 
564     if (!nc->peer) {
565         return;
566     }
567 
568     if (get_vhost_net(nc->peer) &&
569         nc->peer->info->type == NET_CLIENT_DRIVER_TAP) {
570         vhost_net_virtqueue_reset(vdev, nc, queue_index);
571     }
572 
573     flush_or_purge_queued_packets(nc);
574 }
575 
576 static void virtio_net_queue_enable(VirtIODevice *vdev, uint32_t queue_index)
577 {
578     VirtIONet *n = VIRTIO_NET(vdev);
579     NetClientState *nc;
580     int r;
581 
582     /* validate queue_index and skip for cvq */
583     if (queue_index >= n->max_queue_pairs * 2) {
584         return;
585     }
586 
587     nc = qemu_get_subqueue(n->nic, vq2q(queue_index));
588 
589     if (!nc->peer || !vdev->vhost_started) {
590         return;
591     }
592 
593     if (get_vhost_net(nc->peer) &&
594         nc->peer->info->type == NET_CLIENT_DRIVER_TAP) {
595         r = vhost_net_virtqueue_restart(vdev, nc, queue_index);
596         if (r < 0) {
597             error_report("unable to restart vhost net virtqueue: %d, "
598                             "when resetting the queue", queue_index);
599         }
600     }
601 }
602 
603 static void virtio_net_reset(VirtIODevice *vdev)
604 {
605     VirtIONet *n = VIRTIO_NET(vdev);
606     int i;
607 
608     /* Reset back to compatibility mode */
609     n->promisc = 1;
610     n->allmulti = 0;
611     n->alluni = 0;
612     n->nomulti = 0;
613     n->nouni = 0;
614     n->nobcast = 0;
615     /* multiqueue is disabled by default */
616     n->curr_queue_pairs = 1;
617     timer_del(n->announce_timer.tm);
618     n->announce_timer.round = 0;
619     n->status &= ~VIRTIO_NET_S_ANNOUNCE;
620 
621     /* Flush any MAC and VLAN filter table state */
622     n->mac_table.in_use = 0;
623     n->mac_table.first_multi = 0;
624     n->mac_table.multi_overflow = 0;
625     n->mac_table.uni_overflow = 0;
626     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
627     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
628     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
629     memset(n->vlans, 0, MAX_VLAN >> 3);
630 
631     /* Flush any async TX */
632     for (i = 0;  i < n->max_queue_pairs; i++) {
633         flush_or_purge_queued_packets(qemu_get_subqueue(n->nic, i));
634     }
635 }
636 
637 static void peer_test_vnet_hdr(VirtIONet *n)
638 {
639     NetClientState *nc = qemu_get_queue(n->nic);
640     if (!nc->peer) {
641         return;
642     }
643 
644     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
645 }
646 
647 static int peer_has_vnet_hdr(VirtIONet *n)
648 {
649     return n->has_vnet_hdr;
650 }
651 
652 static int peer_has_ufo(VirtIONet *n)
653 {
654     if (!peer_has_vnet_hdr(n))
655         return 0;
656 
657     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
658 
659     return n->has_ufo;
660 }
661 
662 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
663                                        int version_1, int hash_report)
664 {
665     int i;
666     NetClientState *nc;
667 
668     n->mergeable_rx_bufs = mergeable_rx_bufs;
669 
670     if (version_1) {
671         n->guest_hdr_len = hash_report ?
672             sizeof(struct virtio_net_hdr_v1_hash) :
673             sizeof(struct virtio_net_hdr_mrg_rxbuf);
674         n->rss_data.populate_hash = !!hash_report;
675     } else {
676         n->guest_hdr_len = n->mergeable_rx_bufs ?
677             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
678             sizeof(struct virtio_net_hdr);
679     }
680 
681     for (i = 0; i < n->max_queue_pairs; i++) {
682         nc = qemu_get_subqueue(n->nic, i);
683 
684         if (peer_has_vnet_hdr(n) &&
685             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
686             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
687             n->host_hdr_len = n->guest_hdr_len;
688         }
689     }
690 }
691 
692 static int virtio_net_max_tx_queue_size(VirtIONet *n)
693 {
694     NetClientState *peer = n->nic_conf.peers.ncs[0];
695 
696     /*
697      * Backends other than vhost-user or vhost-vdpa don't support max queue
698      * size.
699      */
700     if (!peer) {
701         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
702     }
703 
704     switch(peer->info->type) {
705     case NET_CLIENT_DRIVER_VHOST_USER:
706     case NET_CLIENT_DRIVER_VHOST_VDPA:
707         return VIRTQUEUE_MAX_SIZE;
708     default:
709         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
710     };
711 }
712 
713 static int peer_attach(VirtIONet *n, int index)
714 {
715     NetClientState *nc = qemu_get_subqueue(n->nic, index);
716 
717     if (!nc->peer) {
718         return 0;
719     }
720 
721     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
722         vhost_set_vring_enable(nc->peer, 1);
723     }
724 
725     if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
726         return 0;
727     }
728 
729     if (n->max_queue_pairs == 1) {
730         return 0;
731     }
732 
733     return tap_enable(nc->peer);
734 }
735 
736 static int peer_detach(VirtIONet *n, int index)
737 {
738     NetClientState *nc = qemu_get_subqueue(n->nic, index);
739 
740     if (!nc->peer) {
741         return 0;
742     }
743 
744     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
745         vhost_set_vring_enable(nc->peer, 0);
746     }
747 
748     if (nc->peer->info->type !=  NET_CLIENT_DRIVER_TAP) {
749         return 0;
750     }
751 
752     return tap_disable(nc->peer);
753 }
754 
755 static void virtio_net_set_queue_pairs(VirtIONet *n)
756 {
757     int i;
758     int r;
759 
760     if (n->nic->peer_deleted) {
761         return;
762     }
763 
764     for (i = 0; i < n->max_queue_pairs; i++) {
765         if (i < n->curr_queue_pairs) {
766             r = peer_attach(n, i);
767             assert(!r);
768         } else {
769             r = peer_detach(n, i);
770             assert(!r);
771         }
772     }
773 }
774 
775 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
776 
777 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
778                                         Error **errp)
779 {
780     VirtIONet *n = VIRTIO_NET(vdev);
781     NetClientState *nc = qemu_get_queue(n->nic);
782 
783     /* Firstly sync all virtio-net possible supported features */
784     features |= n->host_features;
785 
786     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
787 
788     if (!peer_has_vnet_hdr(n)) {
789         virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
790         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
791         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
792         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
793 
794         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
795         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
796         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
797         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
798 
799         virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT);
800     }
801 
802     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
803         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
804         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
805     }
806 
807     if (!get_vhost_net(nc->peer)) {
808         virtio_add_feature(&features, VIRTIO_F_RING_RESET);
809         return features;
810     }
811 
812     if (!ebpf_rss_is_loaded(&n->ebpf_rss)) {
813         virtio_clear_feature(&features, VIRTIO_NET_F_RSS);
814     }
815     features = vhost_net_get_features(get_vhost_net(nc->peer), features);
816     vdev->backend_features = features;
817 
818     if (n->mtu_bypass_backend &&
819             (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) {
820         features |= (1ULL << VIRTIO_NET_F_MTU);
821     }
822 
823     /*
824      * Since GUEST_ANNOUNCE is emulated the feature bit could be set without
825      * enabled. This happens in the vDPA case.
826      *
827      * Make sure the feature set is not incoherent, as the driver could refuse
828      * to start.
829      *
830      * TODO: QEMU is able to emulate a CVQ just for guest_announce purposes,
831      * helping guest to notify the new location with vDPA devices that does not
832      * support it.
833      */
834     if (!virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_CTRL_VQ)) {
835         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ANNOUNCE);
836     }
837 
838     return features;
839 }
840 
841 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
842 {
843     uint64_t features = 0;
844 
845     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
846      * but also these: */
847     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
848     virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
849     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
850     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
851     virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
852 
853     return features;
854 }
855 
856 static void virtio_net_apply_guest_offloads(VirtIONet *n)
857 {
858     qemu_set_offload(qemu_get_queue(n->nic)->peer,
859             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
860             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
861             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
862             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
863             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
864 }
865 
866 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
867 {
868     static const uint64_t guest_offloads_mask =
869         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
870         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
871         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
872         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
873         (1ULL << VIRTIO_NET_F_GUEST_UFO);
874 
875     return guest_offloads_mask & features;
876 }
877 
878 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
879 {
880     VirtIODevice *vdev = VIRTIO_DEVICE(n);
881     return virtio_net_guest_offloads_by_features(vdev->guest_features);
882 }
883 
884 typedef struct {
885     VirtIONet *n;
886     DeviceState *dev;
887 } FailoverDevice;
888 
889 /**
890  * Set the failover primary device
891  *
892  * @opaque: FailoverId to setup
893  * @opts: opts for device we are handling
894  * @errp: returns an error if this function fails
895  */
896 static int failover_set_primary(DeviceState *dev, void *opaque)
897 {
898     FailoverDevice *fdev = opaque;
899     PCIDevice *pci_dev = (PCIDevice *)
900         object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE);
901 
902     if (!pci_dev) {
903         return 0;
904     }
905 
906     if (!g_strcmp0(pci_dev->failover_pair_id, fdev->n->netclient_name)) {
907         fdev->dev = dev;
908         return 1;
909     }
910 
911     return 0;
912 }
913 
914 /**
915  * Find the primary device for this failover virtio-net
916  *
917  * @n: VirtIONet device
918  * @errp: returns an error if this function fails
919  */
920 static DeviceState *failover_find_primary_device(VirtIONet *n)
921 {
922     FailoverDevice fdev = {
923         .n = n,
924     };
925 
926     qbus_walk_children(sysbus_get_default(), failover_set_primary, NULL,
927                        NULL, NULL, &fdev);
928     return fdev.dev;
929 }
930 
931 static void failover_add_primary(VirtIONet *n, Error **errp)
932 {
933     Error *err = NULL;
934     DeviceState *dev = failover_find_primary_device(n);
935 
936     if (dev) {
937         return;
938     }
939 
940     if (!n->primary_opts) {
941         error_setg(errp, "Primary device not found");
942         error_append_hint(errp, "Virtio-net failover will not work. Make "
943                           "sure primary device has parameter"
944                           " failover_pair_id=%s\n", n->netclient_name);
945         return;
946     }
947 
948     dev = qdev_device_add_from_qdict(n->primary_opts,
949                                      n->primary_opts_from_json,
950                                      &err);
951     if (err) {
952         qobject_unref(n->primary_opts);
953         n->primary_opts = NULL;
954     } else {
955         object_unref(OBJECT(dev));
956     }
957     error_propagate(errp, err);
958 }
959 
960 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
961 {
962     VirtIONet *n = VIRTIO_NET(vdev);
963     Error *err = NULL;
964     int i;
965 
966     if (n->mtu_bypass_backend &&
967             !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) {
968         features &= ~(1ULL << VIRTIO_NET_F_MTU);
969     }
970 
971     virtio_net_set_multiqueue(n,
972                               virtio_has_feature(features, VIRTIO_NET_F_RSS) ||
973                               virtio_has_feature(features, VIRTIO_NET_F_MQ));
974 
975     virtio_net_set_mrg_rx_bufs(n,
976                                virtio_has_feature(features,
977                                                   VIRTIO_NET_F_MRG_RXBUF),
978                                virtio_has_feature(features,
979                                                   VIRTIO_F_VERSION_1),
980                                virtio_has_feature(features,
981                                                   VIRTIO_NET_F_HASH_REPORT));
982 
983     n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
984         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4);
985     n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
986         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6);
987     n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS);
988 
989     if (n->has_vnet_hdr) {
990         n->curr_guest_offloads =
991             virtio_net_guest_offloads_by_features(features);
992         virtio_net_apply_guest_offloads(n);
993     }
994 
995     for (i = 0;  i < n->max_queue_pairs; i++) {
996         NetClientState *nc = qemu_get_subqueue(n->nic, i);
997 
998         if (!get_vhost_net(nc->peer)) {
999             continue;
1000         }
1001         vhost_net_ack_features(get_vhost_net(nc->peer), features);
1002 
1003         /*
1004          * keep acked_features in NetVhostUserState up-to-date so it
1005          * can't miss any features configured by guest virtio driver.
