2 * Virtio Network Device
4 * Copyright IBM, Corp. 2007
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
14 #include "qemu/osdep.h"
16 #include "hw/virtio/virtio.h"
18 #include "net/checksum.h"
20 #include "qemu/error-report.h"
21 #include "qemu/timer.h"
22 #include "hw/virtio/virtio-net.h"
23 #include "net/vhost_net.h"
24 #include "hw/virtio/virtio-bus.h"
25 #include "qapi/qmp/qjson.h"
26 #include "qapi-event.h"
27 #include "hw/virtio/virtio-access.h"
29 #define VIRTIO_NET_VM_VERSION 11
31 #define MAC_TABLE_ENTRIES 64
32 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */
34 /* previously fixed value */
35 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
36 /* for now, only allow larger queues; with virtio-1, guest can downsize */
37 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
40 * Calculate the number of bytes up to and including the given 'field' of
43 #define endof(container, field) \
44 (offsetof(container, field) + sizeof(((container *)0)->field))
46 typedef struct VirtIOFeature {
51 static VirtIOFeature feature_sizes[] = {
52 {.flags = 1 << VIRTIO_NET_F_MAC,
53 .end = endof(struct virtio_net_config, mac)},
54 {.flags = 1 << VIRTIO_NET_F_STATUS,
55 .end = endof(struct virtio_net_config, status)},
56 {.flags = 1 << VIRTIO_NET_F_MQ,
57 .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
58 {.flags = 1 << VIRTIO_NET_F_MTU,
59 .end = endof(struct virtio_net_config, mtu)},
63 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
65 VirtIONet *n = qemu_get_nic_opaque(nc);
67 return &n->vqs[nc->queue_index];
70 static int vq2q(int queue_index)
72 return queue_index / 2;
76 * - we could suppress RX interrupt if we were so inclined.
79 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
81 VirtIONet *n = VIRTIO_NET(vdev);
82 struct virtio_net_config netcfg;
84 virtio_stw_p(vdev, &netcfg.status, n->status);
85 virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues);
86 virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
87 memcpy(netcfg.mac, n->mac, ETH_ALEN);
88 memcpy(config, &netcfg, n->config_size);
91 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
93 VirtIONet *n = VIRTIO_NET(vdev);
94 struct virtio_net_config netcfg = {};
96 memcpy(&netcfg, config, n->config_size);
98 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
99 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
100 memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
101 memcpy(n->mac, netcfg.mac, ETH_ALEN);
102 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
106 static bool virtio_net_started(VirtIONet *n, uint8_t status)
108 VirtIODevice *vdev = VIRTIO_DEVICE(n);
109 return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
110 (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
113 static void virtio_net_announce_timer(void *opaque)
115 VirtIONet *n = opaque;
116 VirtIODevice *vdev = VIRTIO_DEVICE(n);
118 n->announce_counter--;
119 n->status |= VIRTIO_NET_S_ANNOUNCE;
120 virtio_notify_config(vdev);
123 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
125 VirtIODevice *vdev = VIRTIO_DEVICE(n);
126 NetClientState *nc = qemu_get_queue(n->nic);
127 int queues = n->multiqueue ? n->max_queues : 1;
129 if (!get_vhost_net(nc->peer)) {
133 if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
134 !!n->vhost_started) {
137 if (!n->vhost_started) {
140 if (n->needs_vnet_hdr_swap) {
141 error_report("backend does not support %s vnet headers; "
142 "falling back on userspace virtio",
143 virtio_is_big_endian(vdev) ? "BE" : "LE");
147 /* Any packets outstanding? Purge them to avoid touching rings
148 * when vhost is running.
150 for (i = 0; i < queues; i++) {
151 NetClientState *qnc = qemu_get_subqueue(n->nic, i);
153 /* Purge both directions: TX and RX. */
154 qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
155 qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
158 if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
159 r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
161 error_report("%uBytes MTU not supported by the backend",
168 n->vhost_started = 1;
169 r = vhost_net_start(vdev, n->nic->ncs, queues);
171 error_report("unable to start vhost net: %d: "
172 "falling back on userspace virtio", -r);
173 n->vhost_started = 0;
176 vhost_net_stop(vdev, n->nic->ncs, queues);
177 n->vhost_started = 0;
181 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
182 NetClientState *peer,
185 if (virtio_is_big_endian(vdev)) {
186 return qemu_set_vnet_be(peer, enable);
188 return qemu_set_vnet_le(peer, enable);
192 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
193 int queues, bool enable)
197 for (i = 0; i < queues; i++) {
198 if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
201 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
211 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
213 VirtIODevice *vdev = VIRTIO_DEVICE(n);
214 int queues = n->multiqueue ? n->max_queues : 1;
216 if (virtio_net_started(n, status)) {
217 /* Before using the device, we tell the network backend about the
218 * endianness to use when parsing vnet headers. If the backend
219 * can't do it, we fallback onto fixing the headers in the core
222 n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
224 } else if (virtio_net_started(n, vdev->status)) {
