xref: /openbmc/qemu/hw/net/virtio-net.c (revision ba379542)
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 "monitor/monitor.h"
46 #include "hw/pci/pci_device.h"
47 #include "net_rx_pkt.h"
48 #include "hw/virtio/vhost.h"
49 #include "sysemu/qtest.h"
50 
51 #define VIRTIO_NET_VM_VERSION    11
52 
53 /* previously fixed value */
54 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
55 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
56 
57 /* for now, only allow larger queue_pairs; with virtio-1, guest can downsize */
58 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
59 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
60 
61 #define VIRTIO_NET_IP4_ADDR_SIZE   8        /* ipv4 saddr + daddr */
62 
63 #define VIRTIO_NET_TCP_FLAG         0x3F
64 #define VIRTIO_NET_TCP_HDR_LENGTH   0xF000
65 
66 /* IPv4 max payload, 16 bits in the header */
67 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
68 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
69 
70 /* header length value in ip header without option */
71 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
72 
73 #define VIRTIO_NET_IP6_ADDR_SIZE   32      /* ipv6 saddr + daddr */
74 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
75 
76 /* Purge coalesced packets timer interval, This value affects the performance
77    a lot, and should be tuned carefully, '300000'(300us) is the recommended
78    value to pass the WHQL test, '50000' can gain 2x netperf throughput with
79    tso/gso/gro 'off'. */
80 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
81 
82 #define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \
83                                          VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \
84                                          VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \
85                                          VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \
86                                          VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \
87                                          VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \
88                                          VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \
89                                          VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \
90                                          VIRTIO_NET_RSS_HASH_TYPE_UDP_EX)
91 
92 static const VirtIOFeature feature_sizes[] = {
93     {.flags = 1ULL << VIRTIO_NET_F_MAC,
94      .end = endof(struct virtio_net_config, mac)},
95     {.flags = 1ULL << VIRTIO_NET_F_STATUS,
96      .end = endof(struct virtio_net_config, status)},
97     {.flags = 1ULL << VIRTIO_NET_F_MQ,
98      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
99     {.flags = 1ULL << VIRTIO_NET_F_MTU,
100      .end = endof(struct virtio_net_config, mtu)},
101     {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX,
102      .end = endof(struct virtio_net_config, duplex)},
103     {.flags = (1ULL << VIRTIO_NET_F_RSS) | (1ULL << VIRTIO_NET_F_HASH_REPORT),
104      .end = endof(struct virtio_net_config, supported_hash_types)},
105     {}
106 };
107 
108 static const VirtIOConfigSizeParams cfg_size_params = {
109     .min_size = endof(struct virtio_net_config, mac),
110     .max_size = sizeof(struct virtio_net_config),
111     .feature_sizes = feature_sizes
112 };
113 
114 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
115 {
116     VirtIONet *n = qemu_get_nic_opaque(nc);
117 
118     return &n->vqs[nc->queue_index];
119 }
120 
121 static int vq2q(int queue_index)
122 {
123     return queue_index / 2;
124 }
125 
126 static void flush_or_purge_queued_packets(NetClientState *nc)
127 {
128     if (!nc->peer) {
129         return;
130     }
131 
132     qemu_flush_or_purge_queued_packets(nc->peer, true);
133     assert(!virtio_net_get_subqueue(nc)->async_tx.elem);
134 }
135 
136 /* TODO
137  * - we could suppress RX interrupt if we were so inclined.
138  */
139 
140 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
141 {
142     VirtIONet *n = VIRTIO_NET(vdev);
143     struct virtio_net_config netcfg;
144     NetClientState *nc = qemu_get_queue(n->nic);
145     static const MACAddr zero = { .a = { 0, 0, 0, 0, 0, 0 } };
146 
147     int ret = 0;
148     memset(&netcfg, 0 , sizeof(struct virtio_net_config));
149     virtio_stw_p(vdev, &netcfg.status, n->status);
150     virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queue_pairs);
151     virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
152     memcpy(netcfg.mac, n->mac, ETH_ALEN);
153     virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed);
154     netcfg.duplex = n->net_conf.duplex;
155     netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE;
156     virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length,
157                  virtio_host_has_feature(vdev, VIRTIO_NET_F_RSS) ?
158                  VIRTIO_NET_RSS_MAX_TABLE_LEN : 1);
159     virtio_stl_p(vdev, &netcfg.supported_hash_types,
160                  VIRTIO_NET_RSS_SUPPORTED_HASHES);
161     memcpy(config, &netcfg, n->config_size);
162 
163     /*
164      * Is this VDPA? No peer means not VDPA: there's no way to
165      * disconnect/reconnect a VDPA peer.
166      */
167     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
168         ret = vhost_net_get_config(get_vhost_net(nc->peer), (uint8_t *)&netcfg,
169                                    n->config_size);
170         if (ret == -1) {
171             return;
172         }
173 
174         /*
175          * Some NIC/kernel combinations present 0 as the mac address.  As that
176          * is not a legal address, try to proceed with the address from the
177          * QEMU command line in the hope that the address has been configured
178          * correctly elsewhere - just not reported by the device.
179          */
180         if (memcmp(&netcfg.mac, &zero, sizeof(zero)) == 0) {
181             info_report("Zero hardware mac address detected. Ignoring.");
182             memcpy(netcfg.mac, n->mac, ETH_ALEN);
183         }
184 
185         netcfg.status |= virtio_tswap16(vdev,
186                                         n->status & VIRTIO_NET_S_ANNOUNCE);
187         memcpy(config, &netcfg, n->config_size);
188     }
189 }
190 
191 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
192 {
193     VirtIONet *n = VIRTIO_NET(vdev);
194     struct virtio_net_config netcfg = {};
195     NetClientState *nc = qemu_get_queue(n->nic);
196 
197     memcpy(&netcfg, config, n->config_size);
198 
199     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
200         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
201         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
202         memcpy(n->mac, netcfg.mac, ETH_ALEN);
203         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
204     }
205 
206     /*
207      * Is this VDPA? No peer means not VDPA: there's no way to
208      * disconnect/reconnect a VDPA peer.
209      */
210     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
211         vhost_net_set_config(get_vhost_net(nc->peer),
212                              (uint8_t *)&netcfg, 0, n->config_size,
213                              VHOST_SET_CONFIG_TYPE_FRONTEND);
214       }
215 }
216 
217 static bool virtio_net_started(VirtIONet *n, uint8_t status)
218 {
219     VirtIODevice *vdev = VIRTIO_DEVICE(n);
220     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
221         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
222 }
223 
224 static void virtio_net_announce_notify(VirtIONet *net)
225 {
226     VirtIODevice *vdev = VIRTIO_DEVICE(net);
227     trace_virtio_net_announce_notify();
228 
229     net->status |= VIRTIO_NET_S_ANNOUNCE;
230     virtio_notify_config(vdev);
231 }
232 
233 static void virtio_net_announce_timer(void *opaque)
234 {
235     VirtIONet *n = opaque;
236     trace_virtio_net_announce_timer(n->announce_timer.round);
237 
238     n->announce_timer.round--;
239     virtio_net_announce_notify(n);
240 }
241 
242 static void virtio_net_announce(NetClientState *nc)
243 {
244     VirtIONet *n = qemu_get_nic_opaque(nc);
245     VirtIODevice *vdev = VIRTIO_DEVICE(n);
246 
247     /*
248      * Make sure the virtio migration announcement timer isn't running
249      * If it is, let it trigger announcement so that we do not cause
250      * confusion.
251      */
252     if (n->announce_timer.round) {
253         return;
254     }
255 
256     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
257         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
258             virtio_net_announce_notify(n);
259     }
260 }
261 
262 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
263 {
264     VirtIODevice *vdev = VIRTIO_DEVICE(n);
265     NetClientState *nc = qemu_get_queue(n->nic);
266     int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
267     int cvq = virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) ?
268               n->max_ncs - n->max_queue_pairs : 0;
269 
270     if (!get_vhost_net(nc->peer)) {
271         return;
272     }
273 
274     if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
275         !!n->vhost_started) {
276         return;
277     }
278     if (!n->vhost_started) {
279         int r, i;
280 
281         if (n->needs_vnet_hdr_swap) {
282             error_report("backend does not support %s vnet headers; "
283                          "falling back on userspace virtio",
284                          virtio_is_big_endian(vdev) ? "BE" : "LE");
285             return;
286         }
287 
288         /* Any packets outstanding? Purge them to avoid touching rings
289          * when vhost is running.
290          */
291         for (i = 0;  i < queue_pairs; i++) {
292             NetClientState *qnc = qemu_get_subqueue(n->nic, i);
293 
294             /* Purge both directions: TX and RX. */
295             qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
296             qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
297         }
298 
299         if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
300             r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
301             if (r < 0) {
302                 error_report("%uBytes MTU not supported by the backend",
303                              n->net_conf.mtu);
304 
305                 return;
306             }
307         }
308 
309         n->vhost_started = 1;
310         r = vhost_net_start(vdev, n->nic->ncs, queue_pairs, cvq);
311         if (r < 0) {
312             error_report("unable to start vhost net: %d: "
313                          "falling back on userspace virtio", -r);
314             n->vhost_started = 0;
315         }
316     } else {
317         vhost_net_stop(vdev, n->nic->ncs, queue_pairs, cvq);
318         n->vhost_started = 0;
319     }
320 }
321 
322 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
323                                           NetClientState *peer,
324                                           bool enable)
325 {
326     if (virtio_is_big_endian(vdev)) {
327         return qemu_set_vnet_be(peer, enable);
328     } else {
329         return qemu_set_vnet_le(peer, enable);
330     }
331 }
332 
333 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
334                                        int queue_pairs, bool enable)
335 {
336     int i;
337 
338     for (i = 0; i < queue_pairs; i++) {
339         if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
340             enable) {
341             while (--i >= 0) {
342                 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
343             }
344 
345             return true;
346         }
347     }
348 
349     return false;
350 }
351 
352 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
353 {
354     VirtIODevice *vdev = VIRTIO_DEVICE(n);
355     int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
356 
357     if (virtio_net_started(n, status)) {
358         /* Before using the device, we tell the network backend about the
359          * endianness to use when parsing vnet headers. If the backend
360          * can't do it, we fallback onto fixing the headers in the core
361          * virtio-net code.
362          */
363         n->needs_vnet_hdr_swap = n->has_vnet_hdr &&
364                                  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 peer_test_vnet_hdr(VirtIONet *n)
604 {
605     NetClientState *nc = qemu_get_queue(n->nic);
606     if (!nc->peer) {
607         return;
608     }
609 
610     n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
611 }
612 
613 static int peer_has_vnet_hdr(VirtIONet *n)
614 {
615     return n->has_vnet_hdr;
616 }
617 
618 static int peer_has_ufo(VirtIONet *n)
619 {
620     if (!peer_has_vnet_hdr(n))
621         return 0;
622 
623     n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
624 
625     return n->has_ufo;
626 }
627 
628 static int peer_has_uso(VirtIONet *n)
629 {
630     if (!peer_has_vnet_hdr(n)) {
631         return 0;
632     }
633 
634     return qemu_has_uso(qemu_get_queue(n->nic)->peer);
635 }
636 
637 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
638                                        int version_1, int hash_report)
639 {
640     int i;
641     NetClientState *nc;
642 
643     n->mergeable_rx_bufs = mergeable_rx_bufs;
644 
645     if (version_1) {
646         n->guest_hdr_len = hash_report ?
647             sizeof(struct virtio_net_hdr_v1_hash) :
648             sizeof(struct virtio_net_hdr_mrg_rxbuf);
649         n->rss_data.populate_hash = !!hash_report;
650     } else {
651         n->guest_hdr_len = n->mergeable_rx_bufs ?
652             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
653             sizeof(struct virtio_net_hdr);
654         n->rss_data.populate_hash = false;
655     }
656 
657     for (i = 0; i < n->max_queue_pairs; i++) {
658         nc = qemu_get_subqueue(n->nic, i);
659 
660         if (peer_has_vnet_hdr(n) &&
661             qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
662             qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
663             n->host_hdr_len = n->guest_hdr_len;
664         }
665     }
666 }
667 
668 static int virtio_net_max_tx_queue_size(VirtIONet *n)
669 {
670     NetClientState *peer = n->nic_conf.peers.ncs[0];
671 
672     /*
673      * Backends other than vhost-user or vhost-vdpa don't support max queue
674      * size.
675      */
676     if (!peer) {
677         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
678     }
679 
680     switch(peer->info->type) {
681     case NET_CLIENT_DRIVER_VHOST_USER:
682     case NET_CLIENT_DRIVER_VHOST_VDPA:
683         return VIRTQUEUE_MAX_SIZE;
684     default:
685         return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
686     };
687 }
688 
689 static int peer_attach(VirtIONet *n, int index)
690 {
691     NetClientState *nc = qemu_get_subqueue(n->nic, index);
692 
693     if (!nc->peer) {
694         return 0;
695     }
696 
697     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
698         vhost_set_vring_enable(nc->peer, 1);
699     }
700 
701     if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
702         return 0;
703     }
704 
705     if (n->max_queue_pairs == 1) {
706         return 0;
707     }
708 
709     return tap_enable(nc->peer);
710 }
711 
712 static int peer_detach(VirtIONet *n, int index)
713 {
714     NetClientState *nc = qemu_get_subqueue(n->nic, index);
715 
716     if (!nc->peer) {
717         return 0;
718     }
719 
720     if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
721         vhost_set_vring_enable(nc->peer, 0);
722     }
723 
724     if (nc->peer->info->type !=  NET_CLIENT_DRIVER_TAP) {
725         return 0;
726     }
727 
728     return tap_disable(nc->peer);
729 }
730 
731 static void virtio_net_set_queue_pairs(VirtIONet *n)
732 {
733     int i;
734     int r;
735 
736     if (n->nic->peer_deleted) {
737         return;
738     }
739 
740     for (i = 0; i < n->max_queue_pairs; i++) {
741         if (i < n->curr_queue_pairs) {
742             r = peer_attach(n, i);
743             assert(!r);
744         } else {
745             r = peer_detach(n, i);
746             assert(!r);
747         }
748     }
749 }
750 
751 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
752 
753 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
754                                         Error **errp)
755 {
756     VirtIONet *n = VIRTIO_NET(vdev);
757     NetClientState *nc = qemu_get_queue(n->nic);
758 
759     /* Firstly sync all virtio-net possible supported features */
760     features |= n->host_features;
761 
762     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
763 
764     if (!peer_has_vnet_hdr(n)) {
765         virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
766         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
767         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
768         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
769 
770         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
771         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
772         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
773         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
774 
775         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_USO);
776         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO4);
777         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO6);
778 
779         virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT);
780     }
781 
782     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
783         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
784         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
785     }
786 
787     if (!peer_has_uso(n)) {
788         virtio_clear_feature(&features, VIRTIO_NET_F_HOST_USO);
789         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO4);
790         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO6);
791     }
792 
793     if (!get_vhost_net(nc->peer)) {
794         return features;
795     }
796 
797     if (!ebpf_rss_is_loaded(&n->ebpf_rss)) {
798         virtio_clear_feature(&features, VIRTIO_NET_F_RSS);
799     }
800     features = vhost_net_get_features(get_vhost_net(nc->peer), features);
801     vdev->backend_features = features;
802 
803     if (n->mtu_bypass_backend &&
804             (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) {
805         features |= (1ULL << VIRTIO_NET_F_MTU);
806     }
807 
808     /*
809      * Since GUEST_ANNOUNCE is emulated the feature bit could be set without
810      * enabled. This happens in the vDPA case.