1006          */
1007         vhost_net_save_acked_features(nc->peer);
1008     }
1009 
1010     if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
1011         memset(n->vlans, 0, MAX_VLAN >> 3);
1012     } else {
1013         memset(n->vlans, 0xff, MAX_VLAN >> 3);
1014     }
1015 
1016     if (virtio_has_feature(features, VIRTIO_NET_F_STANDBY)) {
1017         qapi_event_send_failover_negotiated(n->netclient_name);
1018         qatomic_set(&n->failover_primary_hidden, false);
1019         failover_add_primary(n, &err);
1020         if (err) {
1021             if (!qtest_enabled()) {
1022                 warn_report_err(err);
1023             } else {
1024                 error_free(err);
1025             }
1026         }
1027     }
1028 }
1029 
1030 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
1031                                      struct iovec *iov, unsigned int iov_cnt)
1032 {
1033     uint8_t on;
1034     size_t s;
1035     NetClientState *nc = qemu_get_queue(n->nic);
1036 
1037     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
1038     if (s != sizeof(on)) {
1039         return VIRTIO_NET_ERR;
1040     }
1041 
1042     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
1043         n->promisc = on;
1044     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
1045         n->allmulti = on;
1046     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
1047         n->alluni = on;
1048     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
1049         n->nomulti = on;
1050     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
1051         n->nouni = on;
1052     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
1053         n->nobcast = on;
1054     } else {
1055         return VIRTIO_NET_ERR;
1056     }
1057 
1058     rxfilter_notify(nc);
1059 
1060     return VIRTIO_NET_OK;
1061 }
1062 
1063 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
1064                                      struct iovec *iov, unsigned int iov_cnt)
1065 {
1066     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1067     uint64_t offloads;
1068     size_t s;
1069 
1070     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1071         return VIRTIO_NET_ERR;
1072     }
1073 
1074     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
1075     if (s != sizeof(offloads)) {
1076         return VIRTIO_NET_ERR;
1077     }
1078 
1079     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
1080         uint64_t supported_offloads;
1081 
1082         offloads = virtio_ldq_p(vdev, &offloads);
1083 
1084         if (!n->has_vnet_hdr) {
1085             return VIRTIO_NET_ERR;
1086         }
1087 
1088         n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
1089             virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4);
1090         n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
1091             virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6);
1092         virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT);
1093 
1094         supported_offloads = virtio_net_supported_guest_offloads(n);
1095         if (offloads & ~supported_offloads) {
1096             return VIRTIO_NET_ERR;
1097         }
1098 
1099         n->curr_guest_offloads = offloads;
1100         virtio_net_apply_guest_offloads(n);
1101 
1102         return VIRTIO_NET_OK;
1103     } else {
1104         return VIRTIO_NET_ERR;
1105     }
1106 }
1107 
1108 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
1109                                  struct iovec *iov, unsigned int iov_cnt)
1110 {
1111     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1112     struct virtio_net_ctrl_mac mac_data;
1113     size_t s;
1114     NetClientState *nc = qemu_get_queue(n->nic);
1115 
1116     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
1117         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
1118             return VIRTIO_NET_ERR;
1119         }
1120         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
1121         assert(s == sizeof(n->mac));
1122         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
1123         rxfilter_notify(nc);
1124 
1125         return VIRTIO_NET_OK;
1126     }
1127 
1128     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
1129         return VIRTIO_NET_ERR;
1130     }
1131 
1132     int in_use = 0;
1133     int first_multi = 0;
1134     uint8_t uni_overflow = 0;
1135     uint8_t multi_overflow = 0;
1136     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1137 
1138     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1139                    sizeof(mac_data.entries));
1140     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1141     if (s != sizeof(mac_data.entries)) {
1142         goto error;
1143     }
1144     iov_discard_front(&iov, &iov_cnt, s);
1145 
1146     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
1147         goto error;
1148     }
1149 
1150     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
1151         s = iov_to_buf(iov, iov_cnt, 0, macs,
1152                        mac_data.entries * ETH_ALEN);
1153         if (s != mac_data.entries * ETH_ALEN) {
1154             goto error;
1155         }
1156         in_use += mac_data.entries;
1157     } else {
1158         uni_overflow = 1;
1159     }
1160 
1161     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
1162 
1163     first_multi = in_use;
1164 
1165     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1166                    sizeof(mac_data.entries));
1167     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1168     if (s != sizeof(mac_data.entries)) {
1169         goto error;
1170     }
1171 
1172     iov_discard_front(&iov, &iov_cnt, s);
1173 
1174     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
1175         goto error;
1176     }
1177 
1178     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
1179         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
1180                        mac_data.entries * ETH_ALEN);
1181         if (s != mac_data.entries * ETH_ALEN) {
1182             goto error;
1183         }
1184         in_use += mac_data.entries;
1185     } else {
1186         multi_overflow = 1;
1187     }
1188 
1189     n->mac_table.in_use = in_use;
1190     n->mac_table.first_multi = first_multi;
1191     n->mac_table.uni_overflow = uni_overflow;
1192     n->mac_table.multi_overflow = multi_overflow;
1193     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
1194     g_free(macs);
1195     rxfilter_notify(nc);
1196 
1197     return VIRTIO_NET_OK;
1198 
1199 error:
1200     g_free(macs);
1201     return VIRTIO_NET_ERR;
1202 }
1203 
1204 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
1205                                         struct iovec *iov, unsigned int iov_cnt)
1206 {
1207     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1208     uint16_t vid;
1209     size_t s;
1210     NetClientState *nc = qemu_get_queue(n->nic);
1211 
1212     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
1213     vid = virtio_lduw_p(vdev, &vid);
1214     if (s != sizeof(vid)) {
1215         return VIRTIO_NET_ERR;
1216     }
1217 
1218     if (vid >= MAX_VLAN)
1219         return VIRTIO_NET_ERR;
1220 
1221     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
1222         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
1223     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
1224         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
1225     else
1226         return VIRTIO_NET_ERR;
1227 
1228     rxfilter_notify(nc);
1229 
1230     return VIRTIO_NET_OK;
1231 }
1232 
1233 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
1234                                       struct iovec *iov, unsigned int iov_cnt)
1235 {
1236     trace_virtio_net_handle_announce(n->announce_timer.round);
1237     if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
1238         n->status & VIRTIO_NET_S_ANNOUNCE) {
1239         n->status &= ~VIRTIO_NET_S_ANNOUNCE;
1240         if (n->announce_timer.round) {
1241             qemu_announce_timer_step(&n->announce_timer);
1242         }
1243         return VIRTIO_NET_OK;
1244     } else {
1245         return VIRTIO_NET_ERR;
1246     }
1247 }
1248 
1249 static void virtio_net_detach_epbf_rss(VirtIONet *n);
1250 
1251 static void virtio_net_disable_rss(VirtIONet *n)
1252 {
1253     if (n->rss_data.enabled) {
1254         trace_virtio_net_rss_disable();
1255     }
1256     n->rss_data.enabled = false;
1257 
1258     virtio_net_detach_epbf_rss(n);
1259 }
1260 
1261 static bool virtio_net_attach_ebpf_to_backend(NICState *nic, int prog_fd)
1262 {
1263     NetClientState *nc = qemu_get_peer(qemu_get_queue(nic), 0);
1264     if (nc == NULL || nc->info->set_steering_ebpf == NULL) {
1265         return false;
1266     }
1267 
1268     return nc->info->set_steering_ebpf(nc, prog_fd);
1269 }
1270 
1271 static void rss_data_to_rss_config(struct VirtioNetRssData *data,
1272                                    struct EBPFRSSConfig *config)
1273 {
1274     config->redirect = data->redirect;
1275     config->populate_hash = data->populate_hash;
1276     config->hash_types = data->hash_types;
1277     config->indirections_len = data->indirections_len;
1278     config->default_queue = data->default_queue;
1279 }
1280 
1281 static bool virtio_net_attach_epbf_rss(VirtIONet *n)
1282 {
1283     struct EBPFRSSConfig config = {};
1284 
1285     if (!ebpf_rss_is_loaded(&n->ebpf_rss)) {
1286         return false;
1287     }
1288 
1289     rss_data_to_rss_config(&n->rss_data, &config);
1290 
1291     if (!ebpf_rss_set_all(&n->ebpf_rss, &config,
1292                           n->rss_data.indirections_table, n->rss_data.key)) {
1293         return false;
1294     }
1295 
1296     if (!virtio_net_attach_ebpf_to_backend(n->nic, n->ebpf_rss.program_fd)) {
1297         return false;
1298     }
1299 
1300     return true;
1301 }
1302 
1303 static void virtio_net_detach_epbf_rss(VirtIONet *n)
1304 {
1305     virtio_net_attach_ebpf_to_backend(n->nic, -1);
1306 }
1307 
1308 static bool virtio_net_load_ebpf(VirtIONet *n)
1309 {
1310     if (!virtio_net_attach_ebpf_to_backend(n->nic, -1)) {
1311         /* backend does't support steering ebpf */
1312         return false;
1313     }
1314 
1315     return ebpf_rss_load(&n->ebpf_rss);
1316 }
1317 
1318 static void virtio_net_unload_ebpf(VirtIONet *n)
1319 {
1320     virtio_net_attach_ebpf_to_backend(n->nic, -1);
1321     ebpf_rss_unload(&n->ebpf_rss);
1322 }
1323 
1324 static uint16_t virtio_net_handle_rss(VirtIONet *n,
1325                                       struct iovec *iov,
1326                                       unsigned int iov_cnt,
1327                                       bool do_rss)
1328 {
1329     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1330     struct virtio_net_rss_config cfg;
1331     size_t s, offset = 0, size_get;
1332     uint16_t queue_pairs, i;
1333     struct {
1334         uint16_t us;
1335         uint8_t b;
1336     } QEMU_PACKED temp;
1337     const char *err_msg = "";
1338     uint32_t err_value = 0;
1339 
1340     if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) {
1341         err_msg = "RSS is not negotiated";
1342         goto error;
1343     }
1344     if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) {
1345         err_msg = "Hash report is not negotiated";
1346         goto error;
1347     }
1348     size_get = offsetof(struct virtio_net_rss_config, indirection_table);
1349     s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get);
1350     if (s != size_get) {
1351         err_msg = "Short command buffer";
1352         err_value = (uint32_t)s;
1353         goto error;
1354     }
1355     n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types);
1356     n->rss_data.indirections_len =
1357         virtio_lduw_p(vdev, &cfg.indirection_table_mask);
1358     n->rss_data.indirections_len++;
1359     if (!do_rss) {
1360         n->rss_data.indirections_len = 1;
1361     }
1362     if (!is_power_of_2(n->rss_data.indirections_len)) {
1363         err_msg = "Invalid size of indirection table";
1364         err_value = n->rss_data.indirections_len;
1365         goto error;
1366     }
1367     if (n->rss_data.indirections_len > VIRTIO_NET_RSS_MAX_TABLE_LEN) {
1368         err_msg = "Too large indirection table";
1369         err_value = n->rss_data.indirections_len;
1370         goto error;
1371     }
1372     n->rss_data.default_queue = do_rss ?
1373         virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0;
1374     if (n->rss_data.default_queue >= n->max_queue_pairs) {
1375         err_msg = "Invalid default queue";
1376         err_value = n->rss_data.default_queue;
1377         goto error;
1378     }
1379     offset += size_get;
1380     size_get = sizeof(uint16_t) * n->rss_data.indirections_len;
1381     g_free(n->rss_data.indirections_table);
1382     n->rss_data.indirections_table = g_malloc(size_get);
1383     if (!n->rss_data.indirections_table) {
1384         err_msg = "Can't allocate indirections table";
1385         err_value = n->rss_data.indirections_len;
1386         goto error;
1387     }
1388     s = iov_to_buf(iov, iov_cnt, offset,
1389                    n->rss_data.indirections_table, size_get);
1390     if (s != size_get) {
1391         err_msg = "Short indirection table buffer";
1392         err_value = (uint32_t)s;
1393         goto error;
1394     }
1395     for (i = 0; i < n->rss_data.indirections_len; ++i) {
1396         uint16_t val = n->rss_data.indirections_table[i];
1397         n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val);
1398     }
1399     offset += size_get;
1400     size_get = sizeof(temp);
1401     s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get);
1402     if (s != size_get) {
1403         err_msg = "Can't get queue_pairs";
1404         err_value = (uint32_t)s;
1405         goto error;
1406     }
1407     queue_pairs = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queue_pairs;
1408     if (queue_pairs == 0 || queue_pairs > n->max_queue_pairs) {
1409         err_msg = "Invalid number of queue_pairs";
1410         err_value = queue_pairs;
1411         goto error;
1412     }
1413     if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) {
1414         err_msg = "Invalid key size";
1415         err_value = temp.b;
1416         goto error;
1417     }
1418     if (!temp.b && n->rss_data.hash_types) {
1419         err_msg = "No key provided";
1420         err_value = 0;
1421         goto error;
1422     }
1423     if (!temp.b && !n->rss_data.hash_types) {
1424         virtio_net_disable_rss(n);
1425         return queue_pairs;
1426     }
1427     offset += size_get;
1428     size_get = temp.b;
1429     s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get);
1430     if (s != size_get) {
1431         err_msg = "Can get key buffer";
1432         err_value = (uint32_t)s;
1433         goto error;
1434     }
1435     n->rss_data.enabled = true;
1436 
1437     if (!n->rss_data.populate_hash) {
1438         if (!virtio_net_attach_epbf_rss(n)) {
1439             /* EBPF must be loaded for vhost */
1440             if (get_vhost_net(qemu_get_queue(n->nic)->peer)) {
1441                 warn_report("Can't load eBPF RSS for vhost");
1442                 goto error;
1443             }
1444             /* fallback to software RSS */
1445             warn_report("Can't load eBPF RSS - fallback to software RSS");
1446             n->rss_data.enabled_software_rss = true;
1447         }
1448     } else {
1449         /* use software RSS for hash populating */
1450         /* and detach eBPF if was loaded before */
1451         virtio_net_detach_epbf_rss(n);
1452         n->rss_data.enabled_software_rss = true;
1453     }
1454 
1455     trace_virtio_net_rss_enable(n->rss_data.hash_types,
1456                                 n->rss_data.indirections_len,
1457                                 temp.b);
1458     return queue_pairs;
1459 error:
1460     trace_virtio_net_rss_error(err_msg, err_value);
1461     virtio_net_disable_rss(n);
1462     return 0;
1463 }
1464 
1465 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
1466                                 struct iovec *iov, unsigned int iov_cnt)
1467 {
1468     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1469     uint16_t queue_pairs;
1470     NetClientState *nc = qemu_get_queue(n->nic);
1471 
1472     virtio_net_disable_rss(n);
1473     if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) {
1474         queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, false);
1475         return queue_pairs ? VIRTIO_NET_OK : VIRTIO_NET_ERR;
1476     }
1477     if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
1478         queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, true);
1479     } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
1480         struct virtio_net_ctrl_mq mq;
1481         size_t s;
1482         if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) {
1483             return VIRTIO_NET_ERR;
1484         }
1485         s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
1486         if (s != sizeof(mq)) {
1487             return VIRTIO_NET_ERR;
1488         }
1489         queue_pairs = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
1490 
1491     } else {
1492         return VIRTIO_NET_ERR;
1493     }
1494 
1495     if (queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1496         queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1497         queue_pairs > n->max_queue_pairs ||
1498         !n->multiqueue) {
1499         return VIRTIO_NET_ERR;
1500     }
1501 
1502     n->curr_queue_pairs = queue_pairs;
1503     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
1504         /*
1505          * Avoid updating the backend for a vdpa device: We're only interested
1506          * in updating the device model queues.