225 /* After using the device, we need to reset the network backend to
226 * the default (guest native endianness), otherwise the guest may
227 * lose network connectivity if it is rebooted into a different
230 virtio_net_set_vnet_endian(vdev, n->nic->ncs, queues, false);
234 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq)
236 unsigned int dropped = virtqueue_drop_all(vq);
238 virtio_notify(vdev, vq);
242 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
244 VirtIONet *n = VIRTIO_NET(vdev);
247 uint8_t queue_status;
249 virtio_net_vnet_endian_status(n, status);
250 virtio_net_vhost_status(n, status);
252 for (i = 0; i < n->max_queues; i++) {
253 NetClientState *ncs = qemu_get_subqueue(n->nic, i);
257 if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
260 queue_status = status;
263 virtio_net_started(n, queue_status) && !n->vhost_started;
266 qemu_flush_queued_packets(ncs);
269 if (!q->tx_waiting) {
275 timer_mod(q->tx_timer,
276 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
278 qemu_bh_schedule(q->tx_bh);
282 timer_del(q->tx_timer);
284 qemu_bh_cancel(q->tx_bh);
286 if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 &&
287 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK)) {
288 /* if tx is waiting we are likely have some packets in tx queue
289 * and disabled notification */
291 virtio_queue_set_notification(q->tx_vq, 1);
292 virtio_net_drop_tx_queue_data(vdev, q->tx_vq);
298 static void virtio_net_set_link_status(NetClientState *nc)
300 VirtIONet *n = qemu_get_nic_opaque(nc);
301 VirtIODevice *vdev = VIRTIO_DEVICE(n);
302 uint16_t old_status = n->status;
305 n->status &= ~VIRTIO_NET_S_LINK_UP;
307 n->status |= VIRTIO_NET_S_LINK_UP;
309 if (n->status != old_status)
310 virtio_notify_config(vdev);
312 virtio_net_set_status(vdev, vdev->status);
315 static void rxfilter_notify(NetClientState *nc)
317 VirtIONet *n = qemu_get_nic_opaque(nc);
319 if (nc->rxfilter_notify_enabled) {
320 gchar *path = object_get_canonical_path(OBJECT(n->qdev));
321 qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
322 n->netclient_name, path, &error_abort);
325 /* disable event notification to avoid events flooding */
326 nc->rxfilter_notify_enabled = 0;
330 static intList *get_vlan_table(VirtIONet *n)
332 intList *list, *entry;
336 for (i = 0; i < MAX_VLAN >> 5; i++) {
337 for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
338 if (n->vlans[i] & (1U << j)) {
339 entry = g_malloc0(sizeof(*entry));
340 entry->value = (i << 5) + j;
350 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
352 VirtIONet *n = qemu_get_nic_opaque(nc);
353 VirtIODevice *vdev = VIRTIO_DEVICE(n);
355 strList *str_list, *entry;
358 info = g_malloc0(sizeof(*info));
359 info->name = g_strdup(nc->name);
360 info->promiscuous = n->promisc;
363 info->unicast = RX_STATE_NONE;
364 } else if (n->alluni) {
365 info->unicast = RX_STATE_ALL;
367 info->unicast = RX_STATE_NORMAL;
371 info->multicast = RX_STATE_NONE;
372 } else if (n->allmulti) {
373 info->multicast = RX_STATE_ALL;
375 info->multicast = RX_STATE_NORMAL;
378 info->broadcast_allowed = n->nobcast;
379 info->multicast_overflow = n->mac_table.multi_overflow;
380 info->unicast_overflow = n->mac_table.uni_overflow;
382 info->main_mac = qemu_mac_strdup_printf(n->mac);
385 for (i = 0; i < n->mac_table.first_multi; i++) {
386 entry = g_malloc0(sizeof(*entry));
387 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
388 entry->next = str_list;
391 info->unicast_table = str_list;
394 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
395 entry = g_malloc0(sizeof(*entry));
396 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
397 entry->next = str_list;
400 info->multicast_table = str_list;
401 info->vlan_table = get_vlan_table(n);
403 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
404 info->vlan = RX_STATE_ALL;
405 } else if (!info->vlan_table) {
406 info->vlan = RX_STATE_NONE;
408 info->vlan = RX_STATE_NORMAL;
411 /* enable event notification after query */
412 nc->rxfilter_notify_enabled = 1;
417 static void virtio_net_reset(VirtIODevice *vdev)
419 VirtIONet *n = VIRTIO_NET(vdev);
421 /* Reset back to compatibility mode */
428 /* multiqueue is disabled by default */
430 timer_del(n->announce_timer);
431 n->announce_counter = 0;
432 n->status &= ~VIRTIO_NET_S_ANNOUNCE;
434 /* Flush any MAC and VLAN filter table state */
435 n->mac_table.in_use = 0;
436 n->mac_table.first_multi = 0;
437 n->mac_table.multi_overflow = 0;
438 n->mac_table.uni_overflow = 0;
439 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
440 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
441 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
442 memset(n->vlans, 0, MAX_VLAN >> 3);
445 static void peer_test_vnet_hdr(VirtIONet *n)
447 NetClientState *nc = qemu_get_queue(n->nic);
452 n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
455 static int peer_has_vnet_hdr(VirtIONet *n)
457 return n->has_vnet_hdr;
460 static int peer_has_ufo(VirtIONet *n)
462 if (!peer_has_vnet_hdr(n))
465 n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
470 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
476 n->mergeable_rx_bufs = mergeable_rx_bufs;
479 n->guest_hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
481 n->guest_hdr_len = n->mergeable_rx_bufs ?