811      *
812      * Make sure the feature set is not incoherent, as the driver could refuse
813      * to start.
814      *
815      * TODO: QEMU is able to emulate a CVQ just for guest_announce purposes,
816      * helping guest to notify the new location with vDPA devices that does not
817      * support it.
818      */
819     if (!virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_CTRL_VQ)) {
820         virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ANNOUNCE);
821     }
822 
823     return features;
824 }
825 
826 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
827 {
828     uint64_t features = 0;
829 
830     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
831      * but also these: */
832     virtio_add_feature(&features, VIRTIO_NET_F_MAC);
833     virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
834     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
835     virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
836     virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
837 
838     return features;
839 }
840 
841 static void virtio_net_apply_guest_offloads(VirtIONet *n)
842 {
843     qemu_set_offload(qemu_get_queue(n->nic)->peer,
844             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
845             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
846             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
847             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
848             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)),
849             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_USO4)),
850             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_USO6)));
851 }
852 
853 static uint64_t virtio_net_guest_offloads_by_features(uint64_t features)
854 {
855     static const uint64_t guest_offloads_mask =
856         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
857         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
858         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
859         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
860         (1ULL << VIRTIO_NET_F_GUEST_UFO)  |
861         (1ULL << VIRTIO_NET_F_GUEST_USO4) |
862         (1ULL << VIRTIO_NET_F_GUEST_USO6);
863 
864     return guest_offloads_mask & features;
865 }
866 
867 uint64_t virtio_net_supported_guest_offloads(const VirtIONet *n)
868 {
869     VirtIODevice *vdev = VIRTIO_DEVICE(n);
870     return virtio_net_guest_offloads_by_features(vdev->guest_features);
871 }
872 
873 typedef struct {
874     VirtIONet *n;
875     DeviceState *dev;
876 } FailoverDevice;
877 
878 /**
879  * Set the failover primary device
880  *
881  * @opaque: FailoverId to setup
882  * @opts: opts for device we are handling
883  * @errp: returns an error if this function fails
884  */
885 static int failover_set_primary(DeviceState *dev, void *opaque)
886 {
887     FailoverDevice *fdev = opaque;
888     PCIDevice *pci_dev = (PCIDevice *)
889         object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE);
890 
891     if (!pci_dev) {
892         return 0;
893     }
894 
895     if (!g_strcmp0(pci_dev->failover_pair_id, fdev->n->netclient_name)) {
896         fdev->dev = dev;
897         return 1;
898     }
899 
900     return 0;
901 }
902 
903 /**
904  * Find the primary device for this failover virtio-net
905  *
906  * @n: VirtIONet device
907  * @errp: returns an error if this function fails
908  */
909 static DeviceState *failover_find_primary_device(VirtIONet *n)
910 {
911     FailoverDevice fdev = {
912         .n = n,
913     };
914 
915     qbus_walk_children(sysbus_get_default(), failover_set_primary, NULL,
916                        NULL, NULL, &fdev);
917     return fdev.dev;
918 }
919 
920 static void failover_add_primary(VirtIONet *n, Error **errp)
921 {
922     Error *err = NULL;
923     DeviceState *dev = failover_find_primary_device(n);
924 
925     if (dev) {
926         return;
927     }
928 
929     if (!n->primary_opts) {
930         error_setg(errp, "Primary device not found");
931         error_append_hint(errp, "Virtio-net failover will not work. Make "
932                           "sure primary device has parameter"
933                           " failover_pair_id=%s\n", n->netclient_name);
934         return;
935     }
936 
937     dev = qdev_device_add_from_qdict(n->primary_opts,
938                                      n->primary_opts_from_json,
939                                      &err);
940     if (err) {
941         qobject_unref(n->primary_opts);
942         n->primary_opts = NULL;
943     } else {
944         object_unref(OBJECT(dev));
945     }
946     error_propagate(errp, err);
947 }
948 
949 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
950 {
951     VirtIONet *n = VIRTIO_NET(vdev);
952     Error *err = NULL;
953     int i;
954 
955     if (n->mtu_bypass_backend &&
956             !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) {
957         features &= ~(1ULL << VIRTIO_NET_F_MTU);
958     }
959 
960     virtio_net_set_multiqueue(n,
961                               virtio_has_feature(features, VIRTIO_NET_F_RSS) ||
962                               virtio_has_feature(features, VIRTIO_NET_F_MQ));
963 
964     virtio_net_set_mrg_rx_bufs(n,
965                                virtio_has_feature(features,
966                                                   VIRTIO_NET_F_MRG_RXBUF),
967                                virtio_has_feature(features,
968                                                   VIRTIO_F_VERSION_1),
969                                virtio_has_feature(features,
970                                                   VIRTIO_NET_F_HASH_REPORT));
971 
972     n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
973         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4);
974     n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
975         virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6);
976     n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS);
977 
978     if (n->has_vnet_hdr) {
979         n->curr_guest_offloads =
980             virtio_net_guest_offloads_by_features(features);
981         virtio_net_apply_guest_offloads(n);
982     }
983 
984     for (i = 0;  i < n->max_queue_pairs; i++) {
985         NetClientState *nc = qemu_get_subqueue(n->nic, i);
986 
987         if (!get_vhost_net(nc->peer)) {
988             continue;
989         }
990         vhost_net_ack_features(get_vhost_net(nc->peer), features);
991 
992         /*
993          * keep acked_features in NetVhostUserState up-to-date so it
994          * can't miss any features configured by guest virtio driver.
995          */
996         vhost_net_save_acked_features(nc->peer);
997     }
998 
999     if (!virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
1000         memset(n->vlans, 0xff, MAX_VLAN >> 3);
1001     }
1002 
1003     if (virtio_has_feature(features, VIRTIO_NET_F_STANDBY)) {
1004         qapi_event_send_failover_negotiated(n->netclient_name);
1005         qatomic_set(&n->failover_primary_hidden, false);
1006         failover_add_primary(n, &err);
1007         if (err) {
1008             if (!qtest_enabled()) {
1009                 warn_report_err(err);
1010             } else {
1011                 error_free(err);
1012             }
1013         }
1014     }
1015 }
1016 
1017 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
1018                                      struct iovec *iov, unsigned int iov_cnt)
1019 {
1020     uint8_t on;
1021     size_t s;
1022     NetClientState *nc = qemu_get_queue(n->nic);
1023 
1024     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
1025     if (s != sizeof(on)) {
1026         return VIRTIO_NET_ERR;
1027     }
1028 
1029     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
1030         n->promisc = on;
1031     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
1032         n->allmulti = on;
1033     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
1034         n->alluni = on;
1035     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
1036         n->nomulti = on;
1037     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
1038         n->nouni = on;
1039     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
1040         n->nobcast = on;
1041     } else {
1042         return VIRTIO_NET_ERR;
1043     }
1044 
1045     rxfilter_notify(nc);
1046 
1047     return VIRTIO_NET_OK;
1048 }
1049 
1050 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
1051                                      struct iovec *iov, unsigned int iov_cnt)
1052 {
1053     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1054     uint64_t offloads;
1055     size_t s;
1056 
1057     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1058         return VIRTIO_NET_ERR;
1059     }
1060 
1061     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
1062     if (s != sizeof(offloads)) {
1063         return VIRTIO_NET_ERR;
1064     }
1065 
1066     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
1067         uint64_t supported_offloads;
1068 
1069         offloads = virtio_ldq_p(vdev, &offloads);
1070 
1071         if (!n->has_vnet_hdr) {
1072             return VIRTIO_NET_ERR;
1073         }
1074 
1075         n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
1076             virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4);
1077         n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
1078             virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6);
1079         virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT);
1080 
1081         supported_offloads = virtio_net_supported_guest_offloads(n);
1082         if (offloads & ~supported_offloads) {
1083             return VIRTIO_NET_ERR;
1084         }
1085 
1086         n->curr_guest_offloads = offloads;
1087         virtio_net_apply_guest_offloads(n);
1088 
1089         return VIRTIO_NET_OK;
1090     } else {
1091         return VIRTIO_NET_ERR;
1092     }
1093 }
1094 
1095 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
1096                                  struct iovec *iov, unsigned int iov_cnt)
1097 {
1098     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1099     struct virtio_net_ctrl_mac mac_data;
1100     size_t s;
1101     NetClientState *nc = qemu_get_queue(n->nic);
1102 
1103     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
1104         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
1105             return VIRTIO_NET_ERR;
1106         }
1107         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
1108         assert(s == sizeof(n->mac));
1109         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
1110         rxfilter_notify(nc);
1111 
1112         return VIRTIO_NET_OK;
1113     }
1114 
1115     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
1116         return VIRTIO_NET_ERR;
1117     }
1118 
1119     int in_use = 0;
1120     int first_multi = 0;
1121     uint8_t uni_overflow = 0;
1122     uint8_t multi_overflow = 0;
1123     uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1124 
1125     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1126                    sizeof(mac_data.entries));
1127     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1128     if (s != sizeof(mac_data.entries)) {
1129         goto error;
1130     }
1131     iov_discard_front(&iov, &iov_cnt, s);
1132 
1133     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
1134         goto error;
1135     }
1136 
1137     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
1138         s = iov_to_buf(iov, iov_cnt, 0, macs,
1139                        mac_data.entries * ETH_ALEN);
1140         if (s != mac_data.entries * ETH_ALEN) {
1141             goto error;
1142         }
1143         in_use += mac_data.entries;
1144     } else {
1145         uni_overflow = 1;
1146     }
1147 
1148     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
1149 
1150     first_multi = in_use;
1151 
1152     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1153                    sizeof(mac_data.entries));
1154     mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1155     if (s != sizeof(mac_data.entries)) {
1156         goto error;
1157     }
1158 
1159     iov_discard_front(&iov, &iov_cnt, s);
1160 
1161     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
1162         goto error;
1163     }
1164 
1165     if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
1166         s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
1167                        mac_data.entries * ETH_ALEN);
1168         if (s != mac_data.entries * ETH_ALEN) {
1169             goto error;
1170         }
1171         in_use += mac_data.entries;
1172     } else {
1173         multi_overflow = 1;
1174     }
1175 
1176     n->mac_table.in_use = in_use;
1177     n->mac_table.first_multi = first_multi;
1178     n->mac_table.uni_overflow = uni_overflow;
1179     n->mac_table.multi_overflow = multi_overflow;
1180     memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
1181     g_free(macs);
1182     rxfilter_notify(nc);
1183 
1184     return VIRTIO_NET_OK;
1185 
1186 error:
1187     g_free(macs);
1188     return VIRTIO_NET_ERR;
1189 }
1190 
1191 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
1192                                         struct iovec *iov, unsigned int iov_cnt)
1193 {
1194     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1195     uint16_t vid;
1196     size_t s;
1197     NetClientState *nc = qemu_get_queue(n->nic);
1198 
1199     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
1200     vid = virtio_lduw_p(vdev, &vid);
1201     if (s != sizeof(vid)) {
1202         return VIRTIO_NET_ERR;
1203     }
1204 
1205     if (vid >= MAX_VLAN)
1206         return VIRTIO_NET_ERR;
1207 
1208     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
1209         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
1210     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
1211         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
1212     else
1213         return VIRTIO_NET_ERR;
1214 
1215     rxfilter_notify(nc);
1216 
1217     return VIRTIO_NET_OK;
1218 }
1219 
1220 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
1221                                       struct iovec *iov, unsigned int iov_cnt)
1222 {
1223     trace_virtio_net_handle_announce(n->announce_timer.round);
1224     if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
1225         n->status & VIRTIO_NET_S_ANNOUNCE) {
1226         n->status &= ~VIRTIO_NET_S_ANNOUNCE;
1227         if (n->announce_timer.round) {
1228             qemu_announce_timer_step(&n->announce_timer);
1229         }
1230         return VIRTIO_NET_OK;
1231     } else {
1232         return VIRTIO_NET_ERR;
1233     }
1234 }
1235 
1236 static bool virtio_net_attach_ebpf_to_backend(NICState *nic, int prog_fd)
1237 {
1238     NetClientState *nc = qemu_get_peer(qemu_get_queue(nic), 0);
1239     if (nc == NULL || nc->info->set_steering_ebpf == NULL) {
1240         return false;
1241     }
1242 
1243     return nc->info->set_steering_ebpf(nc, prog_fd);
1244 }
1245 
1246 static void rss_data_to_rss_config(struct VirtioNetRssData *data,
1247                                    struct EBPFRSSConfig *config)
1248 {
1249     config->redirect = data->redirect;
1250     config->populate_hash = data->populate_hash;
1251     config->hash_types = data->hash_types;
1252     config->indirections_len = data->indirections_len;
1253     config->default_queue = data->default_queue;
1254 }
1255 
1256 static bool virtio_net_attach_epbf_rss(VirtIONet *n)
1257 {
1258     struct EBPFRSSConfig config = {};
1259 
1260     if (!ebpf_rss_is_loaded(&n->ebpf_rss)) {
1261         return false;
1262     }
1263 
1264     rss_data_to_rss_config(&n->rss_data, &config);
1265 
1266     if (!ebpf_rss_set_all(&n->ebpf_rss, &config,
1267                           n->rss_data.indirections_table, n->rss_data.key)) {
1268         return false;
1269     }
1270 
1271     if (!virtio_net_attach_ebpf_to_backend(n->nic, n->ebpf_rss.program_fd)) {
1272         return false;
1273     }
1274 
1275     return true;
1276 }
1277 
1278 static void virtio_net_detach_epbf_rss(VirtIONet *n)
1279 {
1280     virtio_net_attach_ebpf_to_backend(n->nic, -1);
1281 }
1282 
1283 static void virtio_net_commit_rss_config(VirtIONet *n)
1284 {
1285     if (n->rss_data.enabled) {
1286         n->rss_data.enabled_software_rss = n->rss_data.populate_hash;
1287         if (n->rss_data.populate_hash) {
1288             virtio_net_detach_epbf_rss(n);
1289         } else if (!virtio_net_attach_epbf_rss(n)) {
1290             if (get_vhost_net(qemu_get_queue(n->nic)->peer)) {
1291                 warn_report("Can't load eBPF RSS for vhost");
1292             } else {
1293                 warn_report("Can't load eBPF RSS - fallback to software RSS");
1294                 n->rss_data.enabled_software_rss = true;
1295             }
1296         }
1297 
1298         trace_virtio_net_rss_enable(n->rss_data.hash_types,
1299                                     n->rss_data.indirections_len,
1300                                     sizeof(n->rss_data.key));
1301     } else {
1302         virtio_net_detach_epbf_rss(n);
1303         trace_virtio_net_rss_disable();
1304     }
1305 }
1306 
1307 static void virtio_net_disable_rss(VirtIONet *n)
1308 {
1309     if (!n->rss_data.enabled) {
1310         return;
1311     }
1312 
1313     n->rss_data.enabled = false;
1314     virtio_net_commit_rss_config(n);
1315 }
1316 
1317 static bool virtio_net_load_ebpf_fds(VirtIONet *n)
1318 {
1319     int fds[EBPF_RSS_MAX_FDS] = { [0 ... EBPF_RSS_MAX_FDS - 1] = -1};
1320     int ret = true;
1321     int i = 0;
1322 
1323     if (n->nr_ebpf_rss_fds != EBPF_RSS_MAX_FDS) {
1324         warn_report("Expected %d file descriptors but got %d",
1325                     EBPF_RSS_MAX_FDS, n->nr_ebpf_rss_fds);
1326        return false;
1327    }
1328 
1329     for (i = 0; i < n->nr_ebpf_rss_fds; i++) {
1330         fds[i] = monitor_fd_param(monitor_cur(), n->ebpf_rss_fds[i],
1331                                   &error_warn);
1332         if (fds[i] < 0) {
1333             ret = false;
1334             goto exit;
1335         }
1336     }
1337 
1338     ret = ebpf_rss_load_fds(&n->ebpf_rss, fds[0], fds[1], fds[2], fds[3]);
1339 
1340 exit:
1341     if (!ret) {
1342         for (i = 0; i < n->nr_ebpf_rss_fds && fds[i] != -1; i++) {
1343             close(fds[i]);
1344         }
1345     }
1346 
1347     return ret;
1348 }
1349 
1350 static bool virtio_net_load_ebpf(VirtIONet *n)
1351 {
1352     bool ret = false;
1353 
1354     if (virtio_net_attach_ebpf_to_backend(n->nic, -1)) {
1355         if (!(n->ebpf_rss_fds && virtio_net_load_ebpf_fds(n))) {
1356             ret = ebpf_rss_load(&n->ebpf_rss);
1357         }
1358     }
1359 
1360     return ret;
1361 }
1362 
1363 static void virtio_net_unload_ebpf(VirtIONet *n)
1364 {
1365     virtio_net_attach_ebpf_to_backend(n->nic, -1);
1366     ebpf_rss_unload(&n->ebpf_rss);
1367 }
1368 
1369 static uint16_t virtio_net_handle_rss(VirtIONet *n,
1370                                       struct iovec *iov,
1371                                       unsigned int iov_cnt,
1372                                       bool do_rss)
1373 {
1374     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1375     struct virtio_net_rss_config cfg;
1376     size_t s, offset = 0, size_get;
1377     uint16_t queue_pairs, i;
1378     struct {
1379         uint16_t us;
1380         uint8_t b;
1381     } QEMU_PACKED temp;
1382     const char *err_msg = "";
1383     uint32_t err_value = 0;
1384 
1385     if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) {
1386         err_msg = "RSS is not negotiated";
1387         goto error;
1388     }
1389     if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) {
1390         err_msg = "Hash report is not negotiated";
1391         goto error;
1392     }
1393     size_get = offsetof(struct virtio_net_rss_config, indirection_table);
1394     s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get);
1395     if (s != size_get) {
1396         err_msg = "Short command buffer";
1397         err_value = (uint32_t)s;
1398         goto error;
1399     }
1400     n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types);
1401     n->rss_data.indirections_len =
1402         virtio_lduw_p(vdev, &cfg.indirection_table_mask);
1403     n->rss_data.indirections_len++;
1404     if (!do_rss) {
1405         n->rss_data.indirections_len = 1;
1406     }
1407     if (!is_power_of_2(n->rss_data.indirections_len)) {
1408         err_msg = "Invalid size of indirection table";
1409         err_value = n->rss_data.indirections_len;
1410         goto error;
1411     }
1412     if (n->rss_data.indirections_len > VIRTIO_NET_RSS_MAX_TABLE_LEN) {
1413         err_msg = "Too large indirection table";
1414         err_value = n->rss_data.indirections_len;
1415         goto error;
1416     }
1417     n->rss_data.default_queue = do_rss ?