1507          */
1508         return VIRTIO_NET_OK;
1509     }
1510     /* stop the backend before changing the number of queue_pairs to avoid handling a
1511      * disabled queue */
1512     virtio_net_set_status(vdev, vdev->status);
1513     virtio_net_set_queue_pairs(n);
1514 
1515     return VIRTIO_NET_OK;
1516 }
1517 
1518 size_t virtio_net_handle_ctrl_iov(VirtIODevice *vdev,
1519                                   const struct iovec *in_sg, unsigned in_num,
1520                                   const struct iovec *out_sg,
1521                                   unsigned out_num)
1522 {
1523     VirtIONet *n = VIRTIO_NET(vdev);
1524     struct virtio_net_ctrl_hdr ctrl;
1525     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
1526     size_t s;
1527     struct iovec *iov, *iov2;
1528 
1529     if (iov_size(in_sg, in_num) < sizeof(status) ||
1530         iov_size(out_sg, out_num) < sizeof(ctrl)) {
1531         virtio_error(vdev, "virtio-net ctrl missing headers");
1532         return 0;
1533     }
1534 
1535     iov2 = iov = g_memdup2(out_sg, sizeof(struct iovec) * out_num);
1536     s = iov_to_buf(iov, out_num, 0, &ctrl, sizeof(ctrl));
1537     iov_discard_front(&iov, &out_num, sizeof(ctrl));
1538     if (s != sizeof(ctrl)) {
1539         status = VIRTIO_NET_ERR;
1540     } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
1541         status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, out_num);
1542     } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
1543         status = virtio_net_handle_mac(n, ctrl.cmd, iov, out_num);
1544     } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
1545         status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, out_num);
1546     } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
1547         status = virtio_net_handle_announce(n, ctrl.cmd, iov, out_num);
1548     } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
1549         status = virtio_net_handle_mq(n, ctrl.cmd, iov, out_num);
1550     } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
1551         status = virtio_net_handle_offloads(n, ctrl.cmd, iov, out_num);
1552     }
1553 
1554     s = iov_from_buf(in_sg, in_num, 0, &status, sizeof(status));
1555     assert(s == sizeof(status));
1556 
1557     g_free(iov2);
1558     return sizeof(status);
1559 }
1560 
1561 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
1562 {
1563     VirtQueueElement *elem;
1564 
1565     for (;;) {
1566         size_t written;
1567         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
1568         if (!elem) {
1569             break;
1570         }
1571 
1572         written = virtio_net_handle_ctrl_iov(vdev, elem->in_sg, elem->in_num,
1573                                              elem->out_sg, elem->out_num);
1574         if (written > 0) {
1575             virtqueue_push(vq, elem, written);
1576             virtio_notify(vdev, vq);
1577             g_free(elem);
1578         } else {
1579             virtqueue_detach_element(vq, elem, 0);
1580             g_free(elem);
1581             break;
1582         }
1583     }
1584 }
1585 
1586 /* RX */
1587 
1588 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
1589 {
1590     VirtIONet *n = VIRTIO_NET(vdev);
1591     int queue_index = vq2q(virtio_get_queue_index(vq));
1592 
1593     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
1594 }
1595 
1596 static bool virtio_net_can_receive(NetClientState *nc)
1597 {
1598     VirtIONet *n = qemu_get_nic_opaque(nc);
1599     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1600     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1601 
1602     if (!vdev->vm_running) {
1603         return false;
1604     }
1605 
1606     if (nc->queue_index >= n->curr_queue_pairs) {
1607         return false;
1608     }
1609 
1610     if (!virtio_queue_ready(q->rx_vq) ||
1611         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1612         return false;
1613     }
1614 
1615     return true;
1616 }
1617 
1618 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1619 {
1620     VirtIONet *n = q->n;
1621     if (virtio_queue_empty(q->rx_vq) ||
1622         (n->mergeable_rx_bufs &&
1623          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1624         virtio_queue_set_notification(q->rx_vq, 1);
1625 
1626         /* To avoid a race condition where the guest has made some buffers
1627          * available after the above check but before notification was
1628          * enabled, check for available buffers again.
1629          */
1630         if (virtio_queue_empty(q->rx_vq) ||
1631             (n->mergeable_rx_bufs &&
1632              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1633             return 0;
1634         }
1635     }
1636 
1637     virtio_queue_set_notification(q->rx_vq, 0);
1638     return 1;
1639 }
1640 
1641 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1642 {
1643     virtio_tswap16s(vdev, &hdr->hdr_len);
1644     virtio_tswap16s(vdev, &hdr->gso_size);
1645     virtio_tswap16s(vdev, &hdr->csum_start);
1646     virtio_tswap16s(vdev, &hdr->csum_offset);
1647 }
1648 
1649 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1650  * it never finds out that the packets don't have valid checksums.  This
1651  * causes dhclient to get upset.  Fedora's carried a patch for ages to
1652  * fix this with Xen but it hasn't appeared in an upstream release of
1653  * dhclient yet.
1654  *
1655  * To avoid breaking existing guests, we catch udp packets and add
1656  * checksums.  This is terrible but it's better than hacking the guest
1657  * kernels.
1658  *
1659  * N.B. if we introduce a zero-copy API, this operation is no longer free so
1660  * we should provide a mechanism to disable it to avoid polluting the host
1661  * cache.
1662  */
1663 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1664                                         uint8_t *buf, size_t size)
1665 {
1666     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1667         (size > 27 && size < 1500) && /* normal sized MTU */
1668         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1669         (buf[23] == 17) && /* ip.protocol == UDP */
1670         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1671         net_checksum_calculate(buf, size, CSUM_UDP);
1672         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1673     }
1674 }
1675 
1676 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1677                            const void *buf, size_t size)
1678 {
1679     if (n->has_vnet_hdr) {
1680         /* FIXME this cast is evil */
1681         void *wbuf = (void *)buf;
1682         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1683                                     size - n->host_hdr_len);
1684 
1685         if (n->needs_vnet_hdr_swap) {
1686             virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1687         }
1688         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1689     } else {
1690         struct virtio_net_hdr hdr = {
1691             .flags = 0,
1692             .gso_type = VIRTIO_NET_HDR_GSO_NONE
1693         };
1694         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1695     }
1696 }
1697 
1698 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1699 {
1700     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1701     static const uint8_t vlan[] = {0x81, 0x00};
1702     uint8_t *ptr = (uint8_t *)buf;
1703     int i;
1704 
1705     if (n->promisc)
1706         return 1;
1707 
1708     ptr += n->host_hdr_len;
1709 
1710     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1711         int vid = lduw_be_p(ptr + 14) & 0xfff;
1712         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1713             return 0;
1714     }
1715 
1716     if (ptr[0] & 1) { // multicast
1717         if (!memcmp(ptr, bcast, sizeof(bcast))) {
1718             return !n->nobcast;
1719         } else if (n->nomulti) {
1720             return 0;
1721         } else if (n->allmulti || n->mac_table.multi_overflow) {
1722             return 1;
1723         }
1724 
1725         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1726             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1727                 return 1;
1728             }
1729         }
1730     } else { // unicast
1731         if (n->nouni) {
1732             return 0;
1733         } else if (n->alluni || n->mac_table.uni_overflow) {
1734             return 1;
1735         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1736             return 1;
1737         }
1738 
1739         for (i = 0; i < n->mac_table.first_multi; i++) {
1740             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1741                 return 1;
1742             }
1743         }
1744     }
1745 
1746     return 0;
1747 }
1748 
1749 static uint8_t virtio_net_get_hash_type(bool hasip4,
1750                                         bool hasip6,
1751                                         EthL4HdrProto l4hdr_proto,
1752                                         uint32_t types)
1753 {
1754     if (hasip4) {
1755         switch (l4hdr_proto) {
1756         case ETH_L4_HDR_PROTO_TCP:
1757             if (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4) {
1758                 return NetPktRssIpV4Tcp;
1759             }
1760             break;
1761 
1762         case ETH_L4_HDR_PROTO_UDP:
1763             if (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4) {
1764                 return NetPktRssIpV4Udp;
1765             }
1766             break;
1767 
1768         default:
1769             break;
1770         }
1771 
1772         if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
1773             return NetPktRssIpV4;
1774         }
1775     } else if (hasip6) {
1776         switch (l4hdr_proto) {
1777         case ETH_L4_HDR_PROTO_TCP:
1778             if (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) {
1779                 return NetPktRssIpV6TcpEx;
1780             }
1781             if (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv6) {
1782                 return NetPktRssIpV6Tcp;
1783             }
1784             break;
1785 
1786         case ETH_L4_HDR_PROTO_UDP:
1787             if (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) {
1788                 return NetPktRssIpV6UdpEx;
1789             }
1790             if (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv6) {
1791                 return NetPktRssIpV6Udp;
1792             }
1793             break;
1794 
1795         default:
1796             break;
1797         }
1798 
1799         if (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) {
1800             return NetPktRssIpV6Ex;
1801         }
1802         if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv6) {
1803             return NetPktRssIpV6;
1804         }
1805     }
1806     return 0xff;
1807 }
1808 
1809 static void virtio_set_packet_hash(const uint8_t *buf, uint8_t report,
1810                                    uint32_t hash)
1811 {
1812     struct virtio_net_hdr_v1_hash *hdr = (void *)buf;
1813     hdr->hash_value = hash;
1814     hdr->hash_report = report;
1815 }
1816 
1817 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf,
1818                                   size_t size)
1819 {
1820     VirtIONet *n = qemu_get_nic_opaque(nc);
1821     unsigned int index = nc->queue_index, new_index = index;
1822     struct NetRxPkt *pkt = n->rx_pkt;
1823     uint8_t net_hash_type;
1824     uint32_t hash;
1825     bool hasip4, hasip6;
1826     EthL4HdrProto l4hdr_proto;
1827     static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = {
1828         VIRTIO_NET_HASH_REPORT_IPv4,
1829         VIRTIO_NET_HASH_REPORT_TCPv4,
1830         VIRTIO_NET_HASH_REPORT_TCPv6,
1831         VIRTIO_NET_HASH_REPORT_IPv6,
1832         VIRTIO_NET_HASH_REPORT_IPv6_EX,
1833         VIRTIO_NET_HASH_REPORT_TCPv6_EX,
1834         VIRTIO_NET_HASH_REPORT_UDPv4,
1835         VIRTIO_NET_HASH_REPORT_UDPv6,
1836         VIRTIO_NET_HASH_REPORT_UDPv6_EX
1837     };
1838 
1839     net_rx_pkt_set_protocols(pkt, buf + n->host_hdr_len,
1840                              size - n->host_hdr_len);
1841     net_rx_pkt_get_protocols(pkt, &hasip4, &hasip6, &l4hdr_proto);
1842     net_hash_type = virtio_net_get_hash_type(hasip4, hasip6, l4hdr_proto,
1843                                              n->rss_data.hash_types);
1844     if (net_hash_type > NetPktRssIpV6UdpEx) {
1845         if (n->rss_data.populate_hash) {
1846             virtio_set_packet_hash(buf, VIRTIO_NET_HASH_REPORT_NONE, 0);
1847         }
1848         return n->rss_data.redirect ? n->rss_data.default_queue : -1;
1849     }
1850 
1851     hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key);
1852 
1853     if (n->rss_data.populate_hash) {
1854         virtio_set_packet_hash(buf, reports[net_hash_type], hash);
1855     }
1856 
1857     if (n->rss_data.redirect) {
1858         new_index = hash & (n->rss_data.indirections_len - 1);
1859         new_index = n->rss_data.indirections_table[new_index];
1860     }
1861 
1862     return (index == new_index) ? -1 : new_index;
1863 }
1864 
1865 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
1866                                       size_t size, bool no_rss)
1867 {
1868     VirtIONet *n = qemu_get_nic_opaque(nc);
1869     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1870     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1871     VirtQueueElement *elems[VIRTQUEUE_MAX_SIZE];
1872     size_t lens[VIRTQUEUE_MAX_SIZE];
1873     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1874     struct virtio_net_hdr_mrg_rxbuf mhdr;
1875     unsigned mhdr_cnt = 0;
1876     size_t offset, i, guest_offset, j;
1877     ssize_t err;
1878 
1879     if (!virtio_net_can_receive(nc)) {
1880         return -1;
1881     }
1882 
1883     if (!no_rss && n->rss_data.enabled && n->rss_data.enabled_software_rss) {
1884         int index = virtio_net_process_rss(nc, buf, size);
1885         if (index >= 0) {
1886             NetClientState *nc2 = qemu_get_subqueue(n->nic, index);
1887             return virtio_net_receive_rcu(nc2, buf, size, true);
1888         }
1889     }
1890 
1891     /* hdr_len refers to the header we supply to the guest */
1892     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1893         return 0;
1894     }
1895 
1896     if (!receive_filter(n, buf, size))
1897         return size;
1898 
1899     offset = i = 0;
1900 
1901     while (offset < size) {
1902         VirtQueueElement *elem;
1903         int len, total;
1904         const struct iovec *sg;
1905 
1906         total = 0;
1907 
1908         if (i == VIRTQUEUE_MAX_SIZE) {
1909             virtio_error(vdev, "virtio-net unexpected long buffer chain");
1910             err = size;
1911             goto err;
1912         }
1913 
1914         elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1915         if (!elem) {
1916             if (i) {
1917                 virtio_error(vdev, "virtio-net unexpected empty queue: "
1918                              "i %zd mergeable %d offset %zd, size %zd, "
1919                              "guest hdr len %zd, host hdr len %zd "
1920                              "guest features 0x%" PRIx64,
1921                              i, n->mergeable_rx_bufs, offset, size,
1922                              n->guest_hdr_len, n->host_hdr_len,
1923                              vdev->guest_features);
1924             }
1925             err = -1;
1926             goto err;
1927         }
1928 
1929         if (elem->in_num < 1) {
1930             virtio_error(vdev,
1931                          "virtio-net receive queue contains no in buffers");
1932             virtqueue_detach_element(q->rx_vq, elem, 0);
1933             g_free(elem);
1934             err = -1;
1935             goto err;
1936         }
1937 
1938         sg = elem->in_sg;
1939         if (i == 0) {
1940             assert(offset == 0);
1941             if (n->mergeable_rx_bufs) {
1942                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1943                                     sg, elem->in_num,
1944                                     offsetof(typeof(mhdr), num_buffers),
1945                                     sizeof(mhdr.num_buffers));
1946             }
1947 
1948             receive_header(n, sg, elem->in_num, buf, size);
1949             if (n->rss_data.populate_hash) {
1950                 offset = sizeof(mhdr);
1951                 iov_from_buf(sg, elem->in_num, offset,
1952                              buf + offset, n->host_hdr_len - sizeof(mhdr));
1953             }
1954             offset = n->host_hdr_len;
1955             total += n->guest_hdr_len;
1956             guest_offset = n->guest_hdr_len;
1957         } else {
1958             guest_offset = 0;
1959         }
1960 
1961         /* copy in packet.  ugh */
1962         len = iov_from_buf(sg, elem->in_num, guest_offset,
1963                            buf + offset, size - offset);
1964         total += len;
1965         offset += len;
1966         /* If buffers can't be merged, at this point we
1967          * must have consumed the complete packet.
