2 * Virtio Network Device
4 * Copyright IBM, Corp. 2007
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
15 #include "hw/virtio/virtio.h"
17 #include "net/checksum.h"
19 #include "qemu/error-report.h"
20 #include "qemu/timer.h"
21 #include "hw/virtio/virtio-net.h"
22 #include "net/vhost_net.h"
23 #include "hw/virtio/virtio-bus.h"
24 #include "qapi/qmp/qjson.h"
25 #include "monitor/monitor.h"
27 #define VIRTIO_NET_VM_VERSION 11
29 #define MAC_TABLE_ENTRIES 64
30 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */
33 * Calculate the number of bytes up to and including the given 'field' of
36 #define endof(container, field) \
37 (offsetof(container, field) + sizeof(((container *)0)->field))
39 typedef struct VirtIOFeature
{
44 static VirtIOFeature feature_sizes
[] = {
45 {.flags
= 1 << VIRTIO_NET_F_MAC
,
46 .end
= endof(struct virtio_net_config
, mac
)},
47 {.flags
= 1 << VIRTIO_NET_F_STATUS
,
48 .end
= endof(struct virtio_net_config
, status
)},
49 {.flags
= 1 << VIRTIO_NET_F_MQ
,
50 .end
= endof(struct virtio_net_config
, max_virtqueue_pairs
)},
54 static VirtIONetQueue
*virtio_net_get_subqueue(NetClientState
*nc
)
56 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
58 return &n
->vqs
[nc
->queue_index
];
61 static int vq2q(int queue_index
)
63 return queue_index
/ 2;
67 * - we could suppress RX interrupt if we were so inclined.
70 static void virtio_net_get_config(VirtIODevice
*vdev
, uint8_t *config
)
72 VirtIONet
*n
= VIRTIO_NET(vdev
);
73 struct virtio_net_config netcfg
;
75 stw_p(&netcfg
.status
, n
->status
);
76 stw_p(&netcfg
.max_virtqueue_pairs
, n
->max_queues
);
77 memcpy(netcfg
.mac
, n
->mac
, ETH_ALEN
);
78 memcpy(config
, &netcfg
, n
->config_size
);
81 static void virtio_net_set_config(VirtIODevice
*vdev
, const uint8_t *config
)
83 VirtIONet
*n
= VIRTIO_NET(vdev
);
84 struct virtio_net_config netcfg
= {};
86 memcpy(&netcfg
, config
, n
->config_size
);
88 if (!(vdev
->guest_features
>> VIRTIO_NET_F_CTRL_MAC_ADDR
& 1) &&
89 memcmp(netcfg
.mac
, n
->mac
, ETH_ALEN
)) {
90 memcpy(n
->mac
, netcfg
.mac
, ETH_ALEN
);
91 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->mac
);
95 static bool virtio_net_started(VirtIONet
*n
, uint8_t status
)
97 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
98 return (status
& VIRTIO_CONFIG_S_DRIVER_OK
) &&
99 (n
->status
& VIRTIO_NET_S_LINK_UP
) && vdev
->vm_running
;
102 static void virtio_net_vhost_status(VirtIONet
*n
, uint8_t status
)
104 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
105 NetClientState
*nc
= qemu_get_queue(n
->nic
);
106 int queues
= n
->multiqueue
? n
->max_queues
: 1;
111 if (nc
->peer
->info
->type
!= NET_CLIENT_OPTIONS_KIND_TAP
) {
115 if (!tap_get_vhost_net(nc
->peer
)) {
119 if (!!n
->vhost_started
==
120 (virtio_net_started(n
, status
) && !nc
->peer
->link_down
)) {
123 if (!n
->vhost_started
) {
125 if (!vhost_net_query(tap_get_vhost_net(nc
->peer
), vdev
)) {
128 n
->vhost_started
= 1;
129 r
= vhost_net_start(vdev
, n
->nic
->ncs
, queues
);
131 error_report("unable to start vhost net: %d: "
132 "falling back on userspace virtio", -r
);
133 n
->vhost_started
= 0;
136 vhost_net_stop(vdev
, n
->nic
->ncs
, queues
);
137 n
->vhost_started
= 0;
141 static void virtio_net_set_status(struct VirtIODevice
*vdev
, uint8_t status
)
143 VirtIONet
*n
= VIRTIO_NET(vdev
);
146 uint8_t queue_status
;
148 virtio_net_vhost_status(n
, status
);
150 for (i
= 0; i
< n
->max_queues
; i
++) {
153 if ((!n
->multiqueue
&& i
!= 0) || i
>= n
->curr_queues
) {
156 queue_status
= status
;
159 if (!q
->tx_waiting
) {
163 if (virtio_net_started(n
, queue_status
) && !n
->vhost_started
) {
165 qemu_mod_timer(q
->tx_timer
,
166 qemu_get_clock_ns(vm_clock
) + n
->tx_timeout
);
168 qemu_bh_schedule(q
->tx_bh
);
172 qemu_del_timer(q
->tx_timer
);
174 qemu_bh_cancel(q
->tx_bh
);
180 static void virtio_net_set_link_status(NetClientState
*nc
)
182 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