1418         virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0;
1419     if (n->rss_data.default_queue >= n->max_queue_pairs) {
1420         err_msg = "Invalid default queue";
1421         err_value = n->rss_data.default_queue;
1422         goto error;
1423     }
1424     offset += size_get;
1425     size_get = sizeof(uint16_t) * n->rss_data.indirections_len;
1426     g_free(n->rss_data.indirections_table);
1427     n->rss_data.indirections_table = g_malloc(size_get);
1428     if (!n->rss_data.indirections_table) {
1429         err_msg = "Can't allocate indirections table";
1430         err_value = n->rss_data.indirections_len;
1431         goto error;
1432     }
1433     s = iov_to_buf(iov, iov_cnt, offset,
1434                    n->rss_data.indirections_table, size_get);
1435     if (s != size_get) {
1436         err_msg = "Short indirection table buffer";
1437         err_value = (uint32_t)s;
1438         goto error;
1439     }
1440     for (i = 0; i < n->rss_data.indirections_len; ++i) {
1441         uint16_t val = n->rss_data.indirections_table[i];
1442         n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val);
1443     }
1444     offset += size_get;
1445     size_get = sizeof(temp);
1446     s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get);
1447     if (s != size_get) {
1448         err_msg = "Can't get queue_pairs";
1449         err_value = (uint32_t)s;
1450         goto error;
1451     }
1452     queue_pairs = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queue_pairs;
1453     if (queue_pairs == 0 || queue_pairs > n->max_queue_pairs) {
1454         err_msg = "Invalid number of queue_pairs";
1455         err_value = queue_pairs;
1456         goto error;
1457     }
1458     if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) {
1459         err_msg = "Invalid key size";
1460         err_value = temp.b;
1461         goto error;
1462     }
1463     if (!temp.b && n->rss_data.hash_types) {
1464         err_msg = "No key provided";
1465         err_value = 0;
1466         goto error;
1467     }
1468     if (!temp.b && !n->rss_data.hash_types) {
1469         virtio_net_disable_rss(n);
1470         return queue_pairs;
1471     }
1472     offset += size_get;
1473     size_get = temp.b;
1474     s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get);
1475     if (s != size_get) {
1476         err_msg = "Can get key buffer";
1477         err_value = (uint32_t)s;
1478         goto error;
1479     }
1480     n->rss_data.enabled = true;
1481     virtio_net_commit_rss_config(n);
1482     return queue_pairs;
1483 error:
1484     trace_virtio_net_rss_error(err_msg, err_value);
1485     virtio_net_disable_rss(n);
1486     return 0;
1487 }
1488 
1489 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
1490                                 struct iovec *iov, unsigned int iov_cnt)
1491 {
1492     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1493     uint16_t queue_pairs;
1494     NetClientState *nc = qemu_get_queue(n->nic);
1495 
1496     virtio_net_disable_rss(n);
1497     if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) {
1498         queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, false);
1499         return queue_pairs ? VIRTIO_NET_OK : VIRTIO_NET_ERR;
1500     }
1501     if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
1502         queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, true);
1503     } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
1504         struct virtio_net_ctrl_mq mq;
1505         size_t s;
1506         if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) {
1507             return VIRTIO_NET_ERR;
1508         }
1509         s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
1510         if (s != sizeof(mq)) {
1511             return VIRTIO_NET_ERR;
1512         }
1513         queue_pairs = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
1514 
1515     } else {
1516         return VIRTIO_NET_ERR;
1517     }
1518 
1519     if (queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1520         queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1521         queue_pairs > n->max_queue_pairs ||
1522         !n->multiqueue) {
1523         return VIRTIO_NET_ERR;
1524     }
1525 
1526     n->curr_queue_pairs = queue_pairs;
1527     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
1528         /*
1529          * Avoid updating the backend for a vdpa device: We're only interested
1530          * in updating the device model queues.
1531          */
1532         return VIRTIO_NET_OK;
1533     }
1534     /* stop the backend before changing the number of queue_pairs to avoid handling a
1535      * disabled queue */
1536     virtio_net_set_status(vdev, vdev->status);
1537     virtio_net_set_queue_pairs(n);
1538 
1539     return VIRTIO_NET_OK;
1540 }
1541 
1542 size_t virtio_net_handle_ctrl_iov(VirtIODevice *vdev,
1543                                   const struct iovec *in_sg, unsigned in_num,
1544                                   const struct iovec *out_sg,
1545                                   unsigned out_num)
1546 {
1547     VirtIONet *n = VIRTIO_NET(vdev);
1548     struct virtio_net_ctrl_hdr ctrl;
1549     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
1550     size_t s;
1551     struct iovec *iov, *iov2;
1552 
1553     if (iov_size(in_sg, in_num) < sizeof(status) ||
1554         iov_size(out_sg, out_num) < sizeof(ctrl)) {
1555         virtio_error(vdev, "virtio-net ctrl missing headers");
1556         return 0;
1557     }
1558 
1559     iov2 = iov = g_memdup2(out_sg, sizeof(struct iovec) * out_num);
1560     s = iov_to_buf(iov, out_num, 0, &ctrl, sizeof(ctrl));
1561     iov_discard_front(&iov, &out_num, sizeof(ctrl));
1562     if (s != sizeof(ctrl)) {
1563         status = VIRTIO_NET_ERR;
1564     } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
1565         status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, out_num);
1566     } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
1567         status = virtio_net_handle_mac(n, ctrl.cmd, iov, out_num);
1568     } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
1569         status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, out_num);
1570     } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
1571         status = virtio_net_handle_announce(n, ctrl.cmd, iov, out_num);
1572     } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
1573         status = virtio_net_handle_mq(n, ctrl.cmd, iov, out_num);
1574     } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
1575         status = virtio_net_handle_offloads(n, ctrl.cmd, iov, out_num);
1576     }
1577 
1578     s = iov_from_buf(in_sg, in_num, 0, &status, sizeof(status));
1579     assert(s == sizeof(status));
1580 
1581     g_free(iov2);
1582     return sizeof(status);
1583 }
1584 
1585 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
1586 {
1587     VirtQueueElement *elem;
1588 
1589     for (;;) {
1590         size_t written;
1591         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
1592         if (!elem) {
1593             break;
1594         }
1595 
1596         written = virtio_net_handle_ctrl_iov(vdev, elem->in_sg, elem->in_num,
1597                                              elem->out_sg, elem->out_num);
1598         if (written > 0) {
1599             virtqueue_push(vq, elem, written);
1600             virtio_notify(vdev, vq);
1601             g_free(elem);
1602         } else {
1603             virtqueue_detach_element(vq, elem, 0);
1604             g_free(elem);
1605             break;
1606         }
1607     }
1608 }
1609 
1610 /* RX */
1611 
1612 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
1613 {
1614     VirtIONet *n = VIRTIO_NET(vdev);
1615     int queue_index = vq2q(virtio_get_queue_index(vq));
1616 
1617     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
1618 }
1619 
1620 static bool virtio_net_can_receive(NetClientState *nc)
1621 {
1622     VirtIONet *n = qemu_get_nic_opaque(nc);
1623     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1624     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1625 
1626     if (!vdev->vm_running) {
1627         return false;
1628     }
1629 
1630     if (nc->queue_index >= n->curr_queue_pairs) {
1631         return false;
1632     }
1633 
1634     if (!virtio_queue_ready(q->rx_vq) ||
1635         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1636         return false;
1637     }
1638 
1639     return true;
1640 }
1641 
1642 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1643 {
1644     VirtIONet *n = q->n;
1645     if (virtio_queue_empty(q->rx_vq) ||
1646         (n->mergeable_rx_bufs &&
1647          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1648         virtio_queue_set_notification(q->rx_vq, 1);
1649 
1650         /* To avoid a race condition where the guest has made some buffers
1651          * available after the above check but before notification was
1652          * enabled, check for available buffers again.
1653          */
1654         if (virtio_queue_empty(q->rx_vq) ||
1655             (n->mergeable_rx_bufs &&
1656              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1657             return 0;
1658         }
1659     }
1660 
1661     virtio_queue_set_notification(q->rx_vq, 0);
1662     return 1;
1663 }
1664 
1665 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1666 {
1667     virtio_tswap16s(vdev, &hdr->hdr_len);
1668     virtio_tswap16s(vdev, &hdr->gso_size);
1669     virtio_tswap16s(vdev, &hdr->csum_start);
1670     virtio_tswap16s(vdev, &hdr->csum_offset);
1671 }
1672 
1673 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1674  * it never finds out that the packets don't have valid checksums.  This
1675  * causes dhclient to get upset.  Fedora's carried a patch for ages to
1676  * fix this with Xen but it hasn't appeared in an upstream release of
1677  * dhclient yet.
1678  *
1679  * To avoid breaking existing guests, we catch udp packets and add
1680  * checksums.  This is terrible but it's better than hacking the guest
1681  * kernels.
1682  *
1683  * N.B. if we introduce a zero-copy API, this operation is no longer free so
1684  * we should provide a mechanism to disable it to avoid polluting the host
1685  * cache.