1968          * Otherwise, drop it. */
1969         if (!n->mergeable_rx_bufs && offset < size) {
1970             virtqueue_unpop(q->rx_vq, elem, total);
1971             g_free(elem);
1972             err = size;
1973             goto err;
1974         }
1975 
1976         elems[i] = elem;
1977         lens[i] = total;
1978         i++;
1979     }
1980 
1981     if (mhdr_cnt) {
1982         virtio_stw_p(vdev, &mhdr.num_buffers, i);
1983         iov_from_buf(mhdr_sg, mhdr_cnt,
1984                      0,
1985                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
1986     }
1987 
1988     for (j = 0; j < i; j++) {
1989         /* signal other side */
1990         virtqueue_fill(q->rx_vq, elems[j], lens[j], j);
1991         g_free(elems[j]);
1992     }
1993 
1994     virtqueue_flush(q->rx_vq, i);
1995     virtio_notify(vdev, q->rx_vq);
1996 
1997     return size;
1998 
1999 err:
2000     for (j = 0; j < i; j++) {
2001         virtqueue_detach_element(q->rx_vq, elems[j], lens[j]);
2002         g_free(elems[j]);
2003     }
2004 
2005     return err;
2006 }
2007 
2008 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf,
2009                                   size_t size)
2010 {
2011     RCU_READ_LOCK_GUARD();
2012 
2013     return virtio_net_receive_rcu(nc, buf, size, false);
2014 }
2015 
2016 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain,
2017                                          const uint8_t *buf,
2018                                          VirtioNetRscUnit *unit)
2019 {
2020     uint16_t ip_hdrlen;
2021     struct ip_header *ip;
2022 
2023     ip = (struct ip_header *)(buf + chain->n->guest_hdr_len
2024                               + sizeof(struct eth_header));
2025     unit->ip = (void *)ip;
2026     ip_hdrlen = (ip->ip_ver_len & 0xF) << 2;
2027     unit->ip_plen = &ip->ip_len;
2028     unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen);
2029     unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
2030     unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen;
2031 }
2032 
2033 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain,
2034                                          const uint8_t *buf,
2035                                          VirtioNetRscUnit *unit)
2036 {
2037     struct ip6_header *ip6;
2038 
2039     ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len
2040                                  + sizeof(struct eth_header));
2041     unit->ip = ip6;
2042     unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
2043     unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip)
2044                                         + sizeof(struct ip6_header));
2045     unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
2046 
2047     /* There is a difference between payload lenght in ipv4 and v6,
2048        ip header is excluded in ipv6 */
2049     unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen;
2050 }
2051 
2052 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain,
2053                                        VirtioNetRscSeg *seg)
2054 {
2055     int ret;
2056     struct virtio_net_hdr_v1 *h;
2057 
2058     h = (struct virtio_net_hdr_v1 *)seg->buf;
2059     h->flags = 0;
2060     h->gso_type = VIRTIO_NET_HDR_GSO_NONE;
2061 
2062     if (seg->is_coalesced) {
2063         h->rsc.segments = seg->packets;
2064         h->rsc.dup_acks = seg->dup_ack;
2065         h->flags = VIRTIO_NET_HDR_F_RSC_INFO;
2066         if (chain->proto == ETH_P_IP) {
2067             h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2068         } else {
2069             h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2070         }
2071     }
2072 
2073     ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size);
2074     QTAILQ_REMOVE(&chain->buffers, seg, next);
2075     g_free(seg->buf);
2076     g_free(seg);
2077 
2078     return ret;
2079 }
2080 
2081 static void virtio_net_rsc_purge(void *opq)
2082 {
2083     VirtioNetRscSeg *seg, *rn;
2084     VirtioNetRscChain *chain = (VirtioNetRscChain *)opq;
2085 
2086     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) {
2087         if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2088             chain->stat.purge_failed++;
2089             continue;
2090         }
2091     }
2092 
2093     chain->stat.timer++;
2094     if (!QTAILQ_EMPTY(&chain->buffers)) {
2095         timer_mod(chain->drain_timer,
2096               qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
2097     }
2098 }
2099 
2100 static void virtio_net_rsc_cleanup(VirtIONet *n)
2101 {
2102     VirtioNetRscChain *chain, *rn_chain;
2103     VirtioNetRscSeg *seg, *rn_seg;
2104 
2105     QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) {
2106         QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) {
2107             QTAILQ_REMOVE(&chain->buffers, seg, next);
2108             g_free(seg->buf);
2109             g_free(seg);
2110         }
2111 
2112         timer_free(chain->drain_timer);
2113         QTAILQ_REMOVE(&n->rsc_chains, chain, next);
2114         g_free(chain);
2115     }
2116 }
2117 
2118 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain,
2119                                      NetClientState *nc,
2120                                      const uint8_t *buf, size_t size)
2121 {
2122     uint16_t hdr_len;
2123     VirtioNetRscSeg *seg;
2124 
2125     hdr_len = chain->n->guest_hdr_len;
2126     seg = g_new(VirtioNetRscSeg, 1);
2127     seg->buf = g_malloc(hdr_len + sizeof(struct eth_header)
2128         + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD);
2129     memcpy(seg->buf, buf, size);
2130     seg->size = size;
2131     seg->packets = 1;
2132     seg->dup_ack = 0;
2133     seg->is_coalesced = 0;
2134     seg->nc = nc;
2135 
2136     QTAILQ_INSERT_TAIL(&chain->buffers, seg, next);
2137     chain->stat.cache++;
2138 
2139     switch (chain->proto) {
2140     case ETH_P_IP:
2141         virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
2142         break;
2143     case ETH_P_IPV6:
2144         virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit);
2145         break;
2146     default:
2147         g_assert_not_reached();
2148     }
2149 }
2150 
2151 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain,
2152                                          VirtioNetRscSeg *seg,
2153                                          const uint8_t *buf,
2154                                          struct tcp_header *n_tcp,
2155                                          struct tcp_header *o_tcp)
2156 {
2157     uint32_t nack, oack;
2158     uint16_t nwin, owin;
2159 
2160     nack = htonl(n_tcp->th_ack);
2161     nwin = htons(n_tcp->th_win);
2162     oack = htonl(o_tcp->th_ack);
2163     owin = htons(o_tcp->th_win);
2164 
2165     if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) {
2166         chain->stat.ack_out_of_win++;
2167         return RSC_FINAL;
2168     } else if (nack == oack) {
2169         /* duplicated ack or window probe */
2170         if (nwin == owin) {
2171             /* duplicated ack, add dup ack count due to whql test up to 1 */
2172             chain->stat.dup_ack++;
2173             return RSC_FINAL;
2174         } else {
2175             /* Coalesce window update */
2176             o_tcp->th_win = n_tcp->th_win;
2177             chain->stat.win_update++;
2178             return RSC_COALESCE;
2179         }
2180     } else {
2181         /* pure ack, go to 'C', finalize*/
2182         chain->stat.pure_ack++;
2183         return RSC_FINAL;
2184     }
2185 }
2186 
2187 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain,
2188                                             VirtioNetRscSeg *seg,
2189                                             const uint8_t *buf,
2190                                             VirtioNetRscUnit *n_unit)
2191 {
2192     void *data;
2193     uint16_t o_ip_len;
2194     uint32_t nseq, oseq;
2195     VirtioNetRscUnit *o_unit;
2196 
2197     o_unit = &seg->unit;
2198     o_ip_len = htons(*o_unit->ip_plen);
2199     nseq = htonl(n_unit->tcp->th_seq);
2200     oseq = htonl(o_unit->tcp->th_seq);
2201 
2202     /* out of order or retransmitted. */
2203     if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) {
2204         chain->stat.data_out_of_win++;
2205         return RSC_FINAL;
2206     }
2207 
2208     data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen;
2209     if (nseq == oseq) {
2210         if ((o_unit->payload == 0) && n_unit->payload) {
2211             /* From no payload to payload, normal case, not a dup ack or etc */
2212             chain->stat.data_after_pure_ack++;
2213             goto coalesce;
2214         } else {
2215             return virtio_net_rsc_handle_ack(chain, seg, buf,
2216                                              n_unit->tcp, o_unit->tcp);
2217         }
2218     } else if ((nseq - oseq) != o_unit->payload) {
2219         /* Not a consistent packet, out of order */
2220         chain->stat.data_out_of_order++;
2221         return RSC_FINAL;
2222     } else {
2223 coalesce:
2224         if ((o_ip_len + n_unit->payload) > chain->max_payload) {
2225             chain->stat.over_size++;
2226             return RSC_FINAL;
2227         }
2228 
2229         /* Here comes the right data, the payload length in v4/v6 is different,
2230            so use the field value to update and record the new data len */
2231         o_unit->payload += n_unit->payload; /* update new data len */
2232 
2233         /* update field in ip header */
2234         *o_unit->ip_plen = htons(o_ip_len + n_unit->payload);
2235 
2236         /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
2237            for windows guest, while this may change the behavior for linux
2238            guest (only if it uses RSC feature). */
2239         o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags;
2240 
2241         o_unit->tcp->th_ack = n_unit->tcp->th_ack;
2242         o_unit->tcp->th_win = n_unit->tcp->th_win;
2243 
2244         memmove(seg->buf + seg->size, data, n_unit->payload);
2245         seg->size += n_unit->payload;
2246         seg->packets++;
2247         chain->stat.coalesced++;
2248         return RSC_COALESCE;
2249     }
2250 }
2251 
2252 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain,
2253                                         VirtioNetRscSeg *seg,
2254                                         const uint8_t *buf, size_t size,
2255                                         VirtioNetRscUnit *unit)
2256 {
2257     struct ip_header *ip1, *ip2;
2258 
2259     ip1 = (struct ip_header *)(unit->ip);
2260     ip2 = (struct ip_header *)(seg->unit.ip);
2261     if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst)
2262         || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2263         || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2264         chain->stat.no_match++;
2265         return RSC_NO_MATCH;
2266     }
2267 
2268     return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2269 }
2270 
2271 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain,
2272                                         VirtioNetRscSeg *seg,
2273                                         const uint8_t *buf, size_t size,
2274                                         VirtioNetRscUnit *unit)
2275 {
2276     struct ip6_header *ip1, *ip2;
2277 
2278     ip1 = (struct ip6_header *)(unit->ip);
2279     ip2 = (struct ip6_header *)(seg->unit.ip);
2280     if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address))
2281         || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address))
2282         || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2283         || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2284             chain->stat.no_match++;
2285             return RSC_NO_MATCH;
2286     }
2287 
2288     return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2289 }
2290 
2291 /* Packets with 'SYN' should bypass, other flag should be sent after drain
2292  * to prevent out of order */
2293 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain,
2294                                          struct tcp_header *tcp)
2295 {
2296     uint16_t tcp_hdr;
2297     uint16_t tcp_flag;
2298 
2299     tcp_flag = htons(tcp->th_offset_flags);
2300     tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10;
2301     tcp_flag &= VIRTIO_NET_TCP_FLAG;
2302     if (tcp_flag & TH_SYN) {
2303         chain->stat.tcp_syn++;
2304         return RSC_BYPASS;
2305     }
2306 
2307     if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) {
2308         chain->stat.tcp_ctrl_drain++;
2309         return RSC_FINAL;
2310     }
2311 
2312     if (tcp_hdr > sizeof(struct tcp_header)) {
2313         chain->stat.tcp_all_opt++;
2314         return RSC_FINAL;
2315     }
2316 
2317     return RSC_CANDIDATE;
2318 }
2319 
2320 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain,
2321                                          NetClientState *nc,
2322                                          const uint8_t *buf, size_t size,
2323                                          VirtioNetRscUnit *unit)
2324 {
2325     int ret;
2326     VirtioNetRscSeg *seg, *nseg;
2327 
2328     if (QTAILQ_EMPTY(&chain->buffers)) {
2329         chain->stat.empty_cache++;
2330         virtio_net_rsc_cache_buf(chain, nc, buf, size);
2331         timer_mod(chain->drain_timer,
2332               qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
2333         return size;
2334     }
2335 
2336     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2337         if (chain->proto == ETH_P_IP) {
2338             ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
2339         } else {
2340             ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit);