183 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
184 uint16_t old_status
= n
->status
;
187 n
->status
&= ~VIRTIO_NET_S_LINK_UP
;
189 n
->status
|= VIRTIO_NET_S_LINK_UP
;
191 if (n
->status
!= old_status
)
192 virtio_notify_config(vdev
);
194 virtio_net_set_status(vdev
, vdev
->status
);
197 static void rxfilter_notify(NetClientState
*nc
)
200 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
202 if (nc
->rxfilter_notify_enabled
) {
203 if (n
->netclient_name
) {
204 event_data
= qobject_from_jsonf("{ 'name': %s, 'path': %s }",
206 object_get_canonical_path(OBJECT(n
->qdev
)));
208 event_data
= qobject_from_jsonf("{ 'path': %s }",
209 object_get_canonical_path(OBJECT(n
->qdev
)));
211 monitor_protocol_event(QEVENT_NIC_RX_FILTER_CHANGED
, event_data
);
212 qobject_decref(event_data
);
214 /* disable event notification to avoid events flooding */
215 nc
->rxfilter_notify_enabled
= 0;
219 static char *mac_strdup_printf(const uint8_t *mac
)
221 return g_strdup_printf("%.2x:%.2x:%.2x:%.2x:%.2x:%.2x", mac
[0],
222 mac
[1], mac
[2], mac
[3], mac
[4], mac
[5]);
225 static RxFilterInfo
*virtio_net_query_rxfilter(NetClientState
*nc
)
227 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
229 strList
*str_list
, *entry
;
230 intList
*int_list
, *int_entry
;
233 info
= g_malloc0(sizeof(*info
));
234 info
->name
= g_strdup(nc
->name
);
235 info
->promiscuous
= n
->promisc
;
238 info
->unicast
= RX_STATE_NONE
;
239 } else if (n
->alluni
) {
240 info
->unicast
= RX_STATE_ALL
;
242 info
->unicast
= RX_STATE_NORMAL
;
246 info
->multicast
= RX_STATE_NONE
;
247 } else if (n
->allmulti
) {
248 info
->multicast
= RX_STATE_ALL
;
250 info
->multicast
= RX_STATE_NORMAL
;
253 info
->broadcast_allowed
= n
->nobcast
;
254 info
->multicast_overflow
= n
->mac_table
.multi_overflow
;
255 info
->unicast_overflow
= n
->mac_table
.uni_overflow
;
257 info
->main_mac
= mac_strdup_printf(n
->mac
);
260 for (i
= 0; i
< n
->mac_table
.first_multi
; i
++) {
261 entry
= g_malloc0(sizeof(*entry
));
262 entry
->value
= mac_strdup_printf(n
->mac_table
.macs
+ i
* ETH_ALEN
);
263 entry
->next
= str_list
;
266 info
->unicast_table
= str_list
;
269 for (i
= n
->mac_table
.first_multi
; i
< n
->mac_table
.in_use
; i
++) {
270 entry
= g_malloc0(sizeof(*entry
));
271 entry
->value
= mac_strdup_printf(n
->mac_table
.macs
+ i
* ETH_ALEN
);
272 entry
->next
= str_list
;
275 info
->multicast_table
= str_list
;
278 for (i
= 0; i
< MAX_VLAN
>> 5; i
++) {
279 for (j
= 0; n
->vlans
[i
] && j
< 0x1f; j
++) {
280 if (n
->vlans
[i
] & (1U << j
)) {
281 int_entry
= g_malloc0(sizeof(*int_entry
));
282 int_entry
->value
= (i
<< 5) + j
;
283 int_entry
->next
= int_list
;
284 int_list
= int_entry
;
288 info
->vlan_table
= int_list
;
290 /* enable event notification after query */
291 nc
->rxfilter_notify_enabled
= 1;
296 static void virtio_net_reset(VirtIODevice
*vdev
)
298 VirtIONet
*n
= VIRTIO_NET(vdev
);
300 /* Reset back to compatibility mode */
307 /* multiqueue is disabled by default */
310 /* Flush any MAC and VLAN filter table state */
311 n
->mac_table
.in_use
= 0;
312 n
->mac_table
.first_multi
= 0;
313 n
->mac_table
.multi_overflow
= 0;
314 n
->mac_table
.uni_overflow
= 0;
315 memset(n
->mac_table
.macs
, 0, MAC_TABLE_ENTRIES
* ETH_ALEN
);
316 memcpy(&n
->mac
[0], &n
->nic
->conf
->macaddr
, sizeof(n
->mac
));
317 memset(n
->vlans
, 0, MAX_VLAN
>> 3);
320 static void peer_test_vnet_hdr(VirtIONet
*n
)
322 NetClientState
*nc
= qemu_get_queue(n
->nic
);
327 if (nc
->peer
->info
->type
!= NET_CLIENT_OPTIONS_KIND_TAP
) {
331 n
->has_vnet_hdr
= tap_has_vnet_hdr(nc
->peer
);
334 static int peer_has_vnet_hdr(VirtIONet
*n
)
336 return n
->has_vnet_hdr
;
339 static int peer_has_ufo(VirtIONet
*n
)
341 if (!peer_has_vnet_hdr(n
))
344 n
->has_ufo
= tap_has_ufo(qemu_get_queue(n
->nic
)->peer
);
349 static void virtio_net_set_mrg_rx_bufs(VirtIONet
*n
, int mergeable_rx_bufs
)
354 n
->mergeable_rx_bufs
= mergeable_rx_bufs
;
356 n
->guest_hdr_len
= n
->mergeable_rx_bufs
?