1686  */
1687 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1688                                         uint8_t *buf, size_t size)
1689 {
1690     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1691         (size > 27 && size < 1500) && /* normal sized MTU */
1692         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1693         (buf[23] == 17) && /* ip.protocol == UDP */
1694         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1695         net_checksum_calculate(buf, size, CSUM_UDP);
1696         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1697     }
1698 }
1699 
1700 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1701                            const void *buf, size_t size)
1702 {
1703     if (n->has_vnet_hdr) {
1704         /* FIXME this cast is evil */
1705         void *wbuf = (void *)buf;
1706         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1707                                     size - n->host_hdr_len);
1708 
1709         if (n->needs_vnet_hdr_swap) {
1710             virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1711         }
1712         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1713     } else {
1714         struct virtio_net_hdr hdr = {
1715             .flags = 0,
1716             .gso_type = VIRTIO_NET_HDR_GSO_NONE
1717         };
1718         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1719     }
1720 }
1721 
1722 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1723 {
1724     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1725     static const uint8_t vlan[] = {0x81, 0x00};
1726     uint8_t *ptr = (uint8_t *)buf;
1727     int i;
1728 
1729     if (n->promisc)
1730         return 1;
1731 
1732     ptr += n->host_hdr_len;
1733 
1734     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1735         int vid = lduw_be_p(ptr + 14) & 0xfff;
1736         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1737             return 0;
1738     }
1739 
1740     if (ptr[0] & 1) { // multicast
1741         if (!memcmp(ptr, bcast, sizeof(bcast))) {
1742             return !n->nobcast;
1743         } else if (n->nomulti) {
1744             return 0;
1745         } else if (n->allmulti || n->mac_table.multi_overflow) {
1746             return 1;
1747         }
1748 
1749         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1750             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1751                 return 1;
1752             }
1753         }
1754     } else { // unicast
1755         if (n->nouni) {
1756             return 0;
1757         } else if (n->alluni || n->mac_table.uni_overflow) {
1758             return 1;
1759         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1760             return 1;
1761         }
1762 
1763         for (i = 0; i < n->mac_table.first_multi; i++) {
1764             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1765                 return 1;
1766             }
1767         }
1768     }
1769 
1770     return 0;
1771 }
1772 
1773 static uint8_t virtio_net_get_hash_type(bool hasip4,
1774                                         bool hasip6,
1775                                         EthL4HdrProto l4hdr_proto,
1776                                         uint32_t types)
1777 {
1778     if (hasip4) {
1779         switch (l4hdr_proto) {
1780         case ETH_L4_HDR_PROTO_TCP:
1781             if (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4) {
1782                 return NetPktRssIpV4Tcp;
1783             }
1784             break;
1785 
1786         case ETH_L4_HDR_PROTO_UDP:
1787             if (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4) {
1788                 return NetPktRssIpV4Udp;
1789             }
1790             break;
1791 
1792         default:
1793             break;
1794         }
1795 
1796         if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
1797             return NetPktRssIpV4;
1798         }
1799     } else if (hasip6) {
1800         switch (l4hdr_proto) {
1801         case ETH_L4_HDR_PROTO_TCP:
1802             if (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) {
1803                 return NetPktRssIpV6TcpEx;
1804             }
1805             if (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv6) {
1806                 return NetPktRssIpV6Tcp;
1807             }
1808             break;
1809 
1810         case ETH_L4_HDR_PROTO_UDP:
1811             if (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) {
1812                 return NetPktRssIpV6UdpEx;
1813             }
1814             if (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv6) {
1815                 return NetPktRssIpV6Udp;
1816             }
1817             break;
1818 
1819         default:
1820             break;
1821         }
1822 
1823         if (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) {
1824             return NetPktRssIpV6Ex;
1825         }
1826         if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv6) {
1827             return NetPktRssIpV6;
1828         }
1829     }
1830     return 0xff;
1831 }
1832 
1833 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf,
1834                                   size_t size,
1835                                   struct virtio_net_hdr_v1_hash *hdr)
1836 {
1837     VirtIONet *n = qemu_get_nic_opaque(nc);
1838     unsigned int index = nc->queue_index, new_index = index;
1839     struct NetRxPkt *pkt = n->rx_pkt;
1840     uint8_t net_hash_type;
1841     uint32_t hash;
1842     bool hasip4, hasip6;
1843     EthL4HdrProto l4hdr_proto;
1844     static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = {
1845         VIRTIO_NET_HASH_REPORT_IPv4,
1846         VIRTIO_NET_HASH_REPORT_TCPv4,
1847         VIRTIO_NET_HASH_REPORT_TCPv6,
1848         VIRTIO_NET_HASH_REPORT_IPv6,
1849         VIRTIO_NET_HASH_REPORT_IPv6_EX,
1850         VIRTIO_NET_HASH_REPORT_TCPv6_EX,
1851         VIRTIO_NET_HASH_REPORT_UDPv4,
1852         VIRTIO_NET_HASH_REPORT_UDPv6,
1853         VIRTIO_NET_HASH_REPORT_UDPv6_EX
1854     };
1855     struct iovec iov = {
1856         .iov_base = (void *)buf,
1857         .iov_len = size
1858     };
1859 
1860     net_rx_pkt_set_protocols(pkt, &iov, 1, n->host_hdr_len);
1861     net_rx_pkt_get_protocols(pkt, &hasip4, &hasip6, &l4hdr_proto);
1862     net_hash_type = virtio_net_get_hash_type(hasip4, hasip6, l4hdr_proto,
1863                                              n->rss_data.hash_types);
1864     if (net_hash_type > NetPktRssIpV6UdpEx) {
1865         if (n->rss_data.populate_hash) {
1866             hdr->hash_value = VIRTIO_NET_HASH_REPORT_NONE;
1867             hdr->hash_report = 0;
1868         }
1869         return n->rss_data.redirect ? n->rss_data.default_queue : -1;
1870     }
1871 
1872     hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key);
1873 
1874     if (n->rss_data.populate_hash) {
1875         hdr->hash_value = hash;
1876         hdr->hash_report = reports[net_hash_type];
1877     }
1878 
1879     if (n->rss_data.redirect) {
1880         new_index = hash & (n->rss_data.indirections_len - 1);
1881         new_index = n->rss_data.indirections_table[new_index];
1882     }
1883 
1884     return (index == new_index) ? -1 : new_index;
1885 }
1886 
1887 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
1888                                       size_t size, bool no_rss)
1889 {
1890     VirtIONet *n = qemu_get_nic_opaque(nc);
1891     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1892     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1893     VirtQueueElement *elems[VIRTQUEUE_MAX_SIZE];
1894     size_t lens[VIRTQUEUE_MAX_SIZE];
1895     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1896     struct virtio_net_hdr_v1_hash extra_hdr;
1897     unsigned mhdr_cnt = 0;
1898     size_t offset, i, guest_offset, j;
1899     ssize_t err;
1900 
1901     if (!virtio_net_can_receive(nc)) {
1902         return -1;
1903     }
1904 
1905     if (!no_rss && n->rss_data.enabled && n->rss_data.enabled_software_rss) {
1906         int index = virtio_net_process_rss(nc, buf, size, &extra_hdr);
1907         if (index >= 0) {
1908             NetClientState *nc2 = qemu_get_subqueue(n->nic, index);
1909             return virtio_net_receive_rcu(nc2, buf, size, true);
1910         }
1911     }
1912 
1913     /* hdr_len refers to the header we supply to the guest */
1914     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1915         return 0;
1916     }
1917 
1918     if (!receive_filter(n, buf, size))
1919         return size;
1920 
1921     offset = i = 0;
1922 
1923     while (offset < size) {
1924         VirtQueueElement *elem;
1925         int len, total;
1926         const struct iovec *sg;
1927 
1928         total = 0;
1929 
1930         if (i == VIRTQUEUE_MAX_SIZE) {
1931             virtio_error(vdev, "virtio-net unexpected long buffer chain");
1932             err = size;
1933             goto err;
1934         }
1935 
1936         elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1937         if (!elem) {
1938             if (i) {
1939                 virtio_error(vdev, "virtio-net unexpected empty queue: "
1940                              "i %zd mergeable %d offset %zd, size %zd, "
1941                              "guest hdr len %zd, host hdr len %zd "
1942                              "guest features 0x%" PRIx64,
1943                              i, n->mergeable_rx_bufs, offset, size,
1944                              n->guest_hdr_len, n->host_hdr_len,
1945                              vdev->guest_features);
1946             }
1947             err = -1;
1948             goto err;
1949         }
1950 
1951         if (elem->in_num < 1) {
1952             virtio_error(vdev,
1953                          "virtio-net receive queue contains no in buffers");
1954             virtqueue_detach_element(q->rx_vq, elem, 0);
1955             g_free(elem);
1956             err = -1;
1957             goto err;
1958         }
1959 
1960         sg = elem->in_sg;
1961         if (i == 0) {
1962             assert(offset == 0);
1963             if (n->mergeable_rx_bufs) {
1964                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1965                                     sg, elem->in_num,
1966                                     offsetof(typeof(extra_hdr), hdr.num_buffers),
1967                                     sizeof(extra_hdr.hdr.num_buffers));
1968             }
1969 
1970             receive_header(n, sg, elem->in_num, buf, size);
1971             if (n->rss_data.populate_hash) {
1972                 offset = offsetof(typeof(extra_hdr), hash_value);
1973                 iov_from_buf(sg, elem->in_num, offset,
1974                              (char *)&extra_hdr + offset,
1975                              sizeof(extra_hdr.hash_value) +
1976                              sizeof(extra_hdr.hash_report));
1977             }
1978             offset = n->host_hdr_len;
1979             total += n->guest_hdr_len;
1980             guest_offset = n->guest_hdr_len;
1981         } else {
1982             guest_offset = 0;
1983         }
1984 
1985         /* copy in packet.  ugh */
1986         len = iov_from_buf(sg, elem->in_num, guest_offset,
1987                            buf + offset, size - offset);
1988         total += len;
1989         offset += len;
1990         /* If buffers can't be merged, at this point we
1991          * must have consumed the complete packet.
1992          * Otherwise, drop it. */
1993         if (!n->mergeable_rx_bufs && offset < size) {
1994             virtqueue_unpop(q->rx_vq, elem, total);
1995             g_free(elem);
1996             err = size;
1997             goto err;
1998         }
1999 
2000         elems[i] = elem;
2001         lens[i] = total;
2002         i++;
2003     }
2004 
2005     if (mhdr_cnt) {
2006         virtio_stw_p(vdev, &extra_hdr.hdr.num_buffers, i);
2007         iov_from_buf(mhdr_sg, mhdr_cnt,
2008                      0,
2009                      &extra_hdr.hdr.num_buffers,
2010                      sizeof extra_hdr.hdr.num_buffers);
2011     }
2012 
2013     for (j = 0; j < i; j++) {
2014         /* signal other side */
2015         virtqueue_fill(q->rx_vq, elems[j], lens[j], j);
2016         g_free(elems[j]);
2017     }
2018 
2019     virtqueue_flush(q->rx_vq, i);
2020     virtio_notify(vdev, q->rx_vq);
2021 
2022     return size;
2023 
2024 err:
2025     for (j = 0; j < i; j++) {
2026         virtqueue_detach_element(q->rx_vq, elems[j], lens[j]);
2027         g_free(elems[j]);
2028     }
2029 
2030     return err;
2031 }
2032 
2033 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf,
2034                                   size_t size)
2035 {
2036     RCU_READ_LOCK_GUARD();
2037 
2038     return virtio_net_receive_rcu(nc, buf, size, false);
2039 }
2040 
2041 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain,
2042                                          const uint8_t *buf,
2043                                          VirtioNetRscUnit *unit)
2044 {
2045     uint16_t ip_hdrlen;
2046     struct ip_header *ip;
2047 
2048     ip = (struct ip_header *)(buf + chain->n->guest_hdr_len
2049                               + sizeof(struct eth_header));
2050     unit->ip = (void *)ip;
2051     ip_hdrlen = (ip->ip_ver_len & 0xF) << 2;
2052     unit->ip_plen = &ip->ip_len;
2053     unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen);
2054     unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
2055     unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen;
2056 }
2057 
2058 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain,
2059                                          const uint8_t *buf,
2060                                          VirtioNetRscUnit *unit)
2061 {
2062     struct ip6_header *ip6;
2063 
2064     ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len
2065                                  + sizeof(struct eth_header));
2066     unit->ip = ip6;
2067     unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
2068     unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip)
2069                                         + sizeof(struct ip6_header));
2070     unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
2071 
2072     /* There is a difference between payload length in ipv4 and v6,
2073        ip header is excluded in ipv6 */
2074     unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen;
2075 }
2076 
2077 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain,
2078                                        VirtioNetRscSeg *seg)
2079 {
2080     int ret;
2081     struct virtio_net_hdr_v1 *h;
2082 
2083     h = (struct virtio_net_hdr_v1 *)seg->buf;
2084     h->flags = 0;
2085     h->gso_type = VIRTIO_NET_HDR_GSO_NONE;
2086 
2087     if (seg->is_coalesced) {
2088         h->rsc.segments = seg->packets;
2089         h->rsc.dup_acks = seg->dup_ack;
2090         h->flags = VIRTIO_NET_HDR_F_RSC_INFO;
2091         if (chain->proto == ETH_P_IP) {
2092             h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2093         } else {
2094             h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2095         }
2096     }
2097 
2098     ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size);
2099     QTAILQ_REMOVE(&chain->buffers, seg, next);
2100     g_free(seg->buf);
2101     g_free(seg);
2102 
2103     return ret;
2104 }
2105 
2106 static void virtio_net_rsc_purge(void *opq)
2107 {
2108     VirtioNetRscSeg *seg, *rn;
2109     VirtioNetRscChain *chain = (VirtioNetRscChain *)opq;
2110 
2111     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) {
2112         if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2113             chain->stat.purge_failed++;
2114             continue;
2115         }
2116     }
2117 
2118     chain->stat.timer++;
2119     if (!QTAILQ_EMPTY(&chain->buffers)) {
2120         timer_mod(chain->drain_timer,
2121               qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
2122     }
2123 }
2124 
2125 static void virtio_net_rsc_cleanup(VirtIONet *n)
2126 {
2127     VirtioNetRscChain *chain, *rn_chain;
2128     VirtioNetRscSeg *seg, *rn_seg;
2129 
2130     QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) {
2131         QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) {
2132             QTAILQ_REMOVE(&chain->buffers, seg, next);
2133             g_free(seg->buf);
2134             g_free(seg);
2135         }
2136 
2137         timer_free(chain->drain_timer);
2138         QTAILQ_REMOVE(&n->rsc_chains, chain, next);
2139         g_free(chain);
2140     }
2141 }
2142 
2143 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain,
2144                                      NetClientState *nc,
2145                                      const uint8_t *buf, size_t size)
2146 {
2147     uint16_t hdr_len;
2148     VirtioNetRscSeg *seg;
2149 
2150     hdr_len = chain->n->guest_hdr_len;
2151     seg = g_new(VirtioNetRscSeg, 1);
2152     seg->buf = g_malloc(hdr_len + sizeof(struct eth_header)
2153         + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD);
2154     memcpy(seg->buf, buf, size);
2155     seg->size = size;
2156     seg->packets = 1;
2157     seg->dup_ack = 0;
2158     seg->is_coalesced = 0;
2159     seg->nc = nc;
2160 
2161     QTAILQ_INSERT_TAIL(&chain->buffers, seg, next);
2162     chain->stat.cache++;
2163 
2164     switch (chain->proto) {
2165     case ETH_P_IP:
2166         virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
2167         break;
2168     case ETH_P_IPV6:
2169         virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit);
2170         break;
2171     default:
2172         g_assert_not_reached();
2173     }
2174 }
2175 
2176 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain,
2177                                          VirtioNetRscSeg *seg,
2178                                          const uint8_t *buf,
2179                                          struct tcp_header *n_tcp,
2180                                          struct tcp_header *o_tcp)
2181 {
2182     uint32_t nack, oack;
2183     uint16_t nwin, owin;
2184 
2185     nack = htonl(n_tcp->th_ack);
2186     nwin = htons(n_tcp->th_win);
2187     oack = htonl(o_tcp->th_ack);
2188     owin = htons(o_tcp->th_win);
2189 
2190     if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) {
2191         chain->stat.ack_out_of_win++;
2192         return RSC_FINAL;
2193     } else if (nack == oack) {
2194         /* duplicated ack or window probe */
2195         if (nwin == owin) {
2196             /* duplicated ack, add dup ack count due to whql test up to 1 */
2197             chain->stat.dup_ack++;
2198             return RSC_FINAL;
2199         } else {
2200             /* Coalesce window update */
2201             o_tcp->th_win = n_tcp->th_win;
2202             chain->stat.win_update++;
2203             return RSC_COALESCE;
2204         }
2205     } else {
2206         /* pure ack, go to 'C', finalize*/
2207         chain->stat.pure_ack++;
2208         return RSC_FINAL;
2209     }
2210 }
2211 
2212 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain,
2213                                             VirtioNetRscSeg *seg,
2214                                             const uint8_t *buf,
2215                                             VirtioNetRscUnit *n_unit)