2341         }
2342 
2343         if (ret == RSC_FINAL) {
2344             if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2345                 /* Send failed */
2346                 chain->stat.final_failed++;
2347                 return 0;
2348             }
2349 
2350             /* Send current packet */
2351             return virtio_net_do_receive(nc, buf, size);
2352         } else if (ret == RSC_NO_MATCH) {
2353             continue;
2354         } else {
2355             /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
2356             seg->is_coalesced = 1;
2357             return size;
2358         }
2359     }
2360 
2361     chain->stat.no_match_cache++;
2362     virtio_net_rsc_cache_buf(chain, nc, buf, size);
2363     return size;
2364 }
2365 
2366 /* Drain a connection data, this is to avoid out of order segments */
2367 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain,
2368                                         NetClientState *nc,
2369                                         const uint8_t *buf, size_t size,
2370                                         uint16_t ip_start, uint16_t ip_size,
2371                                         uint16_t tcp_port)
2372 {
2373     VirtioNetRscSeg *seg, *nseg;
2374     uint32_t ppair1, ppair2;
2375 
2376     ppair1 = *(uint32_t *)(buf + tcp_port);
2377     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2378         ppair2 = *(uint32_t *)(seg->buf + tcp_port);
2379         if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size)
2380             || (ppair1 != ppair2)) {
2381             continue;
2382         }
2383         if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2384             chain->stat.drain_failed++;
2385         }
2386 
2387         break;
2388     }
2389 
2390     return virtio_net_do_receive(nc, buf, size);
2391 }
2392 
2393 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain,
2394                                             struct ip_header *ip,
2395                                             const uint8_t *buf, size_t size)
2396 {
2397     uint16_t ip_len;
2398 
2399     /* Not an ipv4 packet */
2400     if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) {
2401         chain->stat.ip_option++;
2402         return RSC_BYPASS;
2403     }
2404 
2405     /* Don't handle packets with ip option */
2406     if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) {
2407         chain->stat.ip_option++;
2408         return RSC_BYPASS;
2409     }
2410 
2411     if (ip->ip_p != IPPROTO_TCP) {
2412         chain->stat.bypass_not_tcp++;
2413         return RSC_BYPASS;
2414     }
2415 
2416     /* Don't handle packets with ip fragment */
2417     if (!(htons(ip->ip_off) & IP_DF)) {
2418         chain->stat.ip_frag++;
2419         return RSC_BYPASS;
2420     }
2421 
2422     /* Don't handle packets with ecn flag */
2423     if (IPTOS_ECN(ip->ip_tos)) {
2424         chain->stat.ip_ecn++;
2425         return RSC_BYPASS;
2426     }
2427 
2428     ip_len = htons(ip->ip_len);
2429     if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header))
2430         || ip_len > (size - chain->n->guest_hdr_len -
2431                      sizeof(struct eth_header))) {
2432         chain->stat.ip_hacked++;
2433         return RSC_BYPASS;
2434     }
2435 
2436     return RSC_CANDIDATE;
2437 }
2438 
2439 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain,
2440                                       NetClientState *nc,
2441                                       const uint8_t *buf, size_t size)
2442 {
2443     int32_t ret;
2444     uint16_t hdr_len;
2445     VirtioNetRscUnit unit;
2446 
2447     hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2448 
2449     if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)
2450         + sizeof(struct tcp_header))) {
2451         chain->stat.bypass_not_tcp++;
2452         return virtio_net_do_receive(nc, buf, size);
2453     }
2454 
2455     virtio_net_rsc_extract_unit4(chain, buf, &unit);
2456     if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size)
2457         != RSC_CANDIDATE) {
2458         return virtio_net_do_receive(nc, buf, size);
2459     }
2460 
2461     ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2462     if (ret == RSC_BYPASS) {
2463         return virtio_net_do_receive(nc, buf, size);
2464     } else if (ret == RSC_FINAL) {
2465         return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2466                 ((hdr_len + sizeof(struct eth_header)) + 12),
2467                 VIRTIO_NET_IP4_ADDR_SIZE,
2468                 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header));
2469     }
2470 
2471     return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2472 }
2473 
2474 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain,
2475                                             struct ip6_header *ip6,
2476                                             const uint8_t *buf, size_t size)
2477 {
2478     uint16_t ip_len;
2479 
2480     if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4)
2481         != IP_HEADER_VERSION_6) {
2482         return RSC_BYPASS;
2483     }
2484 
2485     /* Both option and protocol is checked in this */
2486     if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) {
2487         chain->stat.bypass_not_tcp++;
2488         return RSC_BYPASS;
2489     }
2490 
2491     ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
2492     if (ip_len < sizeof(struct tcp_header) ||
2493         ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header)
2494                   - sizeof(struct ip6_header))) {
2495         chain->stat.ip_hacked++;
2496         return RSC_BYPASS;
2497     }
2498 
2499     /* Don't handle packets with ecn flag */
2500     if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) {
2501         chain->stat.ip_ecn++;
2502         return RSC_BYPASS;
2503     }
2504 
2505     return RSC_CANDIDATE;
2506 }
2507 
2508 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc,
2509                                       const uint8_t *buf, size_t size)
2510 {
2511     int32_t ret;
2512     uint16_t hdr_len;
2513     VirtioNetRscChain *chain;
2514     VirtioNetRscUnit unit;
2515 
2516     chain = opq;
2517     hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2518 
2519     if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header)
2520         + sizeof(tcp_header))) {
2521         return virtio_net_do_receive(nc, buf, size);
2522     }
2523 
2524     virtio_net_rsc_extract_unit6(chain, buf, &unit);
2525     if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain,
2526                                                  unit.ip, buf, size)) {
2527         return virtio_net_do_receive(nc, buf, size);
2528     }
2529 
2530     ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2531     if (ret == RSC_BYPASS) {
2532         return virtio_net_do_receive(nc, buf, size);
2533     } else if (ret == RSC_FINAL) {
2534         return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2535                 ((hdr_len + sizeof(struct eth_header)) + 8),
2536                 VIRTIO_NET_IP6_ADDR_SIZE,
2537                 hdr_len + sizeof(struct eth_header)
2538                 + sizeof(struct ip6_header));
2539     }
2540 
2541     return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2542 }
2543 
2544 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n,
2545                                                       NetClientState *nc,
2546                                                       uint16_t proto)
2547 {
2548     VirtioNetRscChain *chain;
2549 
2550     if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) {
2551         return NULL;
2552     }
2553 
2554     QTAILQ_FOREACH(chain, &n->rsc_chains, next) {
2555         if (chain->proto == proto) {
2556             return chain;
2557         }
2558     }
2559 
2560     chain = g_malloc(sizeof(*chain));
2561     chain->n = n;
2562     chain->proto = proto;
2563     if (proto == (uint16_t)ETH_P_IP) {
2564         chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
2565         chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2566     } else {
2567         chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD;
2568         chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2569     }
2570     chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST,
2571                                       virtio_net_rsc_purge, chain);
2572     memset(&chain->stat, 0, sizeof(chain->stat));
2573 
2574     QTAILQ_INIT(&chain->buffers);
2575     QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next);
2576 
2577     return chain;
2578 }
2579 
2580 static ssize_t virtio_net_rsc_receive(NetClientState *nc,
2581                                       const uint8_t *buf,
2582                                       size_t size)
2583 {
2584     uint16_t proto;
2585     VirtioNetRscChain *chain;
2586     struct eth_header *eth;
2587     VirtIONet *n;
2588 
2589     n = qemu_get_nic_opaque(nc);
2590     if (size < (n->host_hdr_len + sizeof(struct eth_header))) {
2591         return virtio_net_do_receive(nc, buf, size);
2592     }
2593 
2594     eth = (struct eth_header *)(buf + n->guest_hdr_len);
2595     proto = htons(eth->h_proto);
2596 
2597     chain = virtio_net_rsc_lookup_chain(n, nc, proto);
2598     if (chain) {
2599         chain->stat.received++;
2600         if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) {
2601             return virtio_net_rsc_receive4(chain, nc, buf, size);
2602         } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) {
2603             return virtio_net_rsc_receive6(chain, nc, buf, size);
2604         }
2605     }
2606     return virtio_net_do_receive(nc, buf, size);
2607 }
2608 
2609 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf,
2610                                   size_t size)
2611 {
2612     VirtIONet *n = qemu_get_nic_opaque(nc);
2613     if ((n->rsc4_enabled || n->rsc6_enabled)) {
2614         return virtio_net_rsc_receive(nc, buf, size);
2615     } else {
2616         return virtio_net_do_receive(nc, buf, size);
2617     }
2618 }
2619 
2620 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
2621 
2622 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
2623 {
2624     VirtIONet *n = qemu_get_nic_opaque(nc);
2625     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
2626     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2627     int ret;
2628 
2629     virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
2630     virtio_notify(vdev, q->tx_vq);
2631 
2632     g_free(q->async_tx.elem);
2633     q->async_tx.elem = NULL;
2634 
2635     virtio_queue_set_notification(q->tx_vq, 1);
2636     ret = virtio_net_flush_tx(q);
2637     if (ret >= n->tx_burst) {
2638         /*
2639          * the flush has been stopped by tx_burst
2640          * we will not receive notification for the
2641          * remainining part, so re-schedule
2642          */
2643         virtio_queue_set_notification(q->tx_vq, 0);
2644         if (q->tx_bh) {
2645             qemu_bh_schedule(q->tx_bh);
2646         } else {
2647             timer_mod(q->tx_timer,
2648                       qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2649         }
2650         q->tx_waiting = 1;
2651     }
2652 }
2653 
2654 /* TX */
2655 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
2656 {
2657     VirtIONet *n = q->n;
2658     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2659     VirtQueueElement *elem;
2660     int32_t num_packets = 0;
2661     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
2662     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2663         return num_packets;
2664     }
2665 
2666     if (q->async_tx.elem) {
2667         virtio_queue_set_notification(q->tx_vq, 0);
2668         return num_packets;
2669     }
2670 
2671     for (;;) {
2672         ssize_t ret;
2673         unsigned int out_num;
2674         struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
2675         struct virtio_net_hdr_mrg_rxbuf mhdr;
2676 
2677         elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
2678         if (!elem) {
2679             break;
2680         }
2681 
2682         out_num = elem->out_num;
2683         out_sg = elem->out_sg;
2684         if (out_num < 1) {
2685             virtio_error(vdev, "virtio-net header not in first element");
2686             virtqueue_detach_element(q->tx_vq, elem, 0);
2687             g_free(elem);
2688             return -EINVAL;
2689         }
2690 
2691         if (n->has_vnet_hdr) {
2692             if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
2693                 n->guest_hdr_len) {
2694                 virtio_error(vdev, "virtio-net header incorrect");
2695                 virtqueue_detach_element(q->tx_vq, elem, 0);
2696                 g_free(elem);
2697                 return -EINVAL;
2698             }
2699             if (n->needs_vnet_hdr_swap) {
2700                 virtio_net_hdr_swap(vdev, (void *) &mhdr);
2701                 sg2[0].iov_base = &mhdr;
2702                 sg2[0].iov_len = n->guest_hdr_len;
2703                 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
2704                                    out_sg, out_num,
2705                                    n->guest_hdr_len, -1);
2706                 if (out_num == VIRTQUEUE_MAX_SIZE) {
2707                     goto drop;
2708                 }
2709                 out_num += 1;
2710                 out_sg = sg2;
2711             }
2712         }
2713         /*
2714          * If host wants to see the guest header as is, we can
2715          * pass it on unchanged. Otherwise, copy just the parts
2716          * that host is interested in.