482 sizeof(struct virtio_net_hdr_mrg_rxbuf) :
483 sizeof(struct virtio_net_hdr);
486 for (i = 0; i < n->max_queues; i++) {
487 nc = qemu_get_subqueue(n->nic, i);
489 if (peer_has_vnet_hdr(n) &&
490 qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
491 qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
492 n->host_hdr_len = n->guest_hdr_len;
497 static int peer_attach(VirtIONet *n, int index)
499 NetClientState *nc = qemu_get_subqueue(n->nic, index);
505 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
506 vhost_set_vring_enable(nc->peer, 1);
509 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
513 return tap_enable(nc->peer);
516 static int peer_detach(VirtIONet *n, int index)
518 NetClientState *nc = qemu_get_subqueue(n->nic, index);
524 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
525 vhost_set_vring_enable(nc->peer, 0);
528 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
532 return tap_disable(nc->peer);
535 static void virtio_net_set_queues(VirtIONet *n)
540 if (n->nic->peer_deleted) {
544 for (i = 0; i < n->max_queues; i++) {
545 if (i < n->curr_queues) {
546 r = peer_attach(n, i);
549 r = peer_detach(n, i);
555 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
557 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
560 VirtIONet *n = VIRTIO_NET(vdev);
561 NetClientState *nc = qemu_get_queue(n->nic);
563 /* Firstly sync all virtio-net possible supported features */
564 features |= n->host_features;
566 virtio_add_feature(&features, VIRTIO_NET_F_MAC);
568 if (!peer_has_vnet_hdr(n)) {
569 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
570 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
571 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
572 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
574 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
575 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
576 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
577 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
580 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
581 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
582 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
585 if (!get_vhost_net(nc->peer)) {
588 return vhost_net_get_features(get_vhost_net(nc->peer), features);
591 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
593 uint64_t features = 0;
595 /* Linux kernel 2.6.25. It understood MAC (as everyone must),
597 virtio_add_feature(&features, VIRTIO_NET_F_MAC);
598 virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
599 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
600 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
601 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
606 static void virtio_net_apply_guest_offloads(VirtIONet *n)
608 qemu_set_offload(qemu_get_queue(n->nic)->peer,
609 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
610 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
611 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
612 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
613 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
616 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
618 static const uint64_t guest_offloads_mask =
619 (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
620 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
621 (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
622 (1ULL << VIRTIO_NET_F_GUEST_ECN) |
623 (1ULL << VIRTIO_NET_F_GUEST_UFO);
625 return guest_offloads_mask & features;
628 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
630 VirtIODevice *vdev = VIRTIO_DEVICE(n);
631 return virtio_net_guest_offloads_by_features(vdev->guest_features);
634 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
636 VirtIONet *n = VIRTIO_NET(vdev);
639 virtio_net_set_multiqueue(n,
640 virtio_has_feature(features, VIRTIO_NET_F_MQ));
642 virtio_net_set_mrg_rx_bufs(n,
643 virtio_has_feature(features,
644 VIRTIO_NET_F_MRG_RXBUF),
645 virtio_has_feature(features,
646 VIRTIO_F_VERSION_1));
648 if (n->has_vnet_hdr) {
649 n->curr_guest_offloads =
650 virtio_net_guest_offloads_by_features(features);
651 virtio_net_apply_guest_offloads(n);
654 for (i = 0; i < n->max_queues; i++) {
655 NetClientState *nc = qemu_get_subqueue(n->nic, i);
657 if (!get_vhost_net(nc->peer)) {
660 vhost_net_ack_features(get_vhost_net(nc->peer), features);
663 if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
664 memset(n->vlans, 0, MAX_VLAN >> 3);
666 memset(n->vlans, 0xff, MAX_VLAN >> 3);
670 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
671 struct iovec *iov, unsigned int iov_cnt)
675 NetClientState *nc = qemu_get_queue(n->nic);
677 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
678 if (s != sizeof(on)) {
679 return VIRTIO_NET_ERR;
682 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
684 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
686 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
688 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
690 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
692 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
695 return VIRTIO_NET_ERR;
700 return VIRTIO_NET_OK;
703 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
704 struct iovec *iov, unsigned int iov_cnt)
706 VirtIODevice *vdev = VIRTIO_DEVICE(n);
710 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
711 return VIRTIO_NET_ERR;
714 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
715 if (s != sizeof(offloads)) {
716 return VIRTIO_NET_ERR;
719 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
720 uint64_t supported_offloads;
722 if (!n->has_vnet_hdr) {
723 return VIRTIO_NET_ERR;
726 supported_offloads = virtio_net_supported_guest_offloads(n);
727 if (offloads & ~supported_offloads) {
728 return VIRTIO_NET_ERR;
731 n->curr_guest_offloads = offloads;
732 virtio_net_apply_guest_offloads(n);
734 return VIRTIO_NET_OK;
736 return VIRTIO_NET_ERR;
740 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
741 struct iovec *iov, unsigned int iov_cnt)
743 VirtIODevice *vdev = VIRTIO_DEVICE(n);
744 struct virtio_net_ctrl_mac mac_data;
746 NetClientState *nc = qemu_get_queue(n->nic);
748 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
749 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
750 return VIRTIO_NET_ERR;
752 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
753 assert(s == sizeof(n->mac));
754 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
757 return VIRTIO_NET_OK;
760 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
761 return VIRTIO_NET_ERR;
766 uint8_t uni_overflow = 0;
767 uint8_t multi_overflow = 0;
768 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
770 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
771 sizeof(mac_data.entries));
772 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
773 if (s != sizeof(mac_data.entries)) {
776 iov_discard_front(&iov, &iov_cnt, s);
778 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
782 if (mac_data.entries <= MAC_TABLE_ENTRIES) {
783 s = iov_to_buf(iov, iov_cnt, 0, macs,
784 mac_data.entries * ETH_ALEN);
785 if (s != mac_data.entries * ETH_ALEN) {
788 in_use += mac_data.entries;
793 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
795 first_multi = in_use;
797 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
798 sizeof(mac_data.entries));
799 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
800 if (s != sizeof(mac_data.entries)) {
804 iov_discard_front(&iov, &iov_cnt, s);
806 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
810 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
811 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
812 mac_data.entries * ETH_ALEN);
813 if (s != mac_data.entries * ETH_ALEN) {
816 in_use += mac_data.entries;
821 n->mac_table.in_use = in_use;
822 n->mac_table.first_multi = first_multi;
823 n->mac_table.uni_overflow = uni_overflow;
824 n->mac_table.multi_overflow = multi_overflow;
825 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
829 return VIRTIO_NET_OK;
833 return VIRTIO_NET_ERR;
836 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
837 struct iovec *iov, unsigned int iov_cnt)
839 VirtIODevice *vdev = VIRTIO_DEVICE(n);
842 NetClientState *nc = qemu_get_queue(n->nic);
844 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
845 vid = virtio_lduw_p(vdev, &vid);
846 if (s != sizeof(vid)) {
847 return VIRTIO_NET_ERR;
851 return VIRTIO_NET_ERR;
853 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
854 n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
855 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
856 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
858 return VIRTIO_NET_ERR;
862 return VIRTIO_NET_OK;