2216 {
2217     void *data;
2218     uint16_t o_ip_len;
2219     uint32_t nseq, oseq;
2220     VirtioNetRscUnit *o_unit;
2221 
2222     o_unit = &seg->unit;
2223     o_ip_len = htons(*o_unit->ip_plen);
2224     nseq = htonl(n_unit->tcp->th_seq);
2225     oseq = htonl(o_unit->tcp->th_seq);
2226 
2227     /* out of order or retransmitted. */
2228     if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) {
2229         chain->stat.data_out_of_win++;
2230         return RSC_FINAL;
2231     }
2232 
2233     data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen;
2234     if (nseq == oseq) {
2235         if ((o_unit->payload == 0) && n_unit->payload) {
2236             /* From no payload to payload, normal case, not a dup ack or etc */
2237             chain->stat.data_after_pure_ack++;
2238             goto coalesce;
2239         } else {
2240             return virtio_net_rsc_handle_ack(chain, seg, buf,
2241                                              n_unit->tcp, o_unit->tcp);
2242         }
2243     } else if ((nseq - oseq) != o_unit->payload) {
2244         /* Not a consistent packet, out of order */
2245         chain->stat.data_out_of_order++;
2246         return RSC_FINAL;
2247     } else {
2248 coalesce:
2249         if ((o_ip_len + n_unit->payload) > chain->max_payload) {
2250             chain->stat.over_size++;
2251             return RSC_FINAL;
2252         }
2253 
2254         /* Here comes the right data, the payload length in v4/v6 is different,
2255            so use the field value to update and record the new data len */
2256         o_unit->payload += n_unit->payload; /* update new data len */
2257 
2258         /* update field in ip header */
2259         *o_unit->ip_plen = htons(o_ip_len + n_unit->payload);
2260 
2261         /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
2262            for windows guest, while this may change the behavior for linux
2263            guest (only if it uses RSC feature). */
2264         o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags;
2265 
2266         o_unit->tcp->th_ack = n_unit->tcp->th_ack;
2267         o_unit->tcp->th_win = n_unit->tcp->th_win;
2268 
2269         memmove(seg->buf + seg->size, data, n_unit->payload);
2270         seg->size += n_unit->payload;
2271         seg->packets++;
2272         chain->stat.coalesced++;
2273         return RSC_COALESCE;
2274     }
2275 }
2276 
2277 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain,
2278                                         VirtioNetRscSeg *seg,
2279                                         const uint8_t *buf, size_t size,
2280                                         VirtioNetRscUnit *unit)
2281 {
2282     struct ip_header *ip1, *ip2;
2283 
2284     ip1 = (struct ip_header *)(unit->ip);
2285     ip2 = (struct ip_header *)(seg->unit.ip);
2286     if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst)
2287         || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2288         || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2289         chain->stat.no_match++;
2290         return RSC_NO_MATCH;
2291     }
2292 
2293     return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2294 }
2295 
2296 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain,
2297                                         VirtioNetRscSeg *seg,
2298                                         const uint8_t *buf, size_t size,
2299                                         VirtioNetRscUnit *unit)
2300 {
2301     struct ip6_header *ip1, *ip2;
2302 
2303     ip1 = (struct ip6_header *)(unit->ip);
2304     ip2 = (struct ip6_header *)(seg->unit.ip);
2305     if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address))
2306         || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address))
2307         || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2308         || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2309             chain->stat.no_match++;
2310             return RSC_NO_MATCH;
2311     }
2312 
2313     return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2314 }
2315 
2316 /* Packets with 'SYN' should bypass, other flag should be sent after drain
2317  * to prevent out of order */
2318 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain,
2319                                          struct tcp_header *tcp)
2320 {
2321     uint16_t tcp_hdr;
2322     uint16_t tcp_flag;
2323 
2324     tcp_flag = htons(tcp->th_offset_flags);
2325     tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10;
2326     tcp_flag &= VIRTIO_NET_TCP_FLAG;
2327     if (tcp_flag & TH_SYN) {
2328         chain->stat.tcp_syn++;
2329         return RSC_BYPASS;
2330     }
2331 
2332     if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) {
2333         chain->stat.tcp_ctrl_drain++;
2334         return RSC_FINAL;
2335     }
2336 
2337     if (tcp_hdr > sizeof(struct tcp_header)) {
2338         chain->stat.tcp_all_opt++;
2339         return RSC_FINAL;
2340     }
2341 
2342     return RSC_CANDIDATE;
2343 }
2344 
2345 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain,
2346                                          NetClientState *nc,
2347                                          const uint8_t *buf, size_t size,
2348                                          VirtioNetRscUnit *unit)
2349 {
2350     int ret;
2351     VirtioNetRscSeg *seg, *nseg;
2352 
2353     if (QTAILQ_EMPTY(&chain->buffers)) {
2354         chain->stat.empty_cache++;
2355         virtio_net_rsc_cache_buf(chain, nc, buf, size);
2356         timer_mod(chain->drain_timer,
2357               qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
2358         return size;
2359     }
2360 
2361     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2362         if (chain->proto == ETH_P_IP) {
2363             ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
2364         } else {
2365             ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit);
2366         }
2367 
2368         if (ret == RSC_FINAL) {
2369             if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2370                 /* Send failed */
2371                 chain->stat.final_failed++;
2372                 return 0;
2373             }
2374 
2375             /* Send current packet */
2376             return virtio_net_do_receive(nc, buf, size);
2377         } else if (ret == RSC_NO_MATCH) {
2378             continue;
2379         } else {
2380             /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
2381             seg->is_coalesced = 1;
2382             return size;
2383         }
2384     }
2385 
2386     chain->stat.no_match_cache++;
2387     virtio_net_rsc_cache_buf(chain, nc, buf, size);
2388     return size;
2389 }
2390 
2391 /* Drain a connection data, this is to avoid out of order segments */
2392 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain,
2393                                         NetClientState *nc,
2394                                         const uint8_t *buf, size_t size,
2395                                         uint16_t ip_start, uint16_t ip_size,
2396                                         uint16_t tcp_port)
2397 {
2398     VirtioNetRscSeg *seg, *nseg;
2399     uint32_t ppair1, ppair2;
2400 
2401     ppair1 = *(uint32_t *)(buf + tcp_port);
2402     QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2403         ppair2 = *(uint32_t *)(seg->buf + tcp_port);
2404         if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size)
2405             || (ppair1 != ppair2)) {
2406             continue;
2407         }
2408         if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2409             chain->stat.drain_failed++;
2410         }
2411 
2412         break;
2413     }
2414 
2415     return virtio_net_do_receive(nc, buf, size);
2416 }
2417 
2418 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain,
2419                                             struct ip_header *ip,
2420                                             const uint8_t *buf, size_t size)
2421 {
2422     uint16_t ip_len;
2423 
2424     /* Not an ipv4 packet */
2425     if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) {
2426         chain->stat.ip_option++;
2427         return RSC_BYPASS;
2428     }
2429 
2430     /* Don't handle packets with ip option */
2431     if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) {
2432         chain->stat.ip_option++;
2433         return RSC_BYPASS;
2434     }
2435 
2436     if (ip->ip_p != IPPROTO_TCP) {
2437         chain->stat.bypass_not_tcp++;
2438         return RSC_BYPASS;
2439     }
2440 
2441     /* Don't handle packets with ip fragment */
2442     if (!(htons(ip->ip_off) & IP_DF)) {
2443         chain->stat.ip_frag++;
2444         return RSC_BYPASS;
2445     }
2446 
2447     /* Don't handle packets with ecn flag */
2448     if (IPTOS_ECN(ip->ip_tos)) {
2449         chain->stat.ip_ecn++;
2450         return RSC_BYPASS;
2451     }
2452 
2453     ip_len = htons(ip->ip_len);
2454     if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header))
2455         || ip_len > (size - chain->n->guest_hdr_len -
2456                      sizeof(struct eth_header))) {
2457         chain->stat.ip_hacked++;
2458         return RSC_BYPASS;
2459     }
2460 
2461     return RSC_CANDIDATE;
2462 }
2463 
2464 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain,
2465                                       NetClientState *nc,
2466                                       const uint8_t *buf, size_t size)
2467 {
2468     int32_t ret;
2469     uint16_t hdr_len;
2470     VirtioNetRscUnit unit;
2471 
2472     hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2473 
2474     if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)
2475         + sizeof(struct tcp_header))) {
2476         chain->stat.bypass_not_tcp++;
2477         return virtio_net_do_receive(nc, buf, size);
2478     }
2479 
2480     virtio_net_rsc_extract_unit4(chain, buf, &unit);
2481     if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size)
2482         != RSC_CANDIDATE) {
2483         return virtio_net_do_receive(nc, buf, size);
2484     }
2485 
2486     ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2487     if (ret == RSC_BYPASS) {
2488         return virtio_net_do_receive(nc, buf, size);
2489     } else if (ret == RSC_FINAL) {
2490         return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2491                 ((hdr_len + sizeof(struct eth_header)) + 12),
2492                 VIRTIO_NET_IP4_ADDR_SIZE,
2493                 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header));
2494     }
2495 
2496     return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2497 }
2498 
2499 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain,
2500                                             struct ip6_header *ip6,
2501                                             const uint8_t *buf, size_t size)
2502 {
2503     uint16_t ip_len;
2504 
2505     if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4)
2506         != IP_HEADER_VERSION_6) {
2507         return RSC_BYPASS;
2508     }
2509 
2510     /* Both option and protocol is checked in this */
2511     if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) {
2512         chain->stat.bypass_not_tcp++;
2513         return RSC_BYPASS;
2514     }
2515 
2516     ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
2517     if (ip_len < sizeof(struct tcp_header) ||
2518         ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header)
2519                   - sizeof(struct ip6_header))) {
2520         chain->stat.ip_hacked++;
2521         return RSC_BYPASS;
2522     }
2523 
2524     /* Don't handle packets with ecn flag */
2525     if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) {
2526         chain->stat.ip_ecn++;
2527         return RSC_BYPASS;
2528     }
2529 
2530     return RSC_CANDIDATE;
2531 }
2532 
2533 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc,
2534                                       const uint8_t *buf, size_t size)
2535 {
2536     int32_t ret;
2537     uint16_t hdr_len;
2538     VirtioNetRscChain *chain;
2539     VirtioNetRscUnit unit;
2540 
2541     chain = opq;
2542     hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2543 
2544     if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header)
2545         + sizeof(tcp_header))) {
2546         return virtio_net_do_receive(nc, buf, size);
2547     }
2548 
2549     virtio_net_rsc_extract_unit6(chain, buf, &unit);
2550     if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain,
2551                                                  unit.ip, buf, size)) {
2552         return virtio_net_do_receive(nc, buf, size);
2553     }
2554 
2555     ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2556     if (ret == RSC_BYPASS) {
2557         return virtio_net_do_receive(nc, buf, size);
2558     } else if (ret == RSC_FINAL) {
2559         return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2560                 ((hdr_len + sizeof(struct eth_header)) + 8),
2561                 VIRTIO_NET_IP6_ADDR_SIZE,
2562                 hdr_len + sizeof(struct eth_header)
2563                 + sizeof(struct ip6_header));
2564     }
2565 
2566     return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2567 }
2568 
2569 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n,
2570                                                       NetClientState *nc,
2571                                                       uint16_t proto)
2572 {
2573     VirtioNetRscChain *chain;
2574 
2575     if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) {
2576         return NULL;
2577     }
2578 
2579     QTAILQ_FOREACH(chain, &n->rsc_chains, next) {
2580         if (chain->proto == proto) {
2581             return chain;
2582         }
2583     }
2584 
2585     chain = g_malloc(sizeof(*chain));
2586     chain->n = n;
2587     chain->proto = proto;
2588     if (proto == (uint16_t)ETH_P_IP) {
2589         chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
2590         chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2591     } else {
2592         chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD;
2593         chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2594     }
2595     chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST,
2596                                       virtio_net_rsc_purge, chain);
2597     memset(&chain->stat, 0, sizeof(chain->stat));
2598 
2599     QTAILQ_INIT(&chain->buffers);
2600     QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next);
2601 
2602     return chain;
2603 }
2604 
2605 static ssize_t virtio_net_rsc_receive(NetClientState *nc,
2606                                       const uint8_t *buf,
2607                                       size_t size)
2608 {
2609     uint16_t proto;
2610     VirtioNetRscChain *chain;
2611     struct eth_header *eth;
2612     VirtIONet *n;
2613 
2614     n = qemu_get_nic_opaque(nc);
2615     if (size < (n->host_hdr_len + sizeof(struct eth_header))) {
2616         return virtio_net_do_receive(nc, buf, size);
2617     }
2618 
2619     eth = (struct eth_header *)(buf + n->guest_hdr_len);
2620     proto = htons(eth->h_proto);
2621 
2622     chain = virtio_net_rsc_lookup_chain(n, nc, proto);
2623     if (chain) {
2624         chain->stat.received++;
2625         if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) {
2626             return virtio_net_rsc_receive4(chain, nc, buf, size);
2627         } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) {
2628             return virtio_net_rsc_receive6(chain, nc, buf, size);
2629         }
2630     }
2631     return virtio_net_do_receive(nc, buf, size);
2632 }
2633 
2634 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf,
2635                                   size_t size)
2636 {
2637     VirtIONet *n = qemu_get_nic_opaque(nc);
2638     if ((n->rsc4_enabled || n->rsc6_enabled)) {
2639         return virtio_net_rsc_receive(nc, buf, size);
2640     } else {
2641         return virtio_net_do_receive(nc, buf, size);
2642     }
2643 }
2644 
2645 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
2646 
2647 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
2648 {
2649     VirtIONet *n = qemu_get_nic_opaque(nc);
2650     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
2651     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2652     int ret;
2653 
2654     virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
2655     virtio_notify(vdev, q->tx_vq);
2656 
2657     g_free(q->async_tx.elem);
2658     q->async_tx.elem = NULL;
2659 
2660     virtio_queue_set_notification(q->tx_vq, 1);
2661     ret = virtio_net_flush_tx(q);
2662     if (ret >= n->tx_burst) {
2663         /*
2664          * the flush has been stopped by tx_burst
2665          * we will not receive notification for the
2666          * remainining part, so re-schedule
2667          */
2668         virtio_queue_set_notification(q->tx_vq, 0);
2669         if (q->tx_bh) {
2670             qemu_bh_schedule(q->tx_bh);
2671         } else {
2672             timer_mod(q->tx_timer,
2673                       qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2674         }
2675         q->tx_waiting = 1;
2676     }
2677 }
2678 
2679 /* TX */
2680 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
2681 {
2682     VirtIONet *n = q->n;
2683     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2684     VirtQueueElement *elem;
2685     int32_t num_packets = 0;
2686     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
2687     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2688         return num_packets;
2689     }
2690 
2691     if (q->async_tx.elem) {
2692         virtio_queue_set_notification(q->tx_vq, 0);
2693         return num_packets;
2694     }
2695 
2696     for (;;) {
2697         ssize_t ret;
2698         unsigned int out_num;
2699         struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
2700         struct virtio_net_hdr vhdr;
2701 
2702         elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
2703         if (!elem) {
2704             break;
2705         }
2706 
2707         out_num = elem->out_num;
2708         out_sg = elem->out_sg;
2709         if (out_num < 1) {
2710             virtio_error(vdev, "virtio-net header not in first element");
2711             goto detach;
2712         }
2713 
2714         if (n->needs_vnet_hdr_swap) {
2715             if (iov_to_buf(out_sg, out_num, 0, &vhdr, sizeof(vhdr)) <
2716                 sizeof(vhdr)) {
2717                 virtio_error(vdev, "virtio-net header incorrect");
2718                 goto detach;
2719             }
2720             virtio_net_hdr_swap(vdev, &vhdr);
2721             sg2[0].iov_base = &vhdr;
2722             sg2[0].iov_len = sizeof(vhdr);
2723             out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1, out_sg, out_num,
2724                                sizeof(vhdr), -1);
2725             if (out_num == VIRTQUEUE_MAX_SIZE) {
2726                 goto drop;
2727             }
2728             out_num += 1;
2729             out_sg = sg2;
2730         }
2731         /*
2732          * If host wants to see the guest header as is, we can
2733          * pass it on unchanged. Otherwise, copy just the parts
2734          * that host is interested in.