357 sizeof(struct virtio_net_hdr_mrg_rxbuf
) : sizeof(struct virtio_net_hdr
);
359 for (i
= 0; i
< n
->max_queues
; i
++) {
360 nc
= qemu_get_subqueue(n
->nic
, i
);
362 if (peer_has_vnet_hdr(n
) &&
363 tap_has_vnet_hdr_len(nc
->peer
, n
->guest_hdr_len
)) {
364 tap_set_vnet_hdr_len(nc
->peer
, n
->guest_hdr_len
);
365 n
->host_hdr_len
= n
->guest_hdr_len
;
370 static int peer_attach(VirtIONet
*n
, int index
)
372 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, index
);
378 if (nc
->peer
->info
->type
!= NET_CLIENT_OPTIONS_KIND_TAP
) {
382 return tap_enable(nc
->peer
);
385 static int peer_detach(VirtIONet
*n
, int index
)
387 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, index
);
393 if (nc
->peer
->info
->type
!= NET_CLIENT_OPTIONS_KIND_TAP
) {
397 return tap_disable(nc
->peer
);
400 static void virtio_net_set_queues(VirtIONet
*n
)
404 for (i
= 0; i
< n
->max_queues
; i
++) {
405 if (i
< n
->curr_queues
) {
406 assert(!peer_attach(n
, i
));
408 assert(!peer_detach(n
, i
));
413 static void virtio_net_set_multiqueue(VirtIONet
*n
, int multiqueue
);
415 static uint32_t virtio_net_get_features(VirtIODevice
*vdev
, uint32_t features
)
417 VirtIONet
*n
= VIRTIO_NET(vdev
);
418 NetClientState
*nc
= qemu_get_queue(n
->nic
);
420 features
|= (1 << VIRTIO_NET_F_MAC
);
422 if (!peer_has_vnet_hdr(n
)) {
423 features
&= ~(0x1 << VIRTIO_NET_F_CSUM
);
424 features
&= ~(0x1 << VIRTIO_NET_F_HOST_TSO4
);
425 features
&= ~(0x1 << VIRTIO_NET_F_HOST_TSO6
);
426 features
&= ~(0x1 << VIRTIO_NET_F_HOST_ECN
);
428 features
&= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM
);
429 features
&= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4
);
430 features
&= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6
);
431 features
&= ~(0x1 << VIRTIO_NET_F_GUEST_ECN
);
434 if (!peer_has_vnet_hdr(n
) || !peer_has_ufo(n
)) {
435 features
&= ~(0x1 << VIRTIO_NET_F_GUEST_UFO
);
436 features
&= ~(0x1 << VIRTIO_NET_F_HOST_UFO
);
439 if (!nc
->peer
|| nc
->peer
->info
->type
!= NET_CLIENT_OPTIONS_KIND_TAP
) {
442 if (!tap_get_vhost_net(nc
->peer
)) {
445 return vhost_net_get_features(tap_get_vhost_net(nc
->peer
), features
);
448 static uint32_t virtio_net_bad_features(VirtIODevice
*vdev
)
450 uint32_t features
= 0;
452 /* Linux kernel 2.6.25. It understood MAC (as everyone must),
454 features
|= (1 << VIRTIO_NET_F_MAC
);
455 features
|= (1 << VIRTIO_NET_F_CSUM
);
456 features
|= (1 << VIRTIO_NET_F_HOST_TSO4
);
457 features
|= (1 << VIRTIO_NET_F_HOST_TSO6
);
458 features
|= (1 << VIRTIO_NET_F_HOST_ECN
);
463 static void virtio_net_apply_guest_offloads(VirtIONet
*n
)
465 tap_set_offload(qemu_get_subqueue(n
->nic
, 0)->peer
,
466 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_CSUM
)),
467 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_TSO4
)),
468 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_TSO6
)),
469 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_ECN
)),
470 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_UFO
)));
473 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features
)
475 static const uint64_t guest_offloads_mask
=
476 (1ULL << VIRTIO_NET_F_GUEST_CSUM
) |
477 (1ULL << VIRTIO_NET_F_GUEST_TSO4
) |
478 (1ULL << VIRTIO_NET_F_GUEST_TSO6
) |
479 (1ULL << VIRTIO_NET_F_GUEST_ECN
) |
480 (1ULL << VIRTIO_NET_F_GUEST_UFO
);
482 return guest_offloads_mask
& features
;
485 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet
*n
)
487 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
488 return virtio_net_guest_offloads_by_features(vdev
->guest_features
);
491 static void virtio_net_set_features(VirtIODevice
*vdev
, uint32_t features
)
493 VirtIONet
*n
= VIRTIO_NET(vdev
);
496 virtio_net_set_multiqueue(n
, !!(features
& (1 << VIRTIO_NET_F_MQ