865 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
866 struct iovec *iov, unsigned int iov_cnt)
868 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
869 n->status & VIRTIO_NET_S_ANNOUNCE) {
870 n->status &= ~VIRTIO_NET_S_ANNOUNCE;
871 if (n->announce_counter) {
872 timer_mod(n->announce_timer,
873 qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
874 self_announce_delay(n->announce_counter));
876 return VIRTIO_NET_OK;
878 return VIRTIO_NET_ERR;
882 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
883 struct iovec *iov, unsigned int iov_cnt)
885 VirtIODevice *vdev = VIRTIO_DEVICE(n);
886 struct virtio_net_ctrl_mq mq;
890 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
891 if (s != sizeof(mq)) {
892 return VIRTIO_NET_ERR;
895 if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
896 return VIRTIO_NET_ERR;
899 queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
901 if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
902 queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
903 queues > n->max_queues ||
905 return VIRTIO_NET_ERR;
908 n->curr_queues = queues;
909 /* stop the backend before changing the number of queues to avoid handling a
911 virtio_net_set_status(vdev, vdev->status);
912 virtio_net_set_queues(n);
914 return VIRTIO_NET_OK;
917 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
919 VirtIONet *n = VIRTIO_NET(vdev);
920 struct virtio_net_ctrl_hdr ctrl;
921 virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
922 VirtQueueElement *elem;
924 struct iovec *iov, *iov2;
925 unsigned int iov_cnt;
928 elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
932 if (iov_size(elem->in_sg, elem->in_num) < sizeof(status) ||
933 iov_size(elem->out_sg, elem->out_num) < sizeof(ctrl)) {
934 virtio_error(vdev, "virtio-net ctrl missing headers");
935 virtqueue_detach_element(vq, elem, 0);
940 iov_cnt = elem->out_num;
941 iov2 = iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num);
942 s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
943 iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
944 if (s != sizeof(ctrl)) {
945 status = VIRTIO_NET_ERR;
946 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
947 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
948 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
949 status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
950 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
951 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
952 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
953 status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
954 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
955 status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
956 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
957 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
960 s = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, sizeof(status));
961 assert(s == sizeof(status));
963 virtqueue_push(vq, elem, sizeof(status));
964 virtio_notify(vdev, vq);
972 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
974 VirtIONet *n = VIRTIO_NET(vdev);
975 int queue_index = vq2q(virtio_get_queue_index(vq));
977 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
980 static int virtio_net_can_receive(NetClientState *nc)
982 VirtIONet *n = qemu_get_nic_opaque(nc);
983 VirtIODevice *vdev = VIRTIO_DEVICE(n);
984 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
986 if (!vdev->vm_running) {
990 if (nc->queue_index >= n->curr_queues) {
994 if (!virtio_queue_ready(q->rx_vq) ||
995 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1002 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1004 VirtIONet *n = q->n;
1005 if (virtio_queue_empty(q->rx_vq) ||
1006 (n->mergeable_rx_bufs &&
1007 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1008 virtio_queue_set_notification(q->rx_vq, 1);
1010 /* To avoid a race condition where the guest has made some buffers
1011 * available after the above check but before notification was
1012 * enabled, check for available buffers again.
1014 if (virtio_queue_empty(q->rx_vq) ||
1015 (n->mergeable_rx_bufs &&
1016 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1021 virtio_queue_set_notification(q->rx_vq, 0);
1025 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1027 virtio_tswap16s(vdev, &hdr->hdr_len);
1028 virtio_tswap16s(vdev, &hdr->gso_size);
1029 virtio_tswap16s(vdev, &hdr->csum_start);
1030 virtio_tswap16s(vdev, &hdr->csum_offset);
1033 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1034 * it never finds out that the packets don't have valid checksums. This
1035 * causes dhclient to get upset. Fedora's carried a patch for ages to
1036 * fix this with Xen but it hasn't appeared in an upstream release of
1039 * To avoid breaking existing guests, we catch udp packets and add
1040 * checksums. This is terrible but it's better than hacking the guest
1043 * N.B. if we introduce a zero-copy API, this operation is no longer free so
1044 * we should provide a mechanism to disable it to avoid polluting the host
1047 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1048 uint8_t *buf, size_t size)
1050 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1051 (size > 27 && size < 1500) && /* normal sized MTU */
1052 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1053 (buf[23] == 17) && /* ip.protocol == UDP */
1054 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1055 net_checksum_calculate(buf, size);
1056 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1060 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1061 const void *buf, size_t size)
1063 if (n->has_vnet_hdr) {
1064 /* FIXME this cast is evil */
1065 void *wbuf = (void *)buf;
1066 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1067 size - n->host_hdr_len);
1069 if (n->needs_vnet_hdr_swap) {
1070 virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1072 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1074 struct virtio_net_hdr hdr = {
1076 .gso_type = VIRTIO_NET_HDR_GSO_NONE
1078 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1082 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1084 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1085 static const uint8_t vlan[] = {0x81, 0x00};
1086 uint8_t *ptr = (uint8_t *)buf;
1092 ptr += n->host_hdr_len;
1094 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1095 int vid = lduw_be_p(ptr + 14) & 0xfff;
1096 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1100 if (ptr[0] & 1) { // multicast
1101 if (!memcmp(ptr, bcast, sizeof(bcast))) {
1103 } else if (n->nomulti) {
1105 } else if (n->allmulti || n->mac_table.multi_overflow) {
1109 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1110 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1117 } else if (n->alluni || n->mac_table.uni_overflow) {
1119 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1123 for (i = 0; i < n->mac_table.first_multi; i++) {
1124 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1133 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
1135 VirtIONet *n = qemu_get_nic_opaque(nc);
1136 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1137 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1138 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1139 struct virtio_net_hdr_mrg_rxbuf mhdr;
1140 unsigned mhdr_cnt = 0;
1141 size_t offset, i, guest_offset;
1143 if (!virtio_net_can_receive(nc)) {
1147 /* hdr_len refers to the header we supply to the guest */
1148 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1152 if (!receive_filter(n, buf, size))
1157 while (offset < size) {
1158 VirtQueueElement *elem;
1160 const struct iovec *sg;
1164 elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1167 virtio_error(vdev, "virtio-net unexpected empty queue: "
1168 "i %zd mergeable %d offset %zd, size %zd, "
1169 "guest hdr len %zd, host hdr len %zd "
1170 "guest features 0x%" PRIx64,
1171 i, n->mergeable_rx_bufs, offset, size,
1172 n->guest_hdr_len, n->host_hdr_len,
1173 vdev->guest_features);
1178 if (elem->in_num < 1) {
1180 "virtio-net receive queue contains no in buffers");
1181 virtqueue_detach_element(q->rx_vq, elem, 0);
1188 assert(offset == 0);
1189 if (n->mergeable_rx_bufs) {
1190 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1192 offsetof(typeof(mhdr), num_buffers),
1193 sizeof(mhdr.num_buffers));
1196 receive_header(n, sg, elem->in_num, buf, size);
1197 offset = n->host_hdr_len;
1198 total += n->guest_hdr_len;
1199 guest_offset = n->guest_hdr_len;
1204 /* copy in packet. ugh */
1205 len = iov_from_buf(sg, elem->in_num, guest_offset,
1206 buf + offset, size - offset);
1209 /* If buffers can't be merged, at this point we
1210 * must have consumed the complete packet.