2735          */
2736         assert(n->host_hdr_len <= n->guest_hdr_len);
2737         if (n->host_hdr_len != n->guest_hdr_len) {
2738             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
2739                                        out_sg, out_num,
2740                                        0, n->host_hdr_len);
2741             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
2742                              out_sg, out_num,
2743                              n->guest_hdr_len, -1);
2744             out_num = sg_num;
2745             out_sg = sg;
2746 
2747             if (out_num < 1) {
2748                 virtio_error(vdev, "virtio-net nothing to send");
2749                 goto detach;
2750             }
2751         }
2752 
2753         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
2754                                       out_sg, out_num, virtio_net_tx_complete);
2755         if (ret == 0) {
2756             virtio_queue_set_notification(q->tx_vq, 0);
2757             q->async_tx.elem = elem;
2758             return -EBUSY;
2759         }
2760 
2761 drop:
2762         virtqueue_push(q->tx_vq, elem, 0);
2763         virtio_notify(vdev, q->tx_vq);
2764         g_free(elem);
2765 
2766         if (++num_packets >= n->tx_burst) {
2767             break;
2768         }
2769     }
2770     return num_packets;
2771 
2772 detach:
2773     virtqueue_detach_element(q->tx_vq, elem, 0);
2774     g_free(elem);
2775     return -EINVAL;
2776 }
2777 
2778 static void virtio_net_tx_timer(void *opaque);
2779 
2780 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
2781 {
2782     VirtIONet *n = VIRTIO_NET(vdev);
2783     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2784 
2785     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2786         virtio_net_drop_tx_queue_data(vdev, vq);
2787         return;
2788     }
2789 
2790     /* This happens when device was stopped but VCPU wasn't. */
2791     if (!vdev->vm_running) {
2792         q->tx_waiting = 1;
2793         return;
2794     }
2795 
2796     if (q->tx_waiting) {
2797         /* We already have queued packets, immediately flush */
2798         timer_del(q->tx_timer);
2799         virtio_net_tx_timer(q);
2800     } else {
2801         /* re-arm timer to flush it (and more) on next tick */
2802         timer_mod(q->tx_timer,
2803                   qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2804         q->tx_waiting = 1;
2805         virtio_queue_set_notification(vq, 0);
2806     }
2807 }
2808 
2809 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
2810 {
2811     VirtIONet *n = VIRTIO_NET(vdev);
2812     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2813 
2814     if (unlikely(n->vhost_started)) {
2815         return;
2816     }
2817 
2818     if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2819         virtio_net_drop_tx_queue_data(vdev, vq);
2820         return;
2821     }
2822 
2823     if (unlikely(q->tx_waiting)) {
2824         return;
2825     }
2826     q->tx_waiting = 1;
2827     /* This happens when device was stopped but VCPU wasn't. */
2828     if (!vdev->vm_running) {
2829         return;
2830     }
2831     virtio_queue_set_notification(vq, 0);
2832     qemu_bh_schedule(q->tx_bh);
2833 }
2834 
2835 static void virtio_net_tx_timer(void *opaque)
2836 {
2837     VirtIONetQueue *q = opaque;
2838     VirtIONet *n = q->n;
2839     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2840     int ret;
2841 
2842     /* This happens when device was stopped but BH wasn't. */
2843     if (!vdev->vm_running) {
2844         /* Make sure tx waiting is set, so we'll run when restarted. */
2845         assert(q->tx_waiting);
2846         return;
2847     }
2848 
2849     q->tx_waiting = 0;
2850 
2851     /* Just in case the driver is not ready on more */
2852     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2853         return;
2854     }
2855 
2856     ret = virtio_net_flush_tx(q);
2857     if (ret == -EBUSY || ret == -EINVAL) {
2858         return;
2859     }
2860     /*
2861      * If we flush a full burst of packets, assume there are
2862      * more coming and immediately rearm
2863      */
2864     if (ret >= n->tx_burst) {
2865         q->tx_waiting = 1;
2866         timer_mod(q->tx_timer,
2867                   qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2868         return;
2869     }
2870     /*
2871      * If less than a full burst, re-enable notification and flush
2872      * anything that may have come in while we weren't looking.  If
2873      * we find something, assume the guest is still active and rearm
2874      */
2875     virtio_queue_set_notification(q->tx_vq, 1);
2876     ret = virtio_net_flush_tx(q);
2877     if (ret > 0) {
2878         virtio_queue_set_notification(q->tx_vq, 0);
2879         q->tx_waiting = 1;
2880         timer_mod(q->tx_timer,
2881                   qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2882     }
2883 }
2884 
2885 static void virtio_net_tx_bh(void *opaque)
2886 {
2887     VirtIONetQueue *q = opaque;
2888     VirtIONet *n = q->n;
2889     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2890     int32_t ret;
2891 
2892     /* This happens when device was stopped but BH wasn't. */
2893     if (!vdev->vm_running) {
2894         /* Make sure tx waiting is set, so we'll run when restarted. */
2895         assert(q->tx_waiting);
2896         return;
2897     }
2898 
2899     q->tx_waiting = 0;
2900 
2901     /* Just in case the driver is not ready on more */
2902     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
2903         return;
2904     }
2905 
2906     ret = virtio_net_flush_tx(q);
2907     if (ret == -EBUSY || ret == -EINVAL) {
2908         return; /* Notification re-enable handled by tx_complete or device
2909                  * broken */
2910     }
2911 
2912     /* If we flush a full burst of packets, assume there are
2913      * more coming and immediately reschedule */
2914     if (ret >= n->tx_burst) {
2915         qemu_bh_schedule(q->tx_bh);
2916         q->tx_waiting = 1;
2917         return;
2918     }
2919 
2920     /* If less than a full burst, re-enable notification and flush
2921      * anything that may have come in while we weren't looking.  If
2922      * we find something, assume the guest is still active and reschedule */
2923     virtio_queue_set_notification(q->tx_vq, 1);
2924     ret = virtio_net_flush_tx(q);
2925     if (ret == -EINVAL) {
2926         return;
2927     } else if (ret > 0) {
2928         virtio_queue_set_notification(q->tx_vq, 0);
2929         qemu_bh_schedule(q->tx_bh);
2930         q->tx_waiting = 1;
2931     }
2932 }
2933 
2934 static void virtio_net_add_queue(VirtIONet *n, int index)
2935 {
2936     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2937 
2938     n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
2939                                            virtio_net_handle_rx);
2940 
2941     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
2942         n->vqs[index].tx_vq =
2943             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2944                              virtio_net_handle_tx_timer);
2945         n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
2946                                               virtio_net_tx_timer,
2947                                               &n->vqs[index]);
2948     } else {
2949         n->vqs[index].tx_vq =
2950             virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2951                              virtio_net_handle_tx_bh);
2952         n->vqs[index].tx_bh = qemu_bh_new_guarded(virtio_net_tx_bh, &n->vqs[index],
2953                                                   &DEVICE(vdev)->mem_reentrancy_guard);
2954     }
2955 
2956     n->vqs[index].tx_waiting = 0;
2957     n->vqs[index].n = n;
2958 }
2959 
2960 static void virtio_net_del_queue(VirtIONet *n, int index)
2961 {
2962     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2963     VirtIONetQueue *q = &n->vqs[index];
2964     NetClientState *nc = qemu_get_subqueue(n->nic, index);
2965 
2966     qemu_purge_queued_packets(nc);
2967 
2968     virtio_del_queue(vdev, index * 2);
2969     if (q->tx_timer) {
2970         timer_free(q->tx_timer);
2971         q->tx_timer = NULL;
2972     } else {
2973         qemu_bh_delete(q->tx_bh);
2974         q->tx_bh = NULL;
2975     }
2976     q->tx_waiting = 0;
2977     virtio_del_queue(vdev, index * 2 + 1);
2978 }
2979 
2980 static void virtio_net_change_num_queue_pairs(VirtIONet *n, int new_max_queue_pairs)
2981 {
2982     VirtIODevice *vdev = VIRTIO_DEVICE(n);
2983     int old_num_queues = virtio_get_num_queues(vdev);
2984     int new_num_queues = new_max_queue_pairs * 2 + 1;
2985     int i;
2986 
2987     assert(old_num_queues >= 3);
2988     assert(old_num_queues % 2 == 1);
2989 
2990     if (old_num_queues == new_num_queues) {
2991         return;
2992     }
2993 
2994     /*
2995      * We always need to remove and add ctrl vq if
2996      * old_num_queues != new_num_queues. Remove ctrl_vq first,
2997      * and then we only enter one of the following two loops.
2998      */
2999     virtio_del_queue(vdev, old_num_queues - 1);
3000 
3001     for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
3002         /* new_num_queues < old_num_queues */
3003         virtio_net_del_queue(n, i / 2);
3004     }
3005 
3006     for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
3007         /* new_num_queues > old_num_queues */
3008         virtio_net_add_queue(n, i / 2);
3009     }
3010 
3011     /* add ctrl_vq last */
3012     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
3013 }
3014 
3015 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
3016 {
3017     int max = multiqueue ? n->max_queue_pairs : 1;
3018 
3019     n->multiqueue = multiqueue;
3020     virtio_net_change_num_queue_pairs(n, max);
3021 
3022     virtio_net_set_queue_pairs(n);
3023 }
3024 
3025 static int virtio_net_post_load_device(void *opaque, int version_id)
3026 {
3027     VirtIONet *n = opaque;
3028     VirtIODevice *vdev = VIRTIO_DEVICE(n);
3029     int i, link_down;
3030 
3031     trace_virtio_net_post_load_device();
3032     virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
3033                                virtio_vdev_has_feature(vdev,
3034                                                        VIRTIO_F_VERSION_1),
3035                                virtio_vdev_has_feature(vdev,
3036                                                        VIRTIO_NET_F_HASH_REPORT));
3037 
3038     /* MAC_TABLE_ENTRIES may be different from the saved image */
3039     if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
3040         n->mac_table.in_use = 0;
3041     }
3042 
3043     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
3044         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
3045     }
3046 
3047     /*
3048      * curr_guest_offloads will be later overwritten by the
3049      * virtio_set_features_nocheck call done from the virtio_load.
3050      * Here we make sure it is preserved and restored accordingly
3051      * in the virtio_net_post_load_virtio callback.
3052      */
3053     n->saved_guest_offloads = n->curr_guest_offloads;
3054 
3055     virtio_net_set_queue_pairs(n);
3056 
3057     /* Find the first multicast entry in the saved MAC filter */
3058     for (i = 0; i < n->mac_table.in_use; i++) {
3059         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
3060             break;
3061         }
3062     }
3063     n->mac_table.first_multi = i;
3064 
3065     /* nc.link_down can't be migrated, so infer link_down according
3066      * to link status bit in n->status */
3067     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
3068     for (i = 0; i < n->max_queue_pairs; i++) {
3069         qemu_get_subqueue(n->nic, i)->link_down = link_down;
3070     }
3071 
3072     if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
3073         virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
3074         qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
3075                                   QEMU_CLOCK_VIRTUAL,
3076                                   virtio_net_announce_timer, n);
3077         if (n->announce_timer.round) {
3078             timer_mod(n->announce_timer.tm,
3079                       qemu_clock_get_ms(n->announce_timer.type));
3080         } else {
3081             qemu_announce_timer_del(&n->announce_timer, false);
3082         }
3083     }
3084 
3085     virtio_net_commit_rss_config(n);
3086     return 0;
3087 }
3088 
3089 static int virtio_net_post_load_virtio(VirtIODevice *vdev)
3090 {
3091     VirtIONet *n = VIRTIO_NET(vdev);
3092     /*
3093      * The actual needed state is now in saved_guest_offloads,
3094      * see virtio_net_post_load_device for detail.
3095      * Restore it back and apply the desired offloads.
3096      */
3097     n->curr_guest_offloads = n->saved_guest_offloads;
3098     if (peer_has_vnet_hdr(n)) {
3099         virtio_net_apply_guest_offloads(n);
3100     }
3101 
3102     return 0;
3103 }
3104 
3105 /* tx_waiting field of a VirtIONetQueue */
3106 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = {
3107     .name = "virtio-net-queue-tx_waiting",
3108     .fields = (const VMStateField[]) {
3109         VMSTATE_UINT32(tx_waiting, VirtIONetQueue),
3110         VMSTATE_END_OF_LIST()
3111    },
3112 };
3113 
3114 static bool max_queue_pairs_gt_1(void *opaque, int version_id)
3115 {
3116     return VIRTIO_NET(opaque)->max_queue_pairs > 1;
3117 }
3118 
3119 static bool has_ctrl_guest_offloads(void *opaque, int version_id)
3120 {
3121     return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque),
3122                                    VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3123 }
3124 
3125 static bool mac_table_fits(void *opaque, int version_id)
3126 {
3127     return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES;
3128 }
3129 
3130 static bool mac_table_doesnt_fit(void *opaque, int version_id)
3131 {
3132     return !mac_table_fits(opaque, version_id);
3133 }
3134 
3135 /* This temporary type is shared by all the WITH_TMP methods
3136  * although only some fields are used by each.
3137  */
3138 struct VirtIONetMigTmp {
3139     VirtIONet      *parent;
3140     VirtIONetQueue *vqs_1;
3141     uint16_t        curr_queue_pairs_1;
3142     uint8_t         has_ufo;
3143     uint32_t        has_vnet_hdr;
3144 };
3145 
3146 /* The 2nd and subsequent tx_waiting flags are loaded later than
3147  * the 1st entry in the queue_pairs and only if there's more than one
3148  * entry.  We use the tmp mechanism to calculate a temporary
3149  * pointer and count and also validate the count.