2717          */
2718         assert(n->host_hdr_len <= n->guest_hdr_len);
2719         if (n->host_hdr_len != n->guest_hdr_len) {
2720             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
2721                                        out_sg, out_num,
2722                                        0, n->host_hdr_len);
2723             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
2724                              out_sg, out_num,
2725                              n->guest_hdr_len, -1);
2726             out_num = sg_num;
2727             out_sg = sg;
2728         }
2729 
2730         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
2731                                       out_sg, out_num, virtio_net_tx_complete);
2732         if (ret == 0) {
2733             virtio_queue_set_notification(q->tx_vq, 0);
2734             q->async_tx.elem = elem;
2735             return -EBUSY;
2736         }
2737 
2738 drop:
2739         virtqueue_push(q->tx_vq, elem, 0);
2740         virtio_notify(vdev, q->tx_vq);
2741         g_free(elem);
2742 
2743         if (++num_packets >= n->tx_burst) {
2744             break;
2745         }
2746     }
2747     return num_packets;
2748 }
2749 
2750 static void virtio_net_tx_timer(void *opaque);
2751 
2752 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
2753 {
2754     VirtIONet *n = VIRTIO_NET(vdev);
2755     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2756 
2757     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2758         virtio_net_drop_tx_queue_data(vdev, vq);
2759         return;
2760     }
2761 
2762     /* This happens when device was stopped but VCPU wasn't. */
2763     if (!vdev->vm_running) {
2764         q->tx_waiting = 1;
2765         return;
2766     }
2767 
2768     if (q->tx_waiting) {
2769         /* We already have queued packets, immediately flush */
2770         timer_del(q->tx_timer);
2771         virtio_net_tx_timer(q);
2772     } else {
2773         /* re-arm timer to flush it (and more) on next tick */
2774         timer_mod(q->tx_timer,
2775                   qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2776         q->tx_waiting = 1;
2777         virtio_queue_set_notification(vq, 0);
2778     }
2779 }
2780 
2781 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
2782 {
2783     VirtIONet *n = VIRTIO_NET(vdev);
2784     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2785 
2786     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2787         virtio_net_drop_tx_queue_data(vdev, vq);
2788         return;
2789     }
2790 
2791     if (unlikely(q->tx_waiting)) {
2792         return;
2793     }
2794     q->tx_waiting = 1;
2795     /* This happens when device was stopped but VCPU wasn't. */
2796     if (!vdev->vm_running) {
2797         return;
2798     }
2799     virtio_queue_set_notification(vq, 0);
2800     qemu_bh_schedule(q->tx_bh);
2801 }
2802 
2803 static void virtio_net_tx_timer(void *opaque)
2804 {
2805     VirtIONetQueue *q = opaque;
2806     VirtIONet *n = q->n;
2807     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2808     int ret;
2809 
2810     /* This happens when device was stopped but BH wasn't. */
2811     if (!vdev->vm_running) {
2812         /* Make sure tx waiting is set, so we'll run when restarted. */
2813         assert(q->tx_waiting);
2814         return;
2815     }
2816 
2817     q->tx_waiting = 0;
2818 
2819     /* Just in case the driver is not ready on more */
2820     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2821         return;
2822     }
2823 
2824     ret = virtio_net_flush_tx(q);
2825     if (ret == -EBUSY || ret == -EINVAL) {
2826         return;
2827     }
2828     /*
2829      * If we flush a full burst of packets, assume there are
2830      * more coming and immediately rearm
2831      */
2832     if (ret >= n->tx_burst) {
2833         q->tx_waiting = 1;
2834         timer_mod(q->tx_timer,
2835                   qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2836         return;
2837     }
2838     /*
2839      * If less than a full burst, re-enable notification and flush
2840      * anything that may have come in while we weren't looking.  If
2841      * we find something, assume the guest is still active and rearm
2842      */
2843     virtio_queue_set_notification(q->tx_vq, 1);
2844     ret = virtio_net_flush_tx(q);
2845     if (ret > 0) {
2846         virtio_queue_set_notification(q->tx_vq, 0);
2847         q->tx_waiting = 1;
2848         timer_mod(q->tx_timer,
2849                   qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2850     }
2851 }
2852 
2853 static void virtio_net_tx_bh(void *opaque)
2854 {
2855     VirtIONetQueue *q = opaque;
2856     VirtIONet *n = q->n;
2857     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2858     int32_t ret;
2859 
2860     /* This happens when device was stopped but BH wasn't. */
2861     if (!vdev->vm_running) {
2862         /* Make sure tx waiting is set, so we'll run when restarted. */
2863         assert(q->tx_waiting);
2864         return;
2865     }
2866 
2867     q->tx_waiting = 0;
2868 
2869     /* Just in case the driver is not ready on more */
2870     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
2871         return;
2872     }
2873 
2874     ret = virtio_net_flush_tx(q);
2875     if (ret == -EBUSY || ret == -EINVAL) {
2876         return; /* Notification re-enable handled by tx_complete or device
2877                  * broken */
2878     }
2879 
2880     /* If we flush a full burst of packets, assume there are
2881      * more coming and immediately reschedule */
2882     if (ret >= n->tx_burst) {
2883         qemu_bh_schedule(q->tx_bh);
2884         q->tx_waiting = 1;
2885         return;
2886     }
2887 
2888     /* If less than a full burst, re-enable notification and flush
2889      * anything that may have come in while we weren't looking.  If
2890      * we find something, assume the guest is still active and reschedule */
2891     virtio_queue_set_notification(q->tx_vq, 1);
2892     ret = virtio_net_flush_tx(q);
2893     if (ret == -EINVAL) {
2894         return;
2895     } else if (ret > 0) {
2896         virtio_queue_set_notification(q->tx_vq, 0);
2897         qemu_bh_schedule(q->tx_bh);
2898         q->tx_waiting = 1;
2899     }
2900 }
2901 
2902 static void virtio_net_add_queue(VirtIONet *n, int index)
2903 {
2904     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2905 
2906     n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
2907                                            virtio_net_handle_rx);
2908 
2909     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
2910         n->vqs[index].tx_vq =
2911             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2912                              virtio_net_handle_tx_timer);
2913         n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
2914                                               virtio_net_tx_timer,
2915                                               &n->vqs[index]);
2916     } else {
2917         n->vqs[index].tx_vq =
2918             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2919                              virtio_net_handle_tx_bh);
2920         n->vqs[index].tx_bh = qemu_bh_new_guarded(virtio_net_tx_bh, &n->vqs[index],
2921                                                   &DEVICE(vdev)->mem_reentrancy_guard);
2922     }
2923 
2924     n->vqs[index].tx_waiting = 0;
2925     n->vqs[index].n = n;
2926 }
2927 
2928 static void virtio_net_del_queue(VirtIONet *n, int index)
2929 {
2930     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2931     VirtIONetQueue *q = &n->vqs[index];
2932     NetClientState *nc = qemu_get_subqueue(n->nic, index);
2933 
2934     qemu_purge_queued_packets(nc);
2935 
2936     virtio_del_queue(vdev, index * 2);
2937     if (q->tx_timer) {
2938         timer_free(q->tx_timer);
2939         q->tx_timer = NULL;
2940     } else {
2941         qemu_bh_delete(q->tx_bh);
2942         q->tx_bh = NULL;
2943     }
2944     q->tx_waiting = 0;
2945     virtio_del_queue(vdev, index * 2 + 1);
2946 }
2947 
2948 static void virtio_net_change_num_queue_pairs(VirtIONet *n, int new_max_queue_pairs)
2949 {
2950     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2951     int old_num_queues = virtio_get_num_queues(vdev);
2952     int new_num_queues = new_max_queue_pairs * 2 + 1;
2953     int i;
2954 
2955     assert(old_num_queues >= 3);
2956     assert(old_num_queues % 2 == 1);
2957 
2958     if (old_num_queues == new_num_queues) {
2959         return;
2960     }
2961 
2962     /*
2963      * We always need to remove and add ctrl vq if
2964      * old_num_queues != new_num_queues. Remove ctrl_vq first,
2965      * and then we only enter one of the following two loops.
2966      */
2967     virtio_del_queue(vdev, old_num_queues - 1);
2968 
2969     for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
2970         /* new_num_queues < old_num_queues */
2971         virtio_net_del_queue(n, i / 2);
2972     }
2973 
2974     for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
2975         /* new_num_queues > old_num_queues */
2976         virtio_net_add_queue(n, i / 2);
2977     }
2978 
2979     /* add ctrl_vq last */
2980     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
2981 }
2982 
2983 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
2984 {
2985     int max = multiqueue ? n->max_queue_pairs : 1;
2986 
2987     n->multiqueue = multiqueue;
2988     virtio_net_change_num_queue_pairs(n, max);
2989 
2990     virtio_net_set_queue_pairs(n);
2991 }
2992 
2993 static int virtio_net_post_load_device(void *opaque, int version_id)
2994 {
2995     VirtIONet *n = opaque;
2996     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2997     int i, link_down;
2998 
2999     trace_virtio_net_post_load_device();
3000     virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
3001                                virtio_vdev_has_feature(vdev,
3002                                                        VIRTIO_F_VERSION_1),
3003                                virtio_vdev_has_feature(vdev,
3004                                                        VIRTIO_NET_F_HASH_REPORT));
3005 
3006     /* MAC_TABLE_ENTRIES may be different from the saved image */
3007     if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
3008         n->mac_table.in_use = 0;
3009     }
3010 
3011     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
3012         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
3013     }
3014 
3015     /*
3016      * curr_guest_offloads will be later overwritten by the
3017      * virtio_set_features_nocheck call done from the virtio_load.
3018      * Here we make sure it is preserved and restored accordingly
3019      * in the virtio_net_post_load_virtio callback.
3020      */
3021     n->saved_guest_offloads = n->curr_guest_offloads;
3022 
3023     virtio_net_set_queue_pairs(n);
3024 
3025     /* Find the first multicast entry in the saved MAC filter */
3026     for (i = 0; i < n->mac_table.in_use; i++) {
3027         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
3028             break;
3029         }
3030     }
3031     n->mac_table.first_multi = i;
3032 
3033     /* nc.link_down can't be migrated, so infer link_down according
3034      * to link status bit in n->status */
3035     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
3036     for (i = 0; i < n->max_queue_pairs; i++) {
3037         qemu_get_subqueue(n->nic, i)->link_down = link_down;
3038     }
3039 
3040     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
3041         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
3042         qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
3043                                   QEMU_CLOCK_VIRTUAL,
3044                                   virtio_net_announce_timer, n);
3045         if (n->announce_timer.round) {
3046             timer_mod(n->announce_timer.tm,
3047                       qemu_clock_get_ms(n->announce_timer.type));
3048         } else {
3049             qemu_announce_timer_del(&n->announce_timer, false);
3050         }
3051     }
3052 
3053     if (n->rss_data.enabled) {
3054         n->rss_data.enabled_software_rss = n->rss_data.populate_hash;
3055         if (!n->rss_data.populate_hash) {
3056             if (!virtio_net_attach_epbf_rss(n)) {
3057                 if (get_vhost_net(qemu_get_queue(n->nic)->peer)) {
3058                     warn_report("Can't post-load eBPF RSS for vhost");
3059                 } else {
3060                     warn_report("Can't post-load eBPF RSS - "
3061                                 "fallback to software RSS");
3062                     n->rss_data.enabled_software_rss = true;
3063                 }
3064             }
3065         }
3066 
3067         trace_virtio_net_rss_enable(n->rss_data.hash_types,
3068                                     n->rss_data.indirections_len,
3069                                     sizeof(n->rss_data.key));
3070     } else {
3071         trace_virtio_net_rss_disable();
3072     }
3073     return 0;
3074 }
3075 
3076 static int virtio_net_post_load_virtio(VirtIODevice *vdev)
3077 {
3078     VirtIONet *n = VIRTIO_NET(vdev);
3079     /*
3080      * The actual needed state is now in saved_guest_offloads,
3081      * see virtio_net_post_load_device for detail.
3082      * Restore it back and apply the desired offloads.
3083      */
3084     n->curr_guest_offloads = n->saved_guest_offloads;
3085     if (peer_has_vnet_hdr(n)) {
3086         virtio_net_apply_guest_offloads(n);
3087     }
3088 
3089     return 0;
3090 }
3091 
3092 /* tx_waiting field of a VirtIONetQueue */
3093 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = {
3094     .name = "virtio-net-queue-tx_waiting",
3095     .fields = (VMStateField[]) {
3096         VMSTATE_UINT32(tx_waiting, VirtIONetQueue),
3097         VMSTATE_END_OF_LIST()
3098    },
3099 };
3100 
3101 static bool max_queue_pairs_gt_1(void *opaque, int version_id)
3102 {
3103     return VIRTIO_NET(opaque)->max_queue_pairs > 1;
3104 }
3105 
3106 static bool has_ctrl_guest_offloads(void *opaque, int version_id)
3107 {
3108     return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque),
3109                                    VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3110 }
3111 
3112 static bool mac_table_fits(void *opaque, int version_id)
3113 {
3114     return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES;
3115 }
3116 
3117 static bool mac_table_doesnt_fit(void *opaque, int version_id)
3118 {
3119     return !mac_table_fits(opaque, version_id);
3120 }
3121 
3122 /* This temporary type is shared by all the WITH_TMP methods
3123  * although only some fields are used by each.
3124  */
3125 struct VirtIONetMigTmp {
3126     VirtIONet      *parent;
3127     VirtIONetQueue *vqs_1;
3128     uint16_t        curr_queue_pairs_1;
3129     uint8_t         has_ufo;
3130     uint32_t        has_vnet_hdr;
3131 };
3132 
3133 /* The 2nd and subsequent tx_waiting flags are loaded later than
3134  * the 1st entry in the queue_pairs and only if there's more than one
3135  * entry.  We use the tmp mechanism to calculate a temporary
3136  * pointer and count and also validate the count.