1211 * Otherwise, drop it. */
1212 if (!n->mergeable_rx_bufs && offset < size) {
1213 virtqueue_unpop(q->rx_vq, elem, total);
1218 /* signal other side */
1219 virtqueue_fill(q->rx_vq, elem, total, i++);
1224 virtio_stw_p(vdev, &mhdr.num_buffers, i);
1225 iov_from_buf(mhdr_sg, mhdr_cnt,
1227 &mhdr.num_buffers, sizeof mhdr.num_buffers);
1230 virtqueue_flush(q->rx_vq, i);
1231 virtio_notify(vdev, q->rx_vq);
1236 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1238 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1240 VirtIONet *n = qemu_get_nic_opaque(nc);
1241 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1242 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1244 virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
1245 virtio_notify(vdev, q->tx_vq);
1247 g_free(q->async_tx.elem);
1248 q->async_tx.elem = NULL;
1250 virtio_queue_set_notification(q->tx_vq, 1);
1251 virtio_net_flush_tx(q);
1255 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
1257 VirtIONet *n = q->n;
1258 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1259 VirtQueueElement *elem;
1260 int32_t num_packets = 0;
1261 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
1262 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1266 if (q->async_tx.elem) {
1267 virtio_queue_set_notification(q->tx_vq, 0);
1273 unsigned int out_num;
1274 struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
1275 struct virtio_net_hdr_mrg_rxbuf mhdr;
1277 elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
1282 out_num = elem->out_num;
1283 out_sg = elem->out_sg;
1285 virtio_error(vdev, "virtio-net header not in first element");
1286 virtqueue_detach_element(q->tx_vq, elem, 0);
1291 if (n->has_vnet_hdr) {
1292 if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
1294 virtio_error(vdev, "virtio-net header incorrect");
1295 virtqueue_detach_element(q->tx_vq, elem, 0);
1299 if (n->needs_vnet_hdr_swap) {
1300 virtio_net_hdr_swap(vdev, (void *) &mhdr);
1301 sg2[0].iov_base = &mhdr;
1302 sg2[0].iov_len = n->guest_hdr_len;
1303 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
1305 n->guest_hdr_len, -1);
1306 if (out_num == VIRTQUEUE_MAX_SIZE) {
1314 * If host wants to see the guest header as is, we can
1315 * pass it on unchanged. Otherwise, copy just the parts
1316 * that host is interested in.
1318 assert(n->host_hdr_len <= n->guest_hdr_len);
1319 if (n->host_hdr_len != n->guest_hdr_len) {
1320 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
1322 0, n->host_hdr_len);
1323 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
1325 n->guest_hdr_len, -1);
1330 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
1331 out_sg, out_num, virtio_net_tx_complete);
1333 virtio_queue_set_notification(q->tx_vq, 0);
1334 q->async_tx.elem = elem;
1339 virtqueue_push(q->tx_vq, elem, 0);
1340 virtio_notify(vdev, q->tx_vq);
1343 if (++num_packets >= n->tx_burst) {
1350 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
1352 VirtIONet *n = VIRTIO_NET(vdev);
1353 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1355 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
1356 virtio_net_drop_tx_queue_data(vdev, vq);
1360 /* This happens when device was stopped but VCPU wasn't. */
1361 if (!vdev->vm_running) {
1366 if (q->tx_waiting) {
1367 virtio_queue_set_notification(vq, 1);
1368 timer_del(q->tx_timer);
1370 if (virtio_net_flush_tx(q) == -EINVAL) {
1374 timer_mod(q->tx_timer,
1375 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
1377 virtio_queue_set_notification(vq, 0);
1381 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1383 VirtIONet *n = VIRTIO_NET(vdev);
1384 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1386 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
1387 virtio_net_drop_tx_queue_data(vdev, vq);
1391 if (unlikely(q->tx_waiting)) {
1395 /* This happens when device was stopped but VCPU wasn't. */
1396 if (!vdev->vm_running) {
1399 virtio_queue_set_notification(vq, 0);
1400 qemu_bh_schedule(q->tx_bh);
1403 static void virtio_net_tx_timer(void *opaque)
1405 VirtIONetQueue *q = opaque;
1406 VirtIONet *n = q->n;
1407 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1408 /* This happens when device was stopped but BH wasn't. */
1409 if (!vdev->vm_running) {
1410 /* Make sure tx waiting is set, so we'll run when restarted. */
1411 assert(q->tx_waiting);
1417 /* Just in case the driver is not ready on more */
1418 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1422 virtio_queue_set_notification(q->tx_vq, 1);
1423 virtio_net_flush_tx(q);
1426 static void virtio_net_tx_bh(void *opaque)
1428 VirtIONetQueue *q = opaque;
1429 VirtIONet *n = q->n;
1430 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1433 /* This happens when device was stopped but BH wasn't. */
1434 if (!vdev->vm_running) {
1435 /* Make sure tx waiting is set, so we'll run when restarted. */
1436 assert(q->tx_waiting);
1442 /* Just in case the driver is not ready on more */
1443 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1447 ret = virtio_net_flush_tx(q);