)));
498 virtio_net_set_mrg_rx_bufs(n
, !!(features
& (1 << VIRTIO_NET_F_MRG_RXBUF
)));
500 if (n
->has_vnet_hdr
) {
501 n
->curr_guest_offloads
=
502 virtio_net_guest_offloads_by_features(features
);
503 virtio_net_apply_guest_offloads(n
);
506 for (i
= 0; i
< n
->max_queues
; i
++) {
507 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, i
);
509 if (!nc
->peer
|| nc
->peer
->info
->type
!= NET_CLIENT_OPTIONS_KIND_TAP
) {
512 if (!tap_get_vhost_net(nc
->peer
)) {
515 vhost_net_ack_features(tap_get_vhost_net(nc
->peer
), features
);
519 static int virtio_net_handle_rx_mode(VirtIONet
*n
, uint8_t cmd
,
520 struct iovec
*iov
, unsigned int iov_cnt
)
524 NetClientState
*nc
= qemu_get_queue(n
->nic
);
526 s
= iov_to_buf(iov
, iov_cnt
, 0, &on
, sizeof(on
));
527 if (s
!= sizeof(on
)) {
528 return VIRTIO_NET_ERR
;
531 if (cmd
== VIRTIO_NET_CTRL_RX_PROMISC
) {
533 } else if (cmd
== VIRTIO_NET_CTRL_RX_ALLMULTI
) {
535 } else if (cmd
== VIRTIO_NET_CTRL_RX_ALLUNI
) {
537 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOMULTI
) {
539 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOUNI
) {
541 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOBCAST
) {
544 return VIRTIO_NET_ERR
;
549 return VIRTIO_NET_OK
;
552 static int virtio_net_handle_offloads(VirtIONet
*n
, uint8_t cmd
,
553 struct iovec
*iov
, unsigned int iov_cnt
)
555 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
559 if (!((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
) & vdev
->guest_features
)) {
560 return VIRTIO_NET_ERR
;
563 s
= iov_to_buf(iov
, iov_cnt
, 0, &offloads
, sizeof(offloads
));
564 if (s
!= sizeof(offloads
)) {
565 return VIRTIO_NET_ERR
;
568 if (cmd
== VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET
) {
569 uint64_t supported_offloads
;
571 if (!n
->has_vnet_hdr
) {
572 return VIRTIO_NET_ERR
;
575 supported_offloads
= virtio_net_supported_guest_offloads(n
);
576 if (offloads
& ~supported_offloads
) {
577 return VIRTIO_NET_ERR
;
580 n
->curr_guest_offloads
= offloads
;
581 virtio_net_apply_guest_offloads(n
);
583 return VIRTIO_NET_OK
;
585 return VIRTIO_NET_ERR
;
589 static int virtio_net_handle_mac(VirtIONet
*n
, uint8_t cmd
,
590 struct iovec
*iov
, unsigned int iov_cnt
)
592 struct virtio_net_ctrl_mac mac_data
;
594 NetClientState
*nc
= qemu_get_queue(n
->nic
);
596 if (cmd
== VIRTIO_NET_CTRL_MAC_ADDR_SET
) {
597 if (iov_size(iov
, iov_cnt
) != sizeof(n
->mac
)) {
598 return VIRTIO_NET_ERR
;
600 s
= iov_to_buf(iov
, iov_cnt
, 0, &n
->mac
, sizeof(n
->mac
));
601 assert(s
== sizeof(n
->mac
));
602 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->mac
);
605 return VIRTIO_NET_OK
;
608 if (cmd
!= VIRTIO_NET_CTRL_MAC_TABLE_SET
) {
609 return VIRTIO_NET_ERR
;
612 n
->mac_table
.in_use
= 0;
613 n
->mac_table
.first_multi
= 0;
614 n
->mac_table
.uni_overflow
= 0;
615 n
->mac_table
.multi_overflow
= 0;
616 memset(n
->mac_table
.macs
, 0, MAC_TABLE_ENTRIES
* ETH_ALEN
);
618 s
= iov_to_buf(iov
, iov_cnt
, 0, &mac_data
.entries
,
619 sizeof(mac_data
.entries
));
620 mac_data
.entries
= ldl_p(&mac_data
.entries
);
621 if (s
!= sizeof(mac_data
.entries
)) {
624 iov_discard_front(&iov
, &iov_cnt
, s
);
626 if (mac_data
.entries
* ETH_ALEN
> iov_size(iov
, iov_cnt
)) {
630 if (mac_data
.entries
<= MAC_TABLE_ENTRIES
) {
631 s
= iov_to_buf(iov
, iov_cnt
, 0, n
->mac_table
.macs
,
632 mac_data
.entries
* ETH_ALEN
);
633 if (s
!= mac_data
.entries
* ETH_ALEN
) {
636 n
->mac_table
.in_use
+= mac_data
.entries
;
638 n
->mac_table
.uni_overflow
= 1;
641 iov_discard_front(&iov
, &iov_cnt
, mac_data
.entries
* ETH_ALEN
);
643 n
->mac_table
.first_multi
= n
->mac_table
.in_use
;
645 s
= iov_to_buf(iov
, iov_cnt
, 0, &mac_data
.entries
,
646 sizeof(mac_data
.entries
));
647 mac_data