3137  */
3138 
3139 static int virtio_net_tx_waiting_pre_save(void *opaque)
3140 {
3141     struct VirtIONetMigTmp *tmp = opaque;
3142 
3143     tmp->vqs_1 = tmp->parent->vqs + 1;
3144     tmp->curr_queue_pairs_1 = tmp->parent->curr_queue_pairs - 1;
3145     if (tmp->parent->curr_queue_pairs == 0) {
3146         tmp->curr_queue_pairs_1 = 0;
3147     }
3148 
3149     return 0;
3150 }
3151 
3152 static int virtio_net_tx_waiting_pre_load(void *opaque)
3153 {
3154     struct VirtIONetMigTmp *tmp = opaque;
3155 
3156     /* Reuse the pointer setup from save */
3157     virtio_net_tx_waiting_pre_save(opaque);
3158 
3159     if (tmp->parent->curr_queue_pairs > tmp->parent->max_queue_pairs) {
3160         error_report("virtio-net: curr_queue_pairs %x > max_queue_pairs %x",
3161             tmp->parent->curr_queue_pairs, tmp->parent->max_queue_pairs);
3162 
3163         return -EINVAL;
3164     }
3165 
3166     return 0; /* all good */
3167 }
3168 
3169 static const VMStateDescription vmstate_virtio_net_tx_waiting = {
3170     .name      = "virtio-net-tx_waiting",
3171     .pre_load  = virtio_net_tx_waiting_pre_load,
3172     .pre_save  = virtio_net_tx_waiting_pre_save,
3173     .fields    = (VMStateField[]) {
3174         VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp,
3175                                      curr_queue_pairs_1,
3176                                      vmstate_virtio_net_queue_tx_waiting,
3177                                      struct VirtIONetQueue),
3178         VMSTATE_END_OF_LIST()
3179     },
3180 };
3181 
3182 /* the 'has_ufo' flag is just tested; if the incoming stream has the
3183  * flag set we need to check that we have it
3184  */
3185 static int virtio_net_ufo_post_load(void *opaque, int version_id)
3186 {
3187     struct VirtIONetMigTmp *tmp = opaque;
3188 
3189     if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) {
3190         error_report("virtio-net: saved image requires TUN_F_UFO support");
3191         return -EINVAL;
3192     }
3193 
3194     return 0;
3195 }
3196 
3197 static int virtio_net_ufo_pre_save(void *opaque)
3198 {
3199     struct VirtIONetMigTmp *tmp = opaque;
3200 
3201     tmp->has_ufo = tmp->parent->has_ufo;
3202 
3203     return 0;
3204 }
3205 
3206 static const VMStateDescription vmstate_virtio_net_has_ufo = {
3207     .name      = "virtio-net-ufo",
3208     .post_load = virtio_net_ufo_post_load,
3209     .pre_save  = virtio_net_ufo_pre_save,
3210     .fields    = (VMStateField[]) {
3211         VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp),
3212         VMSTATE_END_OF_LIST()
3213     },
3214 };
3215 
3216 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
3217  * flag set we need to check that we have it
3218  */
3219 static int virtio_net_vnet_post_load(void *opaque, int version_id)
3220 {
3221     struct VirtIONetMigTmp *tmp = opaque;
3222 
3223     if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) {
3224         error_report("virtio-net: saved image requires vnet_hdr=on");
3225         return -EINVAL;
3226     }
3227 
3228     return 0;
3229 }
3230 
3231 static int virtio_net_vnet_pre_save(void *opaque)
3232 {
3233     struct VirtIONetMigTmp *tmp = opaque;
3234 
3235     tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr;
3236 
3237     return 0;
3238 }
3239 
3240 static const VMStateDescription vmstate_virtio_net_has_vnet = {
3241     .name      = "virtio-net-vnet",
3242     .post_load = virtio_net_vnet_post_load,
3243     .pre_save  = virtio_net_vnet_pre_save,
3244     .fields    = (VMStateField[]) {
3245         VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp),
3246         VMSTATE_END_OF_LIST()
3247     },
3248 };
3249 
3250 static bool virtio_net_rss_needed(void *opaque)
3251 {
3252     return VIRTIO_NET(opaque)->rss_data.enabled;
3253 }
3254 
3255 static const VMStateDescription vmstate_virtio_net_rss = {
3256     .name      = "virtio-net-device/rss",
3257     .version_id = 1,
3258     .minimum_version_id = 1,
3259     .needed = virtio_net_rss_needed,
3260     .fields = (VMStateField[]) {
3261         VMSTATE_BOOL(rss_data.enabled, VirtIONet),
3262         VMSTATE_BOOL(rss_data.redirect, VirtIONet),
3263         VMSTATE_BOOL(rss_data.populate_hash, VirtIONet),
3264         VMSTATE_UINT32(rss_data.hash_types, VirtIONet),
3265         VMSTATE_UINT16(rss_data.indirections_len, VirtIONet),
3266         VMSTATE_UINT16(rss_data.default_queue, VirtIONet),
3267         VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet,
3268                             VIRTIO_NET_RSS_MAX_KEY_SIZE),
3269         VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet,
3270                                     rss_data.indirections_len, 0,
3271                                     vmstate_info_uint16, uint16_t),
3272         VMSTATE_END_OF_LIST()
3273     },
3274 };
3275 
3276 static const VMStateDescription vmstate_virtio_net_device = {
3277     .name = "virtio-net-device",
3278     .version_id = VIRTIO_NET_VM_VERSION,
3279     .minimum_version_id = VIRTIO_NET_VM_VERSION,
3280     .post_load = virtio_net_post_load_device,
3281     .fields = (VMStateField[]) {
3282         VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN),
3283         VMSTATE_STRUCT_POINTER(vqs, VirtIONet,
3284                                vmstate_virtio_net_queue_tx_waiting,
3285                                VirtIONetQueue),
3286         VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet),
3287         VMSTATE_UINT16(status, VirtIONet),
3288         VMSTATE_UINT8(promisc, VirtIONet),
3289         VMSTATE_UINT8(allmulti, VirtIONet),
3290         VMSTATE_UINT32(mac_table.in_use, VirtIONet),
3291 
3292         /* Guarded pair: If it fits we load it, else we throw it away
3293          * - can happen if source has a larger MAC table.; post-load
3294          *  sets flags in this case.
3295          */
3296         VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet,
3297                                 0, mac_table_fits, mac_table.in_use,
3298                                  ETH_ALEN),
3299         VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0,
3300                                      mac_table.in_use, ETH_ALEN),
3301 
3302         /* Note: This is an array of uint32's that's always been saved as a
3303          * buffer; hold onto your endiannesses; it's actually used as a bitmap
3304          * but based on the uint.
3305          */
3306         VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3),
3307         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3308                          vmstate_virtio_net_has_vnet),
3309         VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet),
3310         VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet),
3311         VMSTATE_UINT8(alluni, VirtIONet),
3312         VMSTATE_UINT8(nomulti, VirtIONet),
3313         VMSTATE_UINT8(nouni, VirtIONet),
3314         VMSTATE_UINT8(nobcast, VirtIONet),
3315         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3316                          vmstate_virtio_net_has_ufo),
3317         VMSTATE_SINGLE_TEST(max_queue_pairs, VirtIONet, max_queue_pairs_gt_1, 0,
3318                             vmstate_info_uint16_equal, uint16_t),
3319         VMSTATE_UINT16_TEST(curr_queue_pairs, VirtIONet, max_queue_pairs_gt_1),
3320         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3321                          vmstate_virtio_net_tx_waiting),
3322         VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet,
3323                             has_ctrl_guest_offloads),
3324         VMSTATE_END_OF_LIST()
3325    },
3326     .subsections = (const VMStateDescription * []) {
3327         &vmstate_virtio_net_rss,
3328         NULL
3329     }
3330 };
3331 
3332 static NetClientInfo net_virtio_info = {
3333     .type = NET_CLIENT_DRIVER_NIC,
3334     .size = sizeof(NICState),
3335     .can_receive = virtio_net_can_receive,
3336     .receive = virtio_net_receive,
3337     .link_status_changed = virtio_net_set_link_status,
3338     .query_rx_filter = virtio_net_query_rxfilter,
3339     .announce = virtio_net_announce,
3340 };
3341 
3342 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
3343 {
3344     VirtIONet *n = VIRTIO_NET(vdev);
3345     NetClientState *nc;
3346     assert(n->vhost_started);
3347     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ) && idx == 2) {
3348         /* Must guard against invalid features and bogus queue index
3349          * from being set by malicious guest, or penetrated through
3350          * buggy migration stream.
3351          */
3352         if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
3353             qemu_log_mask(LOG_GUEST_ERROR,
3354                           "%s: bogus vq index ignored\n", __func__);
3355             return false;
3356         }
3357         nc = qemu_get_subqueue(n->nic, n->max_queue_pairs);
3358     } else {
3359         nc = qemu_get_subqueue(n->nic, vq2q(idx));
3360     }
3361     /*
3362      * Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1
3363      * as the Marco of configure interrupt's IDX, If this driver does not
3364      * support, the function will return false
3365      */
3366 
3367     if (idx == VIRTIO_CONFIG_IRQ_IDX) {
3368         return vhost_net_config_pending(get_vhost_net(nc->peer));
3369     }
3370     return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
3371 }
3372 
3373 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
3374                                            bool mask)
3375 {
3376     VirtIONet *n = VIRTIO_NET(vdev);
3377     NetClientState *nc;
3378     assert(n->vhost_started);
3379     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ) && idx == 2) {
3380         /* Must guard against invalid features and bogus queue index
3381          * from being set by malicious guest, or penetrated through
3382          * buggy migration stream.
3383          */
3384         if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
3385             qemu_log_mask(LOG_GUEST_ERROR,
3386                           "%s: bogus vq index ignored\n", __func__);
3387             return;
3388         }
3389         nc = qemu_get_subqueue(n->nic, n->max_queue_pairs);
3390     } else {
3391         nc = qemu_get_subqueue(n->nic, vq2q(idx));
3392     }
3393     /*
3394      *Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1
3395      * as the Marco of configure interrupt's IDX, If this driver does not
3396      * support, the function will return
3397      */
3398 
3399     if (idx == VIRTIO_CONFIG_IRQ_IDX) {
3400         vhost_net_config_mask(get_vhost_net(nc->peer), vdev, mask);
3401         return;
3402     }
3403     vhost_net_virtqueue_mask(get_vhost_net(nc->peer), vdev, idx, mask);
3404 }
3405 
3406 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
3407 {
3408     virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
3409 
3410     n->config_size = virtio_get_config_size(&cfg_size_params, host_features);
3411 }
3412 
3413 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
3414                                    const char *type)
3415 {
3416     /*
3417      * The name can be NULL, the netclient name will be type.x.
3418      */
3419     assert(type != NULL);
3420 
3421     g_free(n->netclient_name);
3422     g_free(n->netclient_type);
3423     n->netclient_name = g_strdup(name);
3424     n->netclient_type = g_strdup(type);
3425 }
3426 
3427 static bool failover_unplug_primary(VirtIONet *n, DeviceState *dev)
3428 {
3429     HotplugHandler *hotplug_ctrl;
3430     PCIDevice *pci_dev;
3431     Error *err = NULL;
3432 
3433     hotplug_ctrl = qdev_get_hotplug_handler(dev);
3434     if (hotplug_ctrl) {
3435         pci_dev = PCI_DEVICE(dev);
3436         pci_dev->partially_hotplugged = true;
3437         hotplug_handler_unplug_request(hotplug_ctrl, dev, &err);
3438         if (err) {
3439             error_report_err(err);
3440             return false;
3441         }
3442     } else {
3443         return false;
3444     }
3445     return true;
3446 }
3447 
3448 static bool failover_replug_primary(VirtIONet *n, DeviceState *dev,
3449                                     Error **errp)
3450 {
3451     Error *err = NULL;
3452     HotplugHandler *hotplug_ctrl;
3453     PCIDevice *pdev = PCI_DEVICE(dev);
3454     BusState *primary_bus;
3455 
3456     if (!pdev->partially_hotplugged) {
3457         return true;
3458     }
3459     primary_bus = dev->parent_bus;
3460     if (!primary_bus) {
3461         error_setg(errp, "virtio_net: couldn't find primary bus");
3462         return false;
3463     }
3464     qdev_set_parent_bus(dev, primary_bus, &error_abort);
3465     qatomic_set(&n->failover_primary_hidden, false);
3466     hotplug_ctrl = qdev_get_hotplug_handler(dev);
3467     if (hotplug_ctrl) {
3468         hotplug_handler_pre_plug(hotplug_ctrl, dev, &err);
3469         if (err) {
3470             goto out;
3471         }
3472         hotplug_handler_plug(hotplug_ctrl, dev, &err);
3473     }
3474     pdev->partially_hotplugged = false;
3475 
3476 out:
3477     error_propagate(errp, err);
3478     return !err;
3479 }
3480 
3481 static void virtio_net_handle_migration_primary(VirtIONet *n, MigrationState *s)
3482 {
3483     bool should_be_hidden;
3484     Error *err = NULL;
3485     DeviceState *dev = failover_find_primary_device(n);
3486 
3487     if (!dev) {
3488         return;
3489     }
3490 
3491     should_be_hidden = qatomic_read(&n->failover_primary_hidden);
3492 
3493     if (migration_in_setup(s) && !should_be_hidden) {
3494         if (failover_unplug_primary(n, dev)) {
3495             vmstate_unregister(VMSTATE_IF(dev), qdev_get_vmsd(dev), dev);
3496             qapi_event_send_unplug_primary(dev->id);
3497             qatomic_set(&n->failover_primary_hidden, true);
3498         } else {
3499             warn_report("couldn't unplug primary device");
3500         }
3501     } else if (migration_has_failed(s)) {
3502         /* We already unplugged the device let's plug it back */
3503         if (!failover_replug_primary(n, dev, &err)) {
3504             if (err) {
3505                 error_report_err(err);
3506             }
3507         }
3508     }
3509 }
3510 
3511 static void virtio_net_migration_state_notifier(Notifier *notifier, void *data)
3512 {
3513     MigrationState *s = data;
3514     VirtIONet *n = container_of(notifier, VirtIONet, migration_state);
3515     virtio_net_handle_migration_primary(n, s);
3516 }
3517 
3518 static bool failover_hide_primary_device(DeviceListener *listener,
3519                                          const QDict *device_opts,
3520                                          bool from_json,
3521                                          Error **errp)
3522 {
3523     VirtIONet *n = container_of(listener, VirtIONet, primary_listener);
3524     const char *standby_id;
3525 
3526     if (!device_opts) {
3527         return false;
3528     }
3529 
3530     if (!qdict_haskey(device_opts, "failover_pair_id")) {
3531         return false;
3532     }
3533 
3534     if (!qdict_haskey(device_opts, "id")) {
3535         error_setg(errp, "Device with failover_pair_id needs to have id");
3536         return false;
3537     }
3538 
3539     standby_id = qdict_get_str(device_opts, "failover_pair_id");
3540     if (g_strcmp0(standby_id, n->netclient_name) != 0) {
3541         return false;
3542     }
3543 
3544     /*
3545      * The hide helper can be called several times for a given device.
3546      * Check there is only one primary for a virtio-net device but
3547      * don't duplicate the qdict several times if it's called for the same
3548      * device.
3549      */
3550     if (n->primary_opts) {
3551         const char *old, *new;
3552         /* devices with failover_pair_id always have an id */
3553         old = qdict_get_str(n->primary_opts, "id");
3554         new = qdict_get_str(device_opts, "id");
3555         if (strcmp(old, new) != 0) {
3556             error_setg(errp, "Cannot attach more than one primary device to "
3557                        "'%s': '%s' and '%s'", n->netclient_name, old, new);
3558             return false;
3559         }
3560     } else {
3561         n->primary_opts = qdict_clone_shallow(device_opts);
3562         n->primary_opts_from_json = from_json;
3563     }
3564 
3565     /* failover_primary_hidden is set during feature negotiation */
3566     return qatomic_read(&n->failover_primary_hidden);
3567 }
3568 
3569 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
3570 {
3571     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3572     VirtIONet *n = VIRTIO_NET(dev);
3573     NetClientState *nc;
3574     int i;
3575 
3576     if (n->net_conf.mtu) {
3577         n->host_features |= (1ULL << VIRTIO_NET_F_MTU);
3578     }
3579 
3580     if (n->net_conf.duplex_str) {
3581         if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) {
3582             n->net_conf.duplex = DUPLEX_HALF;
3583         } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) {
3584             n->net_conf.duplex = DUPLEX_FULL;
3585         } else {
3586             error_setg(errp, "'duplex' must be 'half' or 'full'");
3587             return;
3588         }
3589         n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3590     } else {
3591         n->net_conf.duplex = DUPLEX_UNKNOWN;
3592     }
3593 
3594     if (n->net_conf.speed < SPEED_UNKNOWN) {
3595         error_setg(errp, "'speed' must be between 0 and INT_MAX");
3596         return;
3597     }
3598     if (n->net_conf.speed >= 0) {
3599         n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3600     }
3601 
3602     if (n->failover) {
3603         n->primary_listener.hide_device = failover_hide_primary_device;
3604         qatomic_set(&n->failover_primary_hidden, true);
3605         device_listener_register(&n->primary_listener);
3606         n->migration_state.notify = virtio_net_migration_state_notifier;
3607         add_migration_state_change_notifier(&n->migration_state);
3608         n->host_features |= (1ULL << VIRTIO_NET_F_STANDBY);
3609     }
3610 
3611     virtio_net_set_config_size(n, n->host_features);
3612     virtio_init(vdev, VIRTIO_ID_NET, n->config_size);
3613 
3614     /*
3615      * We set a lower limit on RX queue size to what it always was.