1448 if (ret == -EBUSY || ret == -EINVAL) {
1449 return; /* Notification re-enable handled by tx_complete or device
1453 /* If we flush a full burst of packets, assume there are
1454 * more coming and immediately reschedule */
1455 if (ret >= n->tx_burst) {
1456 qemu_bh_schedule(q->tx_bh);
1461 /* If less than a full burst, re-enable notification and flush
1462 * anything that may have come in while we weren't looking. If
1463 * we find something, assume the guest is still active and reschedule */
1464 virtio_queue_set_notification(q->tx_vq, 1);
1465 ret = virtio_net_flush_tx(q);
1466 if (ret == -EINVAL) {
1468 } else if (ret > 0) {
1469 virtio_queue_set_notification(q->tx_vq, 0);
1470 qemu_bh_schedule(q->tx_bh);
1475 static void virtio_net_add_queue(VirtIONet *n, int index)
1477 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1479 n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
1480 virtio_net_handle_rx);
1481 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1482 n->vqs[index].tx_vq =
1483 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1484 n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
1485 virtio_net_tx_timer,
1488 n->vqs[index].tx_vq =
1489 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1490 n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
1493 n->vqs[index].tx_waiting = 0;
1494 n->vqs[index].n = n;
1497 static void virtio_net_del_queue(VirtIONet *n, int index)
1499 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1500 VirtIONetQueue *q = &n->vqs[index];
1501 NetClientState *nc = qemu_get_subqueue(n->nic, index);
1503 qemu_purge_queued_packets(nc);
1505 virtio_del_queue(vdev, index * 2);
1507 timer_del(q->tx_timer);
1508 timer_free(q->tx_timer);
1510 qemu_bh_delete(q->tx_bh);
1512 virtio_del_queue(vdev, index * 2 + 1);
1515 static void virtio_net_change_num_queues(VirtIONet *n, int new_max_queues)
1517 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1518 int old_num_queues = virtio_get_num_queues(vdev);
1519 int new_num_queues = new_max_queues * 2 + 1;
1522 assert(old_num_queues >= 3);
1523 assert(old_num_queues % 2 == 1);
1525 if (old_num_queues == new_num_queues) {
1530 * We always need to remove and add ctrl vq if
1531 * old_num_queues != new_num_queues. Remove ctrl_vq first,
1532 * and then we only enter one of the following too loops.
1534 virtio_del_queue(vdev, old_num_queues - 1);
1536 for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
1537 /* new_num_queues < old_num_queues */
1538 virtio_net_del_queue(n, i / 2);
1541 for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
1542 /* new_num_queues > old_num_queues */
1543 virtio_net_add_queue(n, i / 2);
1546 /* add ctrl_vq last */
1547 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1550 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1552 int max = multiqueue ? n->max_queues : 1;
1554 n->multiqueue = multiqueue;
1555 virtio_net_change_num_queues(n, max);
1557 virtio_net_set_queues(n);
1560 static void virtio_net_save_device(VirtIODevice *vdev, QEMUFile *f)
1562 VirtIONet *n = VIRTIO_NET(vdev);
1565 qemu_put_buffer(f, n->mac, ETH_ALEN);
1566 qemu_put_be32(f, n->vqs[0].tx_waiting);
1567 qemu_put_be32(f, n->mergeable_rx_bufs);
1568 qemu_put_be16(f, n->status);
1569 qemu_put_byte(f, n->promisc);
1570 qemu_put_byte(f, n->allmulti);
1571 qemu_put_be32(f, n->mac_table.in_use);
1572 qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1573 qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1574 qemu_put_be32(f, n->has_vnet_hdr);
1575 qemu_put_byte(f, n->mac_table.multi_overflow);
1576 qemu_put_byte(f, n->mac_table.uni_overflow);
1577 qemu_put_byte(f, n->alluni);
1578 qemu_put_byte(f, n->nomulti);
1579 qemu_put_byte(f, n->nouni);
1580 qemu_put_byte(f, n->nobcast);
1581 qemu_put_byte(f, n->has_ufo);
1582 if (n->max_queues > 1) {
1583 qemu_put_be16(f, n->max_queues);
1584 qemu_put_be16(f, n->curr_queues);
1585 for (i = 1; i < n->curr_queues; i++) {
1586 qemu_put_be32(f, n->vqs[i].tx_waiting);
1590 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1591 qemu_put_be64(f, n->curr_guest_offloads);
1595 static int virtio_net_load_device(VirtIODevice *vdev, QEMUFile *f,
1598 VirtIONet *n = VIRTIO_NET(vdev);
1601 qemu_get_buffer(f, n->mac, ETH_ALEN);
1602 n->vqs[0].tx_waiting = qemu_get_be32(f);
1604 virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f),
1605 virtio_vdev_has_feature(vdev,
1606 VIRTIO_F_VERSION_1));
1608 n->status = qemu_get_be16(f);
1610 n->promisc = qemu_get_byte(f);
1611 n->allmulti = qemu_get_byte(f);
1613 n->mac_table.in_use = qemu_get_be32(f);
1614 /* MAC_TABLE_ENTRIES may be different from the saved image */
1615 if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1616 qemu_get_buffer(f, n->mac_table.macs,
1617 n->mac_table.in_use * ETH_ALEN);
1621 /* Overflow detected - can happen if source has a larger MAC table.
1622 * We simply set overflow flag so there's no need to maintain the
1623 * table of addresses, discard them all.
1624 * Note: 64 bit math to avoid integer overflow.