3150  */
3151 
3152 static int virtio_net_tx_waiting_pre_save(void *opaque)
3153 {
3154     struct VirtIONetMigTmp *tmp = opaque;
3155 
3156     tmp->vqs_1 = tmp->parent->vqs + 1;
3157     tmp->curr_queue_pairs_1 = tmp->parent->curr_queue_pairs - 1;
3158     if (tmp->parent->curr_queue_pairs == 0) {
3159         tmp->curr_queue_pairs_1 = 0;
3160     }
3161 
3162     return 0;
3163 }
3164 
3165 static int virtio_net_tx_waiting_pre_load(void *opaque)
3166 {
3167     struct VirtIONetMigTmp *tmp = opaque;
3168 
3169     /* Reuse the pointer setup from save */
3170     virtio_net_tx_waiting_pre_save(opaque);
3171 
3172     if (tmp->parent->curr_queue_pairs > tmp->parent->max_queue_pairs) {
3173         error_report("virtio-net: curr_queue_pairs %x > max_queue_pairs %x",
3174             tmp->parent->curr_queue_pairs, tmp->parent->max_queue_pairs);
3175 
3176         return -EINVAL;
3177     }
3178 
3179     return 0; /* all good */
3180 }
3181 
3182 static const VMStateDescription vmstate_virtio_net_tx_waiting = {
3183     .name      = "virtio-net-tx_waiting",
3184     .pre_load  = virtio_net_tx_waiting_pre_load,
3185     .pre_save  = virtio_net_tx_waiting_pre_save,
3186     .fields    = (const VMStateField[]) {
3187         VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp,
3188                                      curr_queue_pairs_1,
3189                                      vmstate_virtio_net_queue_tx_waiting,
3190                                      struct VirtIONetQueue),
3191         VMSTATE_END_OF_LIST()
3192     },
3193 };
3194 
3195 /* the 'has_ufo' flag is just tested; if the incoming stream has the
3196  * flag set we need to check that we have it
3197  */
3198 static int virtio_net_ufo_post_load(void *opaque, int version_id)
3199 {
3200     struct VirtIONetMigTmp *tmp = opaque;
3201 
3202     if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) {
3203         error_report("virtio-net: saved image requires TUN_F_UFO support");
3204         return -EINVAL;
3205     }
3206 
3207     return 0;
3208 }
3209 
3210 static int virtio_net_ufo_pre_save(void *opaque)
3211 {
3212     struct VirtIONetMigTmp *tmp = opaque;
3213 
3214     tmp->has_ufo = tmp->parent->has_ufo;
3215 
3216     return 0;
3217 }
3218 
3219 static const VMStateDescription vmstate_virtio_net_has_ufo = {
3220     .name      = "virtio-net-ufo",
3221     .post_load = virtio_net_ufo_post_load,
3222     .pre_save  = virtio_net_ufo_pre_save,
3223     .fields    = (const VMStateField[]) {
3224         VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp),
3225         VMSTATE_END_OF_LIST()
3226     },
3227 };
3228 
3229 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
3230  * flag set we need to check that we have it
3231  */
3232 static int virtio_net_vnet_post_load(void *opaque, int version_id)
3233 {
3234     struct VirtIONetMigTmp *tmp = opaque;
3235 
3236     if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) {
3237         error_report("virtio-net: saved image requires vnet_hdr=on");
3238         return -EINVAL;
3239     }
3240 
3241     return 0;
3242 }
3243 
3244 static int virtio_net_vnet_pre_save(void *opaque)
3245 {
3246     struct VirtIONetMigTmp *tmp = opaque;
3247 
3248     tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr;
3249 
3250     return 0;
3251 }
3252 
3253 static const VMStateDescription vmstate_virtio_net_has_vnet = {
3254     .name      = "virtio-net-vnet",
3255     .post_load = virtio_net_vnet_post_load,
3256     .pre_save  = virtio_net_vnet_pre_save,
3257     .fields    = (const VMStateField[]) {
3258         VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp),
3259         VMSTATE_END_OF_LIST()
3260     },
3261 };
3262 
3263 static bool virtio_net_rss_needed(void *opaque)
3264 {
3265     return VIRTIO_NET(opaque)->rss_data.enabled;
3266 }
3267 
3268 static const VMStateDescription vmstate_virtio_net_rss = {
3269     .name      = "virtio-net-device/rss",
3270     .version_id = 1,
3271     .minimum_version_id = 1,
3272     .needed = virtio_net_rss_needed,
3273     .fields = (const VMStateField[]) {
3274         VMSTATE_BOOL(rss_data.enabled, VirtIONet),
3275         VMSTATE_BOOL(rss_data.redirect, VirtIONet),
3276         VMSTATE_BOOL(rss_data.populate_hash, VirtIONet),
3277         VMSTATE_UINT32(rss_data.hash_types, VirtIONet),
3278         VMSTATE_UINT16(rss_data.indirections_len, VirtIONet),
3279         VMSTATE_UINT16(rss_data.default_queue, VirtIONet),
3280         VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet,
3281                             VIRTIO_NET_RSS_MAX_KEY_SIZE),
3282         VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet,
3283                                     rss_data.indirections_len, 0,
3284                                     vmstate_info_uint16, uint16_t),
3285         VMSTATE_END_OF_LIST()
3286     },
3287 };
3288 
3289 static const VMStateDescription vmstate_virtio_net_device = {
3290     .name = "virtio-net-device",
3291     .version_id = VIRTIO_NET_VM_VERSION,
3292     .minimum_version_id = VIRTIO_NET_VM_VERSION,
3293     .post_load = virtio_net_post_load_device,
3294     .fields = (const VMStateField[]) {
3295         VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN),
3296         VMSTATE_STRUCT_POINTER(vqs, VirtIONet,
3297                                vmstate_virtio_net_queue_tx_waiting,
3298                                VirtIONetQueue),
3299         VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet),
3300         VMSTATE_UINT16(status, VirtIONet),
3301         VMSTATE_UINT8(promisc, VirtIONet),
3302         VMSTATE_UINT8(allmulti, VirtIONet),
3303         VMSTATE_UINT32(mac_table.in_use, VirtIONet),
3304 
3305         /* Guarded pair: If it fits we load it, else we throw it away
3306          * - can happen if source has a larger MAC table.; post-load
3307          *  sets flags in this case.
3308          */
3309         VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet,
3310                                 0, mac_table_fits, mac_table.in_use,
3311                                  ETH_ALEN),
3312         VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0,
3313                                      mac_table.in_use, ETH_ALEN),
3314 
3315         /* Note: This is an array of uint32's that's always been saved as a
3316          * buffer; hold onto your endiannesses; it's actually used as a bitmap
3317          * but based on the uint.
3318          */
3319         VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3),
3320         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3321                          vmstate_virtio_net_has_vnet),
3322         VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet),
3323         VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet),
3324         VMSTATE_UINT8(alluni, VirtIONet),
3325         VMSTATE_UINT8(nomulti, VirtIONet),
3326         VMSTATE_UINT8(nouni, VirtIONet),
3327         VMSTATE_UINT8(nobcast, VirtIONet),
3328         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3329                          vmstate_virtio_net_has_ufo),
3330         VMSTATE_SINGLE_TEST(max_queue_pairs, VirtIONet, max_queue_pairs_gt_1, 0,
3331                             vmstate_info_uint16_equal, uint16_t),
3332         VMSTATE_UINT16_TEST(curr_queue_pairs, VirtIONet, max_queue_pairs_gt_1),
3333         VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3334                          vmstate_virtio_net_tx_waiting),
3335         VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet,
3336                             has_ctrl_guest_offloads),
3337         VMSTATE_END_OF_LIST()
3338     },
3339     .subsections = (const VMStateDescription * const []) {
3340         &vmstate_virtio_net_rss,
3341         NULL
3342     }
3343 };
3344 
3345 static NetClientInfo net_virtio_info = {
3346     .type = NET_CLIENT_DRIVER_NIC,
3347     .size = sizeof(NICState),
3348     .can_receive = virtio_net_can_receive,
3349     .receive = virtio_net_receive,
3350     .link_status_changed = virtio_net_set_link_status,
3351     .query_rx_filter = virtio_net_query_rxfilter,
3352     .announce = virtio_net_announce,
3353 };
3354 
3355 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
3356 {
3357     VirtIONet *n = VIRTIO_NET(vdev);
3358     NetClientState *nc;
3359     assert(n->vhost_started);
3360     if (!n->multiqueue && idx == 2) {
3361         /* Must guard against invalid features and bogus queue index
3362          * from being set by malicious guest, or penetrated through
3363          * buggy migration stream.
3364          */
3365         if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
3366             qemu_log_mask(LOG_GUEST_ERROR,
3367                           "%s: bogus vq index ignored\n", __func__);
3368             return false;
3369         }
3370         nc = qemu_get_subqueue(n->nic, n->max_queue_pairs);
3371     } else {
3372         nc = qemu_get_subqueue(n->nic, vq2q(idx));
3373     }
3374     /*
3375      * Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1
3376      * as the macro of configure interrupt's IDX, If this driver does not
3377      * support, the function will return false
3378      */
3379 
3380     if (idx == VIRTIO_CONFIG_IRQ_IDX) {
3381         return vhost_net_config_pending(get_vhost_net(nc->peer));
3382     }
3383     return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
3384 }
3385 
3386 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
3387                                            bool mask)
3388 {
3389     VirtIONet *n = VIRTIO_NET(vdev);
3390     NetClientState *nc;
3391     assert(n->vhost_started);
3392     if (!n->multiqueue && idx == 2) {
3393         /* Must guard against invalid features and bogus queue index
3394          * from being set by malicious guest, or penetrated through
3395          * buggy migration stream.
3396          */
3397         if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
3398             qemu_log_mask(LOG_GUEST_ERROR,
3399                           "%s: bogus vq index ignored\n", __func__);
3400             return;
3401         }
3402         nc = qemu_get_subqueue(n->nic, n->max_queue_pairs);
3403     } else {
3404         nc = qemu_get_subqueue(n->nic, vq2q(idx));
3405     }
3406     /*
3407      *Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1
3408      * as the macro of configure interrupt's IDX, If this driver does not
3409      * support, the function will return
3410      */
3411 
3412     if (idx == VIRTIO_CONFIG_IRQ_IDX) {
3413         vhost_net_config_mask(get_vhost_net(nc->peer), vdev, mask);
3414         return;
3415     }
3416     vhost_net_virtqueue_mask(get_vhost_net(nc->peer), vdev, idx, mask);
3417 }
3418 
3419 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
3420 {
3421     virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
3422 
3423     n->config_size = virtio_get_config_size(&cfg_size_params, host_features);
3424 }
3425 
3426 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
3427                                    const char *type)
3428 {
3429     /*
3430      * The name can be NULL, the netclient name will be type.x.
3431      */
3432     assert(type != NULL);
3433 
3434     g_free(n->netclient_name);
3435     g_free(n->netclient_type);
3436     n->netclient_name = g_strdup(name);
3437     n->netclient_type = g_strdup(type);
3438 }
3439 
3440 static bool failover_unplug_primary(VirtIONet *n, DeviceState *dev)
3441 {
3442     HotplugHandler *hotplug_ctrl;
3443     PCIDevice *pci_dev;
3444     Error *err = NULL;
3445 
3446     hotplug_ctrl = qdev_get_hotplug_handler(dev);
3447     if (hotplug_ctrl) {
3448         pci_dev = PCI_DEVICE(dev);
3449         pci_dev->partially_hotplugged = true;
3450         hotplug_handler_unplug_request(hotplug_ctrl, dev, &err);
3451         if (err) {
3452             error_report_err(err);
3453             return false;
3454         }
3455     } else {
3456         return false;
3457     }
3458     return true;
3459 }
3460 
3461 static bool failover_replug_primary(VirtIONet *n, DeviceState *dev,
3462                                     Error **errp)
3463 {
3464     Error *err = NULL;
3465     HotplugHandler *hotplug_ctrl;
3466     PCIDevice *pdev = PCI_DEVICE(dev);
3467     BusState *primary_bus;
3468 
3469     if (!pdev->partially_hotplugged) {
3470         return true;
3471     }
3472     primary_bus = dev->parent_bus;
3473     if (!primary_bus) {
3474         error_setg(errp, "virtio_net: couldn't find primary bus");
3475         return false;
3476     }
3477     qdev_set_parent_bus(dev, primary_bus, &error_abort);
3478     qatomic_set(&n->failover_primary_hidden, false);
3479     hotplug_ctrl = qdev_get_hotplug_handler(dev);
3480     if (hotplug_ctrl) {
3481         hotplug_handler_pre_plug(hotplug_ctrl, dev, &err);
3482         if (err) {
3483             goto out;
3484         }
3485         hotplug_handler_plug(hotplug_ctrl, dev, &err);
3486     }
3487     pdev->partially_hotplugged = false;
3488 
3489 out:
3490     error_propagate(errp, err);
3491     return !err;
3492 }
3493 
3494 static void virtio_net_handle_migration_primary(VirtIONet *n, MigrationEvent *e)
3495 {
3496     bool should_be_hidden;
3497     Error *err = NULL;
3498     DeviceState *dev = failover_find_primary_device(n);
3499 
3500     if (!dev) {
3501         return;
3502     }
3503 
3504     should_be_hidden = qatomic_read(&n->failover_primary_hidden);
3505 
3506     if (e->type == MIG_EVENT_PRECOPY_SETUP && !should_be_hidden) {
3507         if (failover_unplug_primary(n, dev)) {
3508             vmstate_unregister(VMSTATE_IF(dev), qdev_get_vmsd(dev), dev);
3509             qapi_event_send_unplug_primary(dev->id);
3510             qatomic_set(&n->failover_primary_hidden, true);
3511         } else {
3512             warn_report("couldn't unplug primary device");
3513         }
3514     } else if (e->type == MIG_EVENT_PRECOPY_FAILED) {
3515         /* We already unplugged the device let's plug it back */
3516         if (!failover_replug_primary(n, dev, &err)) {
3517             if (err) {
3518                 error_report_err(err);
3519             }
3520         }
3521     }
3522 }
3523 
3524 static int virtio_net_migration_state_notifier(NotifierWithReturn *notifier,
3525                                                MigrationEvent *e, Error **errp)
3526 {
3527     VirtIONet *n = container_of(notifier, VirtIONet, migration_state);
3528     virtio_net_handle_migration_primary(n, e);
3529     return 0;
3530 }
3531 
3532 static bool failover_hide_primary_device(DeviceListener *listener,
3533                                          const QDict *device_opts,
3534                                          bool from_json,
3535                                          Error **errp)
3536 {
3537     VirtIONet *n = container_of(listener, VirtIONet, primary_listener);
3538     const char *standby_id;
3539 
3540     if (!device_opts) {
3541         return false;
3542     }
3543 
3544     if (!qdict_haskey(device_opts, "failover_pair_id")) {
3545         return false;
3546     }
3547 
3548     if (!qdict_haskey(device_opts, "id")) {
3549         error_setg(errp, "Device with failover_pair_id needs to have id");
3550         return false;
3551     }
3552 
3553     standby_id = qdict_get_str(device_opts, "failover_pair_id");
3554     if (g_strcmp0(standby_id, n->netclient_name) != 0) {
3555         return false;
3556     }
3557 
3558     /*
3559      * The hide helper can be called several times for a given device.
3560      * Check there is only one primary for a virtio-net device but
3561      * don't duplicate the qdict several times if it's called for the same
3562      * device.
3563      */
3564     if (n->primary_opts) {
3565         const char *old, *new;
3566         /* devices with failover_pair_id always have an id */
3567         old = qdict_get_str(n->primary_opts, "id");
3568         new = qdict_get_str(device_opts, "id");
3569         if (strcmp(old, new) != 0) {
3570             error_setg(errp, "Cannot attach more than one primary device to "
3571                        "'%s': '%s' and '%s'", n->netclient_name, old, new);
3572             return false;
3573         }
3574     } else {
3575         n->primary_opts = qdict_clone_shallow(device_opts);
3576         n->primary_opts_from_json = from_json;
3577     }
3578 
3579     /* failover_primary_hidden is set during feature negotiation */
3580     return qatomic_read(&n->failover_primary_hidden);
3581 }
3582 
3583 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
3584 {
3585     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3586     VirtIONet *n = VIRTIO_NET(dev);
3587     NetClientState *nc;
3588     int i;
3589 
3590     if (n->net_conf.mtu) {
3591         n->host_features |= (1ULL << VIRTIO_NET_F_MTU);
3592     }
3593 
3594     if (n->net_conf.duplex_str) {
3595         if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) {
3596             n->net_conf.duplex = DUPLEX_HALF;
3597         } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) {
3598             n->net_conf.duplex = DUPLEX_FULL;
3599         } else {
3600             error_setg(errp, "'duplex' must be 'half' or 'full'");
3601             return;
3602         }
3603         n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3604     } else {
3605         n->net_conf.duplex = DUPLEX_UNKNOWN;
3606     }
3607 
3608     if (n->net_conf.speed < SPEED_UNKNOWN) {
3609         error_setg(errp, "'speed' must be between 0 and INT_MAX");
3610         return;
3611     }
3612     if (n->net_conf.speed >= 0) {
3613         n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3614     }
3615 
3616     if (n->failover) {
3617         n->primary_listener.hide_device = failover_hide_primary_device;
3618         qatomic_set(&n->failover_primary_hidden, true);
3619         device_listener_register(&n->primary_listener);
3620         migration_add_notifier(&n->migration_state,
3621                                virtio_net_migration_state_notifier);
3622         n->host_features |= (1ULL << VIRTIO_NET_F_STANDBY);
3623     }
3624 
3625     virtio_net_set_config_size(n, n->host_features);
3626     virtio_init(vdev, VIRTIO_ID_NET, n->config_size);
3627 
3628     /*
3629      * We set a lower limit on RX queue size to what it always was.