.entries
= ldl_p(&mac_data
.entries
);
648 if (s
!= sizeof(mac_data
.entries
)) {
652 iov_discard_front(&iov
, &iov_cnt
, s
);
654 if (mac_data
.entries
* ETH_ALEN
!= iov_size(iov
, iov_cnt
)) {
658 if (n
->mac_table
.in_use
+ mac_data
.entries
<= MAC_TABLE_ENTRIES
) {
659 s
= iov_to_buf(iov
, iov_cnt
, 0, n
->mac_table
.macs
,
660 mac_data
.entries
* ETH_ALEN
);
661 if (s
!= mac_data
.entries
* ETH_ALEN
) {
664 n
->mac_table
.in_use
+= mac_data
.entries
;
666 n
->mac_table
.multi_overflow
= 1;
671 return VIRTIO_NET_OK
;
675 return VIRTIO_NET_ERR
;
678 static int virtio_net_handle_vlan_table(VirtIONet
*n
, uint8_t cmd
,
679 struct iovec
*iov
, unsigned int iov_cnt
)
683 NetClientState
*nc
= qemu_get_queue(n
->nic
);
685 s
= iov_to_buf(iov
, iov_cnt
, 0, &vid
, sizeof(vid
));
687 if (s
!= sizeof(vid
)) {
688 return VIRTIO_NET_ERR
;
692 return VIRTIO_NET_ERR
;
694 if (cmd
== VIRTIO_NET_CTRL_VLAN_ADD
)
695 n
->vlans
[vid
>> 5] |= (1U << (vid
& 0x1f));
696 else if (cmd
== VIRTIO_NET_CTRL_VLAN_DEL
)
697 n
->vlans
[vid
>> 5] &= ~(1U << (vid
& 0x1f));
699 return VIRTIO_NET_ERR
;
703 return VIRTIO_NET_OK
;
706 static int virtio_net_handle_mq(VirtIONet
*n
, uint8_t cmd
,
707 struct iovec
*iov
, unsigned int iov_cnt
)
709 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
710 struct virtio_net_ctrl_mq mq
;
714 s
= iov_to_buf(iov
, iov_cnt
, 0, &mq
, sizeof(mq
));
715 if (s
!= sizeof(mq
)) {
716 return VIRTIO_NET_ERR
;
719 if (cmd
!= VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
) {
720 return VIRTIO_NET_ERR
;
723 queues
= lduw_p(&mq
.virtqueue_pairs
);
725 if (queues
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
726 queues
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
727 queues
> n
->max_queues
||
729 return VIRTIO_NET_ERR
;
732 n
->curr_queues
= queues
;
733 /* stop the backend before changing the number of queues to avoid handling a
735 virtio_net_set_status(vdev
, vdev
->status
);
736 virtio_net_set_queues(n
);
738 return VIRTIO_NET_OK
;
740 static void virtio_net_handle_ctrl(VirtIODevice
*vdev
, VirtQueue
*vq
)
742 VirtIONet
*n
= VIRTIO_NET(vdev
);
743 struct virtio_net_ctrl_hdr ctrl
;
744 virtio_net_ctrl_ack status
= VIRTIO_NET_ERR
;
745 VirtQueueElement elem
;
748 unsigned int iov_cnt
;
750 while (virtqueue_pop(vq
, &elem
)) {
751 if (iov_size(elem
.in_sg
, elem
.in_num
) < sizeof(status
) ||
752 iov_size(elem
.out_sg
, elem
.out_num
) < sizeof(ctrl
)) {
753 error_report("virtio-net ctrl missing headers");
758 iov_cnt
= elem
.out_num
;
759 s
= iov_to_buf(iov
, iov_cnt
, 0, &ctrl
, sizeof(ctrl
));
760 iov_discard_front(&iov
, &iov_cnt
, sizeof(ctrl
));
761 if (s
!= sizeof(ctrl
)) {
762 status
= VIRTIO_NET_ERR
;
763 } else if (ctrl
.class == VIRTIO_NET_CTRL_RX
) {
764 status
= virtio_net_handle_rx_mode(n
, ctrl
.cmd
, iov
, iov_cnt
);
765 } else if (ctrl
.class == VIRTIO_NET_CTRL_MAC
) {
766 status
= virtio_net_handle_mac(n
, ctrl
.cmd
, iov
, iov_cnt
);
767 } else if (ctrl
.class == VIRTIO_NET_CTRL_VLAN
) {
768 status
= virtio_net_handle_vlan_table(n
, ctrl
.cmd
, iov
, iov_cnt
);
769 } else if (ctrl
.class == VIRTIO_NET_CTRL_MQ
) {
770 status
= virtio_net_handle_mq(n
, ctrl
.cmd
, iov
, iov_cnt
);
771 } else if (ctrl
.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS
) {
772 status
= virtio_net_handle_offloads(n
, ctrl
.cmd
, iov
, iov_cnt
);
775 s
= iov_from_buf(elem
.in_sg
, elem
.in_num
, 0, &status
, sizeof(status
));
776 assert(s
== sizeof(status
));
778 virtqueue_push(vq
, &elem
, sizeof(status
));
779 virtio_notify(vdev
, vq
);
785 static void virtio_net_handle_rx(VirtIODevice
*vdev
, VirtQueue
*vq
)
787 VirtIONet
*n
= VIRTIO_NET(vdev
);
788 int queue_index
= vq2q(virtio_get_queue_index(vq
));
790 qemu_flush_queued_packets(qemu_get_subqueue(n