1626 for (i = 0; i < (int64_t)n->mac_table.in_use * ETH_ALEN; ++i) {
1629 n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1630 n->mac_table.in_use = 0;
1633 qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1635 if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1636 error_report("virtio-net: saved image requires vnet_hdr=on");
1640 n->mac_table.multi_overflow = qemu_get_byte(f);
1641 n->mac_table.uni_overflow = qemu_get_byte(f);
1643 n->alluni = qemu_get_byte(f);
1644 n->nomulti = qemu_get_byte(f);
1645 n->nouni = qemu_get_byte(f);
1646 n->nobcast = qemu_get_byte(f);
1648 if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1649 error_report("virtio-net: saved image requires TUN_F_UFO support");
1653 if (n->max_queues > 1) {
1654 if (n->max_queues != qemu_get_be16(f)) {
1655 error_report("virtio-net: different max_queues ");
1659 n->curr_queues = qemu_get_be16(f);
1660 if (n->curr_queues > n->max_queues) {
1661 error_report("virtio-net: curr_queues %x > max_queues %x",
1662 n->curr_queues, n->max_queues);
1665 for (i = 1; i < n->curr_queues; i++) {
1666 n->vqs[i].tx_waiting = qemu_get_be32(f);
1670 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1671 n->curr_guest_offloads = qemu_get_be64(f);
1673 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
1676 if (peer_has_vnet_hdr(n)) {
1677 virtio_net_apply_guest_offloads(n);
1680 virtio_net_set_queues(n);
1682 /* Find the first multicast entry in the saved MAC filter */
1683 for (i = 0; i < n->mac_table.in_use; i++) {
1684 if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1688 n->mac_table.first_multi = i;
1690 /* nc.link_down can't be migrated, so infer link_down according
1691 * to link status bit in n->status */
1692 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1693 for (i = 0; i < n->max_queues; i++) {
1694 qemu_get_subqueue(n->nic, i)->link_down = link_down;
1697 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
1698 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
1699 n->announce_counter = SELF_ANNOUNCE_ROUNDS;
1700 timer_mod(n->announce_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL));
1706 static NetClientInfo net_virtio_info = {
1707 .type = NET_CLIENT_DRIVER_NIC,
1708 .size = sizeof(NICState),
1709 .can_receive = virtio_net_can_receive,
1710 .receive = virtio_net_receive,
1711 .link_status_changed = virtio_net_set_link_status,
1712 .query_rx_filter = virtio_net_query_rxfilter,
1715 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1717 VirtIONet *n = VIRTIO_NET(vdev);
1718 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1719 assert(n->vhost_started);
1720 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
1723 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1726 VirtIONet *n = VIRTIO_NET(vdev);
1727 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1728 assert(n->vhost_started);
1729 vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
1733 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
1735 int i, config_size = 0;
1736 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
1738 for (i = 0; feature_sizes[i].flags != 0; i++) {
1739 if (host_features & feature_sizes[i].flags) {
1740 config_size = MAX(feature_sizes[i].end, config_size);
1743 n->config_size = config_size;
1746 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
1750 * The name can be NULL, the netclient name will be type.x.
1752 assert(type != NULL);
1754 g_free(n->netclient_name);
1755 g_free(n->netclient_type);
1756 n->netclient_name = g_strdup(name);
1757 n->netclient_type = g_strdup(type);
1760 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
1762 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1763 VirtIONet *n = VIRTIO_NET(dev);
1767 if (n->net_conf.mtu) {
1768 n->host_features |= (0x1 << VIRTIO_NET_F_MTU);
1771 virtio_net_set_config_size(n, n->host_features);
1772 virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
1775 * We set a lower limit on RX queue size to what it always was.
1776 * Guests that want a smaller ring can always resize it without
1777 * help from us (using virtio 1 and up).
1779 if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
1780 n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
1781 (n->net_conf.rx_queue_size & (n->net_conf.rx_queue_size - 1))) {
1782 error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
1783 "must be a power of 2 between %d and %d.",
1784 n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
1785 VIRTQUEUE_MAX_SIZE);
1786 virtio_cleanup(vdev);
1790 n->max_queues = MAX(n->nic_conf.peers.queues, 1);
1791 if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) {
1792 error_setg(errp, "Invalid number of queues (= %" PRIu32 "), "
1793 "must be a positive integer less than %d.",
1794 n->max_queues, (VIRTIO_QUEUE_MAX - 1) / 2);
1795 virtio_cleanup(vdev);
1798 n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1800 n->tx_timeout = n->net_conf.txtimer;
1802 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1803 && strcmp(n->net_conf.tx, "bh")) {
1804 error_report("virtio-net: "
1805 "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1807 error_report("Defaulting to \"bh\"");
1810 for (i = 0; i < n->max_queues; i++) {
1811 virtio_net_add_queue(n, i);
1814 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1815 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1816 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1817 n->status = VIRTIO_NET_S_LINK_UP;
1818 n->announce_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
1819 virtio_net_announce_timer, n);
1821 if (n->netclient_type) {
1823 * Happen when virtio_net_set_netclient_name has been called.
1825 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1826 n->netclient_type, n->netclient_name, n);
1828 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1829 object_get_typename(OBJECT(dev)), dev->id, n);
1832 peer_test_vnet_hdr(n);
1833 if (peer_has_vnet_hdr(n)) {
1834 for (i = 0; i < n->max_queues; i++) {
1835 qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1837 n->host_hdr_len = sizeof(struct virtio_net_hdr);
1839 n->host_hdr_len = 0;
1842 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1844 n->vqs[0].tx_waiting = 0;
1845 n->tx_burst = n->net_conf.txburst;
1846 virtio_net_set_mrg_rx_bufs(n, 0, 0);
1847 n->promisc = 1; /* for compatibility */
1849 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1851 n->vlans = g_malloc0(MAX_VLAN >> 3);
1853 nc = qemu_get_queue(n->nic);
1854 nc->rxfilter_notify_enabled = 1;
1859 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
1861 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1862 VirtIONet *n = VIRTIO_NET(dev);
1865 /* This will stop vhost backend if appropriate. */
1866 virtio_net_set_status(vdev, 0);
1868 g_free(n->netclient_name);
1869 n->netclient_name = NULL;
1870 g_free(n->netclient_type);
1871 n->netclient_type = NULL;
1873 g_free(n->mac_table.macs);
1876 max_queues = n->multiqueue ? n->max_queues : 1;
1877 for (i = 0; i < max_queues; i++) {
1878 virtio_net_del_queue(n, i);
1881 timer_del(n->announce_timer);
1882 timer_free(n->announce_timer);
1884 qemu_del_nic(n->nic);
1885 virtio_cleanup(vdev);
1888 static void virtio_net_instance_init(Object *obj)
1890 VirtIONet *n = VIRTIO_NET(obj);
1893 * The default config_size is sizeof(struct virtio_net_config).