3616      * Guests that want a smaller ring can always resize it without
3617      * help from us (using virtio 1 and up).
3618      */
3619     if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
3620         n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
3621         !is_power_of_2(n->net_conf.rx_queue_size)) {
3622         error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
3623                    "must be a power of 2 between %d and %d.",
3624                    n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
3625                    VIRTQUEUE_MAX_SIZE);
3626         virtio_cleanup(vdev);
3627         return;
3628     }
3629 
3630     if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE ||
3631         n->net_conf.tx_queue_size > VIRTQUEUE_MAX_SIZE ||
3632         !is_power_of_2(n->net_conf.tx_queue_size)) {
3633         error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), "
3634                    "must be a power of 2 between %d and %d",
3635                    n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE,
3636                    VIRTQUEUE_MAX_SIZE);
3637         virtio_cleanup(vdev);
3638         return;
3639     }
3640 
3641     n->max_ncs = MAX(n->nic_conf.peers.queues, 1);
3642 
3643     /*
3644      * Figure out the datapath queue pairs since the backend could
3645      * provide control queue via peers as well.
3646      */
3647     if (n->nic_conf.peers.queues) {
3648         for (i = 0; i < n->max_ncs; i++) {
3649             if (n->nic_conf.peers.ncs[i]->is_datapath) {
3650                 ++n->max_queue_pairs;
3651             }
3652         }
3653     }
3654     n->max_queue_pairs = MAX(n->max_queue_pairs, 1);
3655 
3656     if (n->max_queue_pairs * 2 + 1 > VIRTIO_QUEUE_MAX) {
3657         error_setg(errp, "Invalid number of queue pairs (= %" PRIu32 "), "
3658                    "must be a positive integer less than %d.",
3659                    n->max_queue_pairs, (VIRTIO_QUEUE_MAX - 1) / 2);
3660         virtio_cleanup(vdev);
3661         return;
3662     }
3663     n->vqs = g_new0(VirtIONetQueue, n->max_queue_pairs);
3664     n->curr_queue_pairs = 1;
3665     n->tx_timeout = n->net_conf.txtimer;
3666 
3667     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
3668                        && strcmp(n->net_conf.tx, "bh")) {
3669         warn_report("virtio-net: "
3670                     "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
3671                     n->net_conf.tx);
3672         error_printf("Defaulting to \"bh\"");
3673     }
3674 
3675     n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n),
3676                                     n->net_conf.tx_queue_size);
3677 
3678     for (i = 0; i < n->max_queue_pairs; i++) {
3679         virtio_net_add_queue(n, i);
3680     }
3681 
3682     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
3683     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
3684     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
3685     n->status = VIRTIO_NET_S_LINK_UP;
3686     qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
3687                               QEMU_CLOCK_VIRTUAL,
3688                               virtio_net_announce_timer, n);
3689     n->announce_timer.round = 0;
3690 
3691     if (n->netclient_type) {
3692         /*
3693          * Happen when virtio_net_set_netclient_name has been called.
3694          */
3695         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3696                               n->netclient_type, n->netclient_name, n);
3697     } else {
3698         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3699                               object_get_typename(OBJECT(dev)), dev->id, n);
3700     }
3701 
3702     for (i = 0; i < n->max_queue_pairs; i++) {
3703         n->nic->ncs[i].do_not_pad = true;
3704     }
3705 
3706     peer_test_vnet_hdr(n);
3707     if (peer_has_vnet_hdr(n)) {
3708         for (i = 0; i < n->max_queue_pairs; i++) {
3709             qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
3710         }
3711         n->host_hdr_len = sizeof(struct virtio_net_hdr);
3712     } else {
3713         n->host_hdr_len = 0;
3714     }
3715 
3716     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
3717 
3718     n->vqs[0].tx_waiting = 0;
3719     n->tx_burst = n->net_conf.txburst;
3720     virtio_net_set_mrg_rx_bufs(n, 0, 0, 0);
3721     n->promisc = 1; /* for compatibility */
3722 
3723     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
3724 
3725     n->vlans = g_malloc0(MAX_VLAN >> 3);
3726 
3727     nc = qemu_get_queue(n->nic);
3728     nc->rxfilter_notify_enabled = 1;
3729 
3730    if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
3731         struct virtio_net_config netcfg = {};
3732         memcpy(&netcfg.mac, &n->nic_conf.macaddr, ETH_ALEN);
3733         vhost_net_set_config(get_vhost_net(nc->peer),
3734             (uint8_t *)&netcfg, 0, ETH_ALEN, VHOST_SET_CONFIG_TYPE_MASTER);
3735     }
3736     QTAILQ_INIT(&n->rsc_chains);
3737     n->qdev = dev;
3738 
3739     net_rx_pkt_init(&n->rx_pkt);
3740 
3741     if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) {
3742         virtio_net_load_ebpf(n);
3743     }
3744 }
3745 
3746 static void virtio_net_device_unrealize(DeviceState *dev)
3747 {
3748     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3749     VirtIONet *n = VIRTIO_NET(dev);
3750     int i, max_queue_pairs;
3751 
3752     if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) {
3753         virtio_net_unload_ebpf(n);
3754     }
3755 
3756     /* This will stop vhost backend if appropriate. */
3757     virtio_net_set_status(vdev, 0);
3758 
3759     g_free(n->netclient_name);
3760     n->netclient_name = NULL;
3761     g_free(n->netclient_type);
3762     n->netclient_type = NULL;
3763 
3764     g_free(n->mac_table.macs);
3765     g_free(n->vlans);
3766 
3767     if (n->failover) {
3768         qobject_unref(n->primary_opts);
3769         device_listener_unregister(&n->primary_listener);
3770         remove_migration_state_change_notifier(&n->migration_state);
3771     } else {
3772         assert(n->primary_opts == NULL);
3773     }
3774 
3775     max_queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
3776     for (i = 0; i < max_queue_pairs; i++) {
3777         virtio_net_del_queue(n, i);
3778     }
3779     /* delete also control vq */
3780     virtio_del_queue(vdev, max_queue_pairs * 2);
3781     qemu_announce_timer_del(&n->announce_timer, false);
3782     g_free(n->vqs);
3783     qemu_del_nic(n->nic);
3784     virtio_net_rsc_cleanup(n);
3785     g_free(n->rss_data.indirections_table);
3786     net_rx_pkt_uninit(n->rx_pkt);
3787     virtio_cleanup(vdev);
3788 }
3789 
3790 static void virtio_net_instance_init(Object *obj)
3791 {
3792     VirtIONet *n = VIRTIO_NET(obj);
3793 
3794     /*
3795      * The default config_size is sizeof(struct virtio_net_config).
3796      * Can be overriden with virtio_net_set_config_size.
3797      */
3798     n->config_size = sizeof(struct virtio_net_config);
3799     device_add_bootindex_property(obj, &n->nic_conf.bootindex,
3800                                   "bootindex", "/ethernet-phy@0",
3801                                   DEVICE(n));
3802 
3803     ebpf_rss_init(&n->ebpf_rss);
3804 }
3805 
3806 static int virtio_net_pre_save(void *opaque)
3807 {
3808     VirtIONet *n = opaque;
3809 
3810     /* At this point, backend must be stopped, otherwise
3811      * it might keep writing to memory. */
3812     assert(!n->vhost_started);
3813 
3814     return 0;
3815 }
3816 
3817 static bool primary_unplug_pending(void *opaque)
3818 {
3819     DeviceState *dev = opaque;
3820     DeviceState *primary;
3821     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3822     VirtIONet *n = VIRTIO_NET(vdev);
3823 
3824     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
3825         return false;
3826     }
3827     primary = failover_find_primary_device(n);
3828     return primary ? primary->pending_deleted_event : false;
3829 }
3830 
3831 static bool dev_unplug_pending(void *opaque)
3832 {
3833     DeviceState *dev = opaque;
3834     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3835 
3836     return vdc->primary_unplug_pending(dev);
3837 }
3838 
3839 static struct vhost_dev *virtio_net_get_vhost(VirtIODevice *vdev)
3840 {
3841     VirtIONet *n = VIRTIO_NET(vdev);
3842     NetClientState *nc = qemu_get_queue(n->nic);
3843     struct vhost_net *net = get_vhost_net(nc->peer);
3844     return &net->dev;
3845 }
3846 
3847 static const VMStateDescription vmstate_virtio_net = {
3848     .name = "virtio-net",
3849     .minimum_version_id = VIRTIO_NET_VM_VERSION,
3850     .version_id = VIRTIO_NET_VM_VERSION,
3851     .fields = (VMStateField[]) {
3852         VMSTATE_VIRTIO_DEVICE,
3853         VMSTATE_END_OF_LIST()
3854     },
3855     .pre_save = virtio_net_pre_save,
3856     .dev_unplug_pending = dev_unplug_pending,
3857 };
3858 
3859 static Property virtio_net_properties[] = {
3860     DEFINE_PROP_BIT64("csum", VirtIONet, host_features,
3861                     VIRTIO_NET_F_CSUM, true),
3862     DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features,
3863                     VIRTIO_NET_F_GUEST_CSUM, true),
3864     DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
3865     DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features,
3866                     VIRTIO_NET_F_GUEST_TSO4, true),
3867     DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features,
3868                     VIRTIO_NET_F_GUEST_TSO6, true),
3869     DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features,
3870                     VIRTIO_NET_F_GUEST_ECN, true),
3871     DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features,
3872                     VIRTIO_NET_F_GUEST_UFO, true),
3873     DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features,
3874                     VIRTIO_NET_F_GUEST_ANNOUNCE, true),
3875     DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features,
3876                     VIRTIO_NET_F_HOST_TSO4, true),
3877     DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features,
3878                     VIRTIO_NET_F_HOST_TSO6, true),
3879     DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features,
3880                     VIRTIO_NET_F_HOST_ECN, true),
3881     DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features,
3882                     VIRTIO_NET_F_HOST_UFO, true),
3883     DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features,
3884                     VIRTIO_NET_F_MRG_RXBUF, true),
3885     DEFINE_PROP_BIT64("status", VirtIONet, host_features,
3886                     VIRTIO_NET_F_STATUS, true),
3887     DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features,
3888                     VIRTIO_NET_F_CTRL_VQ, true),
3889     DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features,
3890                     VIRTIO_NET_F_CTRL_RX, true),
3891     DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features,
3892                     VIRTIO_NET_F_CTRL_VLAN, true),
3893     DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features,
3894                     VIRTIO_NET_F_CTRL_RX_EXTRA, true),
3895     DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features,
3896                     VIRTIO_NET_F_CTRL_MAC_ADDR, true),
3897     DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features,
3898                     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
3899     DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
3900     DEFINE_PROP_BIT64("rss", VirtIONet, host_features,
3901                     VIRTIO_NET_F_RSS, false),
3902     DEFINE_PROP_BIT64("hash", VirtIONet, host_features,
3903                     VIRTIO_NET_F_HASH_REPORT, false),
3904     DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features,
3905                     VIRTIO_NET_F_RSC_EXT, false),
3906     DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout,
3907                        VIRTIO_NET_RSC_DEFAULT_INTERVAL),
3908     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
3909     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
3910                        TX_TIMER_INTERVAL),
3911     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
3912     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
3913     DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
3914                        VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
3915     DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size,
3916                        VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE),
3917     DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
3918     DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend,
3919                      true),
3920     DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN),
3921     DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str),
3922     DEFINE_PROP_BOOL("failover", VirtIONet, failover, false),
3923     DEFINE_PROP_END_OF_LIST(),
3924 };
3925 
3926 static void virtio_net_class_init(ObjectClass *klass, void *data)
3927 {
3928     DeviceClass *dc = DEVICE_CLASS(klass);
3929     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3930 
3931     device_class_set_props(dc, virtio_net_properties);
3932     dc->vmsd = &vmstate_virtio_net;
3933     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
3934     vdc->realize = virtio_net_device_realize;
3935     vdc->unrealize = virtio_net_device_unrealize;
3936     vdc->get_config = virtio_net_get_config;
3937     vdc->set_config = virtio_net_set_config;
3938     vdc->get_features = virtio_net_get_features;
3939     vdc->set_features = virtio_net_set_features;
3940     vdc->bad_features = virtio_net_bad_features;
3941     vdc->reset = virtio_net_reset;
3942     vdc->queue_reset = virtio_net_queue_reset;
3943     vdc->queue_enable = virtio_net_queue_enable;
3944     vdc->set_status = virtio_net_set_status;
3945     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
3946     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
3947     vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
3948     vdc->post_load = virtio_net_post_load_virtio;
3949     vdc->vmsd = &vmstate_virtio_net_device;
3950     vdc->primary_unplug_pending = primary_unplug_pending;
3951     vdc->get_vhost = virtio_net_get_vhost;
3952 }
3953 
3954 static const TypeInfo virtio_net_info = {
3955     .name = TYPE_VIRTIO_NET,
3956     .parent = TYPE_VIRTIO_DEVICE,
3957     .instance_size = sizeof(VirtIONet),
3958     .instance_init = virtio_net_instance_init,
3959     .class_init = virtio_net_class_init,
3960 };
3961 
3962 static void virtio_register_types(void)
3963 {
3964     type_register_static(&virtio_net_info);
3965 }
3966 
3967 type_init(virtio_register_types)
3968