->nic
, queue_index
));
793 static int virtio_net_can_receive(NetClientState
*nc
)
795 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
796 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
797 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
799 if (!vdev
->vm_running
) {
803 if (nc
->queue_index
>= n
->curr_queues
) {
807 if (!virtio_queue_ready(q
->rx_vq
) ||
808 !(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
815 static int virtio_net_has_buffers(VirtIONetQueue
*q
, int bufsize
)
818 if (virtio_queue_empty(q
->rx_vq
) ||
819 (n
->mergeable_rx_bufs
&&
820 !virtqueue_avail_bytes(q
->rx_vq
, bufsize
, 0))) {
821 virtio_queue_set_notification(q
->rx_vq
, 1);
823 /* To avoid a race condition where the guest has made some buffers
824 * available after the above check but before notification was
825 * enabled, check for available buffers again.
827 if (virtio_queue_empty(q
->rx_vq
) ||
828 (n
->mergeable_rx_bufs
&&
829 !virtqueue_avail_bytes(q
->rx_vq
, bufsize
, 0))) {
834 virtio_queue_set_notification(q
->rx_vq
, 0);
838 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
839 * it never finds out that the packets don't have valid checksums. This
840 * causes dhclient to get upset. Fedora's carried a patch for ages to
841 * fix this with Xen but it hasn't appeared in an upstream release of
844 * To avoid breaking existing guests, we catch udp packets and add
845 * checksums. This is terrible but it's better than hacking the guest
848 * N.B. if we introduce a zero-copy API, this operation is no longer free so
849 * we should provide a mechanism to disable it to avoid polluting the host
852 static void work_around_broken_dhclient(struct virtio_net_hdr
*hdr
,
853 uint8_t *buf
, size_t size
)
855 if ((hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) && /* missing csum */
856 (size
> 27 && size
< 1500) && /* normal sized MTU */
857 (buf
[12] == 0x08 && buf
[13] == 0x00) && /* ethertype == IPv4 */
858 (buf
[23] == 17) && /* ip.protocol == UDP */
859 (buf
[34] == 0 && buf
[35] == 67)) { /* udp.srcport == bootps */
860 net_checksum_calculate(buf
, size
);
861 hdr
->flags
&= ~VIRTIO_NET_HDR_F_NEEDS_CSUM
;
865 static void receive_header(VirtIONet
*n
, const struct iovec
*iov
, int iov_cnt
,
866 const void *buf
, size_t size
)
868 if (n
->has_vnet_hdr
) {
869 /* FIXME this cast is evil */
870 void *wbuf
= (void *)buf
;
871 work_around_broken_dhclient(wbuf
, wbuf
+ n
->host_hdr_len
,
872 size
- n
->host_hdr_len
);
873 iov_from_buf(iov
, iov_cnt
, 0, buf
, sizeof(struct virtio_net_hdr
));
875 struct virtio_net_hdr hdr
= {
877 .gso_type
= VIRTIO_NET_HDR_GSO_NONE
879 iov_from_buf(iov
, iov_cnt
, 0, &hdr
, sizeof hdr
);
883 static int receive_filter(VirtIONet
*n
, const uint8_t *buf
, int size
)
885 static const uint8_t bcast
[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
886 static const uint8_t vlan
[] = {0x81, 0x00};
887 uint8_t *ptr
= (uint8_t *)buf
;
893 ptr
+= n
->host_hdr_len
;
895 if (!memcmp(&ptr
[12], vlan
, sizeof(vlan
))) {
896 int vid
= be16_to_cpup((uint16_t *)(ptr
+ 14)) & 0xfff;
897 if (!(n
->vlans
[vid
>> 5] & (1U << (vid
& 0x1f))))
901 if (ptr
[0] & 1) { // multicast
902 if (!memcmp(ptr
, bcast
, sizeof(bcast
))) {
904 } else if (n
->nomulti
) {
906 } else if (n
->allmulti
|| n
->mac_table
.multi_overflow
) {
910 for (i
= n
->mac_table
.first_multi
; i
< n
->mac_table
.in_use
; i
++) {
911 if (!memcmp(ptr
, &n
->mac_table
.macs
[i
* ETH_ALEN
], ETH_ALEN
)) {
918 } else if (n
->alluni
|| n
->mac_table
.uni_overflow
) {
920 } else if (!memcmp(ptr
, n
->mac
, ETH_ALEN
)) {
924 for (i
= 0; i
< n
->mac_table
.first_multi
; i
++) {
925 if (!memcmp(ptr
, &n
->mac_table
.macs
[i
* ETH_ALEN
], ETH_ALEN
)) {
934 static ssize_t
virtio_net_receive(NetClientState
*nc
, const uint8_t *buf
, size_t size
)
936 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
937 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
938 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
939 struct iovec mhdr_sg
[VIRTQUEUE_MAX_SIZE
];
940 struct virtio_net_hdr_mrg_rxbuf mhdr
;
941 unsigned mhdr_cnt
= 0;
942 size_t offset
, i
, guest_offset
;
944 if (!virtio_net_can_receive(nc
)) {
948 /* hdr_len refers to the header we supply to the guest */
949 if (!virtio_net_has_buffers(q
, size
+ n
->guest_hdr_len
- n
->host_hdr_len
)) {
953 if (!receive_filter(n
, buf
, size
))
958 while (offset
< size
) {
959 VirtQueueElement elem
;
961 const struct iovec
*sg
= elem
.in_sg
;
965 if (virtqueue_pop(q
->rx_vq
, &elem
) == 0) {
968 error_report("virtio-net unexpected empty queue: "
969 "i %zd mergeable %d offset %zd, size %zd, "
970 "guest hdr len %zd, host hdr len %zd guest features 0x%x",
971 i
, n
->mergeable_rx_bufs
, offset
, size
,
972 n
->guest_hdr_len
, n
->host_hdr_len
, vdev
->guest_features
);
976 if (elem
.in_num
< 1) {
977 error_report("virtio-net receive queue contains no in buffers");
983 if (n
->mergeable_rx_bufs
) {
984 mhdr_cnt
= iov_copy(mhdr_sg
, ARRAY_SIZE(mhdr_sg
),
986 offsetof(typeof(mhdr
), num_buffers
),
987 sizeof(mhdr
.num_buffers
));
990 receive_header(n
, sg
, elem
.in_num
, buf
, size
);
991 offset
= n
->host_hdr_len
;
992 total
+= n
->guest_hdr_len
;
993 guest_offset
= n
->guest_hdr_len
;
998 /* copy in packet. ugh */
999 len
= iov_from_buf(sg
, elem
.in_num
, guest_offset
,
1000 buf
+ offset
, size
- offset
);
1003 /* If buffers can't be merged, at this point we
1004 * must have consumed the complete packet.
1005 * Otherwise, drop it. */
1006 if (!n
->mergeable_rx_bufs
&& offset
< size
) {
1008 error_report("virtio-net truncated non-mergeable packet: "
1009 "i %zd mergeable %d offset %zd, size %zd, "
1010 "guest hdr len %zd, host hdr len %zd",
1011 i
, n
->mergeable_rx_bufs
,
1012 offset
, size
, n
->guest_hdr_len
, n
->host_hdr_len
);
1017 /* signal other side */
1018 virtqueue_fill(q
->rx_vq
, &elem
, total
, i
++);
1022 stw_p(&mhdr
.num_buffers
, i
);
1023 iov_from_buf(mhdr_sg
, mhdr_cnt
,
1025 &mhdr
.num_buffers
, sizeof mhdr
.num_buffers
);
1028 virtqueue_flush(q
->rx_vq
, i
);
1029 virtio_notify(vdev
, q
->rx_vq
);
1034 static int32_t virtio_net_flush_tx(VirtIONetQueue
*q
);
1036 static void virtio_net_tx_complete(NetClientState
*nc
, ssize_t len
)
1038 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
1039 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
1040 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1042 virtqueue_push(q
->tx_vq
, &q
->async_tx
.elem
, 0);
1043 virtio_notify(vdev
, q
->tx_vq
);
1045 q
->async_tx
.elem
.out_num
= q
->async_tx
.len
= 0;
1047 virtio_queue_set_notification(q
->tx_vq
, 1);
1048 virtio_net_flush_tx(q
);
1052 static int32_t virtio_net_flush_tx(VirtIONetQueue
*q
)
1054 VirtIONet
*n
= q
->n
;
1055 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1056 VirtQueueElement elem
;
1057 int32_t num_packets
= 0;
1058 int queue_index
= vq2q(virtio_get_queue_index(q
->tx_vq
));
1059 if (!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
1063 assert(vdev
->vm_running
);
1065 if (q
->async_tx
.elem
.out_num
) {
1066 virtio_queue_set_notification(q
->tx_vq
, 0);
1070 while (virtqueue_pop(q
->tx_vq
, &elem
)) {
1072 unsigned int out_num
= elem
.out_num
;
1073 struct iovec
*out_sg
= &elem
.out_sg
[0];
1074 struct iovec sg
[VIRTQUEUE_MAX_SIZE
];
1077 error_report("virtio-net header not in first element");
1082 * If host wants to see the guest header as is, we can
1083 * pass it on unchanged. Otherwise, copy just the parts
1084 * that host is interested in.
1086 assert(n
->host_hdr_len
<= n
->guest_hdr_len
);
1087 if (n
->host_hdr_len
!= n
->guest_hdr_len
) {
1088 unsigned sg_num
= iov_copy(sg
, ARRAY_SIZE(sg
),
1090 0, n
->host_hdr_len
);
1091 sg_num
+= iov_copy(sg
+ sg_num
, ARRAY_SIZE(sg
) - sg_num
,
1093 n
->guest_hdr_len
, -1);
1098 len
= n
->guest_hdr_len
;
1100 ret
= qemu_sendv_packet_async(qemu_get_subqueue(n
->nic
, queue_index
),
1101 out_sg
, out_num
, virtio_net_tx_complete
);
1103 virtio_queue_set_notification(q
->tx_vq
, 0);
1104 q
->async_tx
.elem
= elem
;
1105 q
->async_tx
.len
= len
;
1111 virtqueue_push(q
->tx_vq
, &elem
, 0);
1112 virtio_notify(vdev
, q
->tx_vq
);
1114 if (++num_packets
>= n
->tx_burst
) {
1121 static void virtio_net_handle_tx_timer(VirtIODevice
*vdev
, VirtQueue
*vq
)
1123 VirtIONet
*n
= VIRTIO_NET(vdev
);
1124 VirtIONetQueue
*q
= &n
->vqs
[vq2q(virtio_get_queue_index(vq
))];
1126 /* This happens when device was stopped but VCPU wasn't. */
1127 if (!vdev
->vm_running
) {
1132 if (q
->tx_waiting
) {
1133 virtio_queue_set_notification(vq
, 1);
1134 qemu_del_timer(q
->tx_timer
);
1136 virtio_net_flush_tx(q
);
1138 qemu_mod_timer(q
->tx_timer
,
1139 qemu_get_clock_ns(vm_clock
) + n
->tx_timeout
);
1141 virtio_queue_set_notification(vq
, 0);
1145 static void virtio_net_handle_tx_bh(VirtIODevice
*vdev
, VirtQueue
*vq
)
1147 VirtIONet
*n
= VIRTIO_NET(vdev
);
1148 VirtIONetQueue
*q
= &n
->vqs
[vq2q(virtio_get_queue_index(vq
))];
1150 if (unlikely(q
->tx_waiting
)) {
1154 /* This happens when device was stopped but VCPU wasn't. */
1155 if (!vdev
->vm_running
) {
1158 virtio_queue_set_notification(vq
, 0);
1159 qemu_bh_schedule(q
->tx_bh
);
1162 static void virtio_net_tx_timer(void *opaque
)
1164 VirtIONetQueue
*q
= opaque
;
1165 VirtIONet
*n
= q
->n
;
1166 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1167 assert(vdev
->vm_running
);
1171 /* Just in case the driver is not ready on more */
1172 if (!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
1176 virtio_queue_set_notification(q
->tx_vq
, 1);
1177 virtio_net_flush_tx(q
);
1180 static void virtio_net_tx_bh(void *opaque
)
1182 VirtIONetQueue
*q
= opaque
;
1183 VirtIONet
*n
= q
->n
;
1184 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1187 assert(vdev
->vm_running
);
1191 /* Just in case the driver is not ready on more */
1192 if (unlikely(!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
))) {
1196 ret
= virtio_net_flush_tx(q
);
1197 if (ret
== -EBUSY
) {
1198 return; /* Notification re-enable handled by tx_complete */
1201 /* If we flush a full burst of packets, assume there are
1202 * more coming and immediately reschedule */
1203 if (ret
>= n
->tx_burst
) {
1204 qemu_bh_schedule(q
->tx_bh
);
1209 /* If less than a full burst, re-enable notification and flush
1210 * anything that may have come in while we weren't looking. If
1211 * we find something, assume the guest is still active and reschedule */
1212 virtio_queue_set_notification(q
->tx_vq
, 1);
1213 if (virtio_net_flush_tx(q
) > 0) {
1214 virtio_queue_set_notification(q
->tx_vq
, 0);
1215 qemu_bh_schedule(q
->tx_bh
);
1220 static void virtio_net_set_multiqueue