1894 * Can be overriden with virtio_net_set_config_size.
1896 n->config_size = sizeof(struct virtio_net_config);
1897 device_add_bootindex_property(obj, &n->nic_conf.bootindex,
1898 "bootindex", "/ethernet-phy@0",
1902 static void virtio_net_pre_save(void *opaque)
1904 VirtIONet *n = opaque;
1906 /* At this point, backend must be stopped, otherwise
1907 * it might keep writing to memory. */
1908 assert(!n->vhost_started);
1911 static const VMStateDescription vmstate_virtio_net = {
1912 .name = "virtio-net",
1913 .minimum_version_id = VIRTIO_NET_VM_VERSION,
1914 .version_id = VIRTIO_NET_VM_VERSION,
1915 .fields = (VMStateField[]) {
1916 VMSTATE_VIRTIO_DEVICE,
1917 VMSTATE_END_OF_LIST()
1919 .pre_save = virtio_net_pre_save,
1922 static Property virtio_net_properties[] = {
1923 DEFINE_PROP_BIT("csum", VirtIONet, host_features, VIRTIO_NET_F_CSUM, true),
1924 DEFINE_PROP_BIT("guest_csum", VirtIONet, host_features,
1925 VIRTIO_NET_F_GUEST_CSUM, true),
1926 DEFINE_PROP_BIT("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
1927 DEFINE_PROP_BIT("guest_tso4", VirtIONet, host_features,
1928 VIRTIO_NET_F_GUEST_TSO4, true),
1929 DEFINE_PROP_BIT("guest_tso6", VirtIONet, host_features,
1930 VIRTIO_NET_F_GUEST_TSO6, true),
1931 DEFINE_PROP_BIT("guest_ecn", VirtIONet, host_features,
1932 VIRTIO_NET_F_GUEST_ECN, true),
1933 DEFINE_PROP_BIT("guest_ufo", VirtIONet, host_features,
1934 VIRTIO_NET_F_GUEST_UFO, true),
1935 DEFINE_PROP_BIT("guest_announce", VirtIONet, host_features,
1936 VIRTIO_NET_F_GUEST_ANNOUNCE, true),
1937 DEFINE_PROP_BIT("host_tso4", VirtIONet, host_features,
1938 VIRTIO_NET_F_HOST_TSO4, true),
1939 DEFINE_PROP_BIT("host_tso6", VirtIONet, host_features,
1940 VIRTIO_NET_F_HOST_TSO6, true),
1941 DEFINE_PROP_BIT("host_ecn", VirtIONet, host_features,
1942 VIRTIO_NET_F_HOST_ECN, true),
1943 DEFINE_PROP_BIT("host_ufo", VirtIONet, host_features,
1944 VIRTIO_NET_F_HOST_UFO, true),
1945 DEFINE_PROP_BIT("mrg_rxbuf", VirtIONet, host_features,
1946 VIRTIO_NET_F_MRG_RXBUF, true),
1947 DEFINE_PROP_BIT("status", VirtIONet, host_features,
1948 VIRTIO_NET_F_STATUS, true),
1949 DEFINE_PROP_BIT("ctrl_vq", VirtIONet, host_features,
1950 VIRTIO_NET_F_CTRL_VQ, true),
1951 DEFINE_PROP_BIT("ctrl_rx", VirtIONet, host_features,
1952 VIRTIO_NET_F_CTRL_RX, true),
1953 DEFINE_PROP_BIT("ctrl_vlan", VirtIONet, host_features,
1954 VIRTIO_NET_F_CTRL_VLAN, true),
1955 DEFINE_PROP_BIT("ctrl_rx_extra", VirtIONet, host_features,
1956 VIRTIO_NET_F_CTRL_RX_EXTRA, true),
1957 DEFINE_PROP_BIT("ctrl_mac_addr", VirtIONet, host_features,
1958 VIRTIO_NET_F_CTRL_MAC_ADDR, true),
1959 DEFINE_PROP_BIT("ctrl_guest_offloads", VirtIONet, host_features,
1960 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
1961 DEFINE_PROP_BIT("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
1962 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
1963 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
1965 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
1966 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
1967 DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
1968 VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
1969 DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
1970 DEFINE_PROP_END_OF_LIST(),
1973 static void virtio_net_class_init(ObjectClass *klass, void *data)
1975 DeviceClass *dc = DEVICE_CLASS(klass);
1976 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1978 dc->props = virtio_net_properties;
1979 dc->vmsd = &vmstate_virtio_net;
1980 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
1981 vdc->realize = virtio_net_device_realize;
1982 vdc->unrealize = virtio_net_device_unrealize;
1983 vdc->get_config = virtio_net_get_config;
1984 vdc->set_config = virtio_net_set_config;
1985 vdc->get_features = virtio_net_get_features;
1986 vdc->set_features = virtio_net_set_features;
1987 vdc->bad_features = virtio_net_bad_features;
1988 vdc->reset = virtio_net_reset;
1989 vdc->set_status = virtio_net_set_status;
1990 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
1991 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
1992 vdc->load = virtio_net_load_device;
1993 vdc->save = virtio_net_save_device;
1994 vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
1997 static const TypeInfo virtio_net_info = {
1998 .name = TYPE_VIRTIO_NET,
1999 .parent = TYPE_VIRTIO_DEVICE,
2000 .instance_size = sizeof(VirtIONet),
2001 .instance_init = virtio_net_instance_init,
2002 .class_init = virtio_net_class_init,
2005 static void virtio_register_types(void)
2007 type_register_static(&virtio_net_info);
2010 type_init(virtio_register_types)