3630      * Guests that want a smaller ring can always resize it without
3631      * help from us (using virtio 1 and up).
3632      */
3633     if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
3634         n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
3635         !is_power_of_2(n->net_conf.rx_queue_size)) {
3636         error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
3637                    "must be a power of 2 between %d and %d.",
3638                    n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
3639                    VIRTQUEUE_MAX_SIZE);
3640         virtio_cleanup(vdev);
3641         return;
3642     }
3643 
3644     if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE ||
3645         n->net_conf.tx_queue_size > virtio_net_max_tx_queue_size(n) ||
3646         !is_power_of_2(n->net_conf.tx_queue_size)) {
3647         error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), "
3648                    "must be a power of 2 between %d and %d",
3649                    n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE,
3650                    virtio_net_max_tx_queue_size(n));
3651         virtio_cleanup(vdev);
3652         return;
3653     }
3654 
3655     n->max_ncs = MAX(n->nic_conf.peers.queues, 1);
3656 
3657     /*
3658      * Figure out the datapath queue pairs since the backend could
3659      * provide control queue via peers as well.
3660      */
3661     if (n->nic_conf.peers.queues) {
3662         for (i = 0; i < n->max_ncs; i++) {
3663             if (n->nic_conf.peers.ncs[i]->is_datapath) {
3664                 ++n->max_queue_pairs;
3665             }
3666         }
3667     }
3668     n->max_queue_pairs = MAX(n->max_queue_pairs, 1);
3669 
3670     if (n->max_queue_pairs * 2 + 1 > VIRTIO_QUEUE_MAX) {
3671         error_setg(errp, "Invalid number of queue pairs (= %" PRIu32 "), "
3672                    "must be a positive integer less than %d.",
3673                    n->max_queue_pairs, (VIRTIO_QUEUE_MAX - 1) / 2);
3674         virtio_cleanup(vdev);
3675         return;
3676     }
3677     n->vqs = g_new0(VirtIONetQueue, n->max_queue_pairs);
3678     n->curr_queue_pairs = 1;
3679     n->tx_timeout = n->net_conf.txtimer;
3680 
3681     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
3682                        && strcmp(n->net_conf.tx, "bh")) {
3683         warn_report("virtio-net: "
3684                     "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
3685                     n->net_conf.tx);
3686         error_printf("Defaulting to \"bh\"");
3687     }
3688 
3689     n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n),
3690                                     n->net_conf.tx_queue_size);
3691 
3692     virtio_net_add_queue(n, 0);
3693 
3694     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
3695     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
3696     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
3697     n->status = VIRTIO_NET_S_LINK_UP;
3698     qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
3699                               QEMU_CLOCK_VIRTUAL,
3700                               virtio_net_announce_timer, n);
3701     n->announce_timer.round = 0;
3702 
3703     if (n->netclient_type) {
3704         /*
3705          * Happen when virtio_net_set_netclient_name has been called.
3706          */
3707         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3708                               n->netclient_type, n->netclient_name,
3709                               &dev->mem_reentrancy_guard, n);
3710     } else {
3711         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3712                               object_get_typename(OBJECT(dev)), dev->id,
3713                               &dev->mem_reentrancy_guard, n);
3714     }
3715 
3716     for (i = 0; i < n->max_queue_pairs; i++) {
3717         n->nic->ncs[i].do_not_pad = true;
3718     }
3719 
3720     peer_test_vnet_hdr(n);
3721     if (peer_has_vnet_hdr(n)) {
3722         n->host_hdr_len = sizeof(struct virtio_net_hdr);
3723     } else {
3724         n->host_hdr_len = 0;
3725     }
3726 
3727     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
3728 
3729     n->vqs[0].tx_waiting = 0;
3730     n->tx_burst = n->net_conf.txburst;
3731     virtio_net_set_mrg_rx_bufs(n, 0, 0, 0);
3732     n->promisc = 1; /* for compatibility */
3733 
3734     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
3735 
3736     n->vlans = g_malloc0(MAX_VLAN >> 3);
3737 
3738     nc = qemu_get_queue(n->nic);
3739     nc->rxfilter_notify_enabled = 1;
3740 
3741    if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
3742         struct virtio_net_config netcfg = {};
3743         memcpy(&netcfg.mac, &n->nic_conf.macaddr, ETH_ALEN);
3744         vhost_net_set_config(get_vhost_net(nc->peer),
3745             (uint8_t *)&netcfg, 0, ETH_ALEN, VHOST_SET_CONFIG_TYPE_FRONTEND);
3746     }
3747     QTAILQ_INIT(&n->rsc_chains);
3748     n->qdev = dev;
3749 
3750     net_rx_pkt_init(&n->rx_pkt);
3751 
3752     if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) {
3753         virtio_net_load_ebpf(n);
3754     }
3755 }
3756 
3757 static void virtio_net_device_unrealize(DeviceState *dev)
3758 {
3759     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3760     VirtIONet *n = VIRTIO_NET(dev);
3761     int i, max_queue_pairs;
3762 
3763     if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) {
3764         virtio_net_unload_ebpf(n);
3765     }
3766 
3767     /* This will stop vhost backend if appropriate. */
3768     virtio_net_set_status(vdev, 0);
3769 
3770     g_free(n->netclient_name);
3771     n->netclient_name = NULL;
3772     g_free(n->netclient_type);
3773     n->netclient_type = NULL;
3774 
3775     g_free(n->mac_table.macs);
3776     g_free(n->vlans);
3777 
3778     if (n->failover) {
3779         qobject_unref(n->primary_opts);
3780         device_listener_unregister(&n->primary_listener);
3781         migration_remove_notifier(&n->migration_state);
3782     } else {
3783         assert(n->primary_opts == NULL);
3784     }
3785 
3786     max_queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
3787     for (i = 0; i < max_queue_pairs; i++) {
3788         virtio_net_del_queue(n, i);
3789     }
3790     /* delete also control vq */
3791     virtio_del_queue(vdev, max_queue_pairs * 2);
3792     qemu_announce_timer_del(&n->announce_timer, false);
3793     g_free(n->vqs);
3794     qemu_del_nic(n->nic);
3795     virtio_net_rsc_cleanup(n);
3796     g_free(n->rss_data.indirections_table);
3797     net_rx_pkt_uninit(n->rx_pkt);
3798     virtio_cleanup(vdev);
3799 }
3800 
3801 static void virtio_net_reset(VirtIODevice *vdev)
3802 {
3803     VirtIONet *n = VIRTIO_NET(vdev);
3804     int i;
3805 
3806     /* Reset back to compatibility mode */
3807     n->promisc = 1;
3808     n->allmulti = 0;
3809     n->alluni = 0;
3810     n->nomulti = 0;
3811     n->nouni = 0;
3812     n->nobcast = 0;
3813     /* multiqueue is disabled by default */
3814     n->curr_queue_pairs = 1;
3815     timer_del(n->announce_timer.tm);
3816     n->announce_timer.round = 0;
3817     n->status &= ~VIRTIO_NET_S_ANNOUNCE;
3818 
3819     /* Flush any MAC and VLAN filter table state */
3820     n->mac_table.in_use = 0;
3821     n->mac_table.first_multi = 0;
3822     n->mac_table.multi_overflow = 0;
3823     n->mac_table.uni_overflow = 0;
3824     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
3825     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
3826     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
3827     memset(n->vlans, 0, MAX_VLAN >> 3);
3828 
3829     /* Flush any async TX */
3830     for (i = 0;  i < n->max_queue_pairs; i++) {
3831         flush_or_purge_queued_packets(qemu_get_subqueue(n->nic, i));
3832     }
3833 
3834     virtio_net_disable_rss(n);
3835 }
3836 
3837 static void virtio_net_instance_init(Object *obj)
3838 {
3839     VirtIONet *n = VIRTIO_NET(obj);
3840 
3841     /*
3842      * The default config_size is sizeof(struct virtio_net_config).
3843      * Can be overridden with virtio_net_set_config_size.
3844      */
3845     n->config_size = sizeof(struct virtio_net_config);
3846     device_add_bootindex_property(obj, &n->nic_conf.bootindex,
3847                                   "bootindex", "/ethernet-phy@0",
3848                                   DEVICE(n));
3849 
3850     ebpf_rss_init(&n->ebpf_rss);
3851 }
3852 
3853 static int virtio_net_pre_save(void *opaque)
3854 {
3855     VirtIONet *n = opaque;
3856 
3857     /* At this point, backend must be stopped, otherwise
3858      * it might keep writing to memory. */
3859     assert(!n->vhost_started);
3860 
3861     return 0;
3862 }
3863 
3864 static bool primary_unplug_pending(void *opaque)
3865 {
3866     DeviceState *dev = opaque;
3867     DeviceState *primary;
3868     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3869     VirtIONet *n = VIRTIO_NET(vdev);
3870 
3871     if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
3872         return false;
3873     }
3874     primary = failover_find_primary_device(n);
3875     return primary ? primary->pending_deleted_event : false;
3876 }
3877 
3878 static bool dev_unplug_pending(void *opaque)
3879 {
3880     DeviceState *dev = opaque;
3881     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3882 
3883     return vdc->primary_unplug_pending(dev);
3884 }
3885 
3886 static struct vhost_dev *virtio_net_get_vhost(VirtIODevice *vdev)
3887 {
3888     VirtIONet *n = VIRTIO_NET(vdev);
3889     NetClientState *nc = qemu_get_queue(n->nic);
3890     struct vhost_net *net = get_vhost_net(nc->peer);
3891     return &net->dev;
3892 }
3893 
3894 static const VMStateDescription vmstate_virtio_net = {
3895     .name = "virtio-net",
3896     .minimum_version_id = VIRTIO_NET_VM_VERSION,
3897     .version_id = VIRTIO_NET_VM_VERSION,
3898     .fields = (const VMStateField[]) {
3899         VMSTATE_VIRTIO_DEVICE,
3900         VMSTATE_END_OF_LIST()
3901     },
3902     .pre_save = virtio_net_pre_save,
3903     .dev_unplug_pending = dev_unplug_pending,
3904 };
3905 
3906 static Property virtio_net_properties[] = {
3907     DEFINE_PROP_BIT64("csum", VirtIONet, host_features,
3908                     VIRTIO_NET_F_CSUM, true),
3909     DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features,
3910                     VIRTIO_NET_F_GUEST_CSUM, true),
3911     DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
3912     DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features,
3913                     VIRTIO_NET_F_GUEST_TSO4, true),
3914     DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features,
3915                     VIRTIO_NET_F_GUEST_TSO6, true),
3916     DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features,
3917                     VIRTIO_NET_F_GUEST_ECN, true),
3918     DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features,
3919                     VIRTIO_NET_F_GUEST_UFO, true),
3920     DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features,
3921                     VIRTIO_NET_F_GUEST_ANNOUNCE, true),
3922     DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features,
3923                     VIRTIO_NET_F_HOST_TSO4, true),
3924     DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features,
3925                     VIRTIO_NET_F_HOST_TSO6, true),
3926     DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features,
3927                     VIRTIO_NET_F_HOST_ECN, true),
3928     DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features,
3929                     VIRTIO_NET_F_HOST_UFO, true),
3930     DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features,
3931                     VIRTIO_NET_F_MRG_RXBUF, true),
3932     DEFINE_PROP_BIT64("status", VirtIONet, host_features,
3933                     VIRTIO_NET_F_STATUS, true),
3934     DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features,
3935                     VIRTIO_NET_F_CTRL_VQ, true),
3936     DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features,
3937                     VIRTIO_NET_F_CTRL_RX, true),
3938     DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features,
3939                     VIRTIO_NET_F_CTRL_VLAN, true),
3940     DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features,
3941                     VIRTIO_NET_F_CTRL_RX_EXTRA, true),
3942     DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features,
3943                     VIRTIO_NET_F_CTRL_MAC_ADDR, true),
3944     DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features,
3945                     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
3946     DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
3947     DEFINE_PROP_BIT64("rss", VirtIONet, host_features,
3948                     VIRTIO_NET_F_RSS, false),
3949     DEFINE_PROP_BIT64("hash", VirtIONet, host_features,
3950                     VIRTIO_NET_F_HASH_REPORT, false),
3951     DEFINE_PROP_ARRAY("ebpf-rss-fds", VirtIONet, nr_ebpf_rss_fds,
3952                       ebpf_rss_fds, qdev_prop_string, char*),
3953     DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features,
3954                     VIRTIO_NET_F_RSC_EXT, false),
3955     DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout,
3956                        VIRTIO_NET_RSC_DEFAULT_INTERVAL),
3957     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
3958     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
3959                        TX_TIMER_INTERVAL),
3960     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
3961     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
3962     DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
3963                        VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
3964     DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size,
3965                        VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE),
3966     DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
3967     DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend,
3968                      true),
3969     DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN),
3970     DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str),
3971     DEFINE_PROP_BOOL("failover", VirtIONet, failover, false),
3972     DEFINE_PROP_BIT64("guest_uso4", VirtIONet, host_features,
3973                       VIRTIO_NET_F_GUEST_USO4, true),
3974     DEFINE_PROP_BIT64("guest_uso6", VirtIONet, host_features,
3975                       VIRTIO_NET_F_GUEST_USO6, true),
3976     DEFINE_PROP_BIT64("host_uso", VirtIONet, host_features,
3977                       VIRTIO_NET_F_HOST_USO, true),
3978     DEFINE_PROP_END_OF_LIST(),
3979 };
3980 
3981 static void virtio_net_class_init(ObjectClass *klass, void *data)
3982 {
3983     DeviceClass *dc = DEVICE_CLASS(klass);
3984     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3985 
3986     device_class_set_props(dc, virtio_net_properties);
3987     dc->vmsd = &vmstate_virtio_net;
3988     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
3989     vdc->realize = virtio_net_device_realize;
3990     vdc->unrealize = virtio_net_device_unrealize;
3991     vdc->get_config = virtio_net_get_config;
3992     vdc->set_config = virtio_net_set_config;
3993     vdc->get_features = virtio_net_get_features;
3994     vdc->set_features = virtio_net_set_features;
3995     vdc->bad_features = virtio_net_bad_features;
3996     vdc->reset = virtio_net_reset;
3997     vdc->queue_reset = virtio_net_queue_reset;
3998     vdc->queue_enable = virtio_net_queue_enable;
3999     vdc->set_status = virtio_net_set_status;
4000     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
4001     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
4002     vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
4003     vdc->post_load = virtio_net_post_load_virtio;
4004     vdc->vmsd = &vmstate_virtio_net_device;
4005     vdc->primary_unplug_pending = primary_unplug_pending;
4006     vdc->get_vhost = virtio_net_get_vhost;
4007     vdc->toggle_device_iotlb = vhost_toggle_device_iotlb;
4008 }
4009 
4010 static const TypeInfo virtio_net_info = {
4011     .name = TYPE_VIRTIO_NET,
4012     .parent = TYPE_VIRTIO_DEVICE,
4013     .instance_size = sizeof(VirtIONet),
4014     .instance_init = virtio_net_instance_init,
4015     .class_init = virtio_net_class_init,
4016 };
4017 
4018 static void virtio_register_types(void)
4019 {
4020     type_register_static(&virtio_net_info);
4021 }
4022 
4023 type_init(virtio_register_types)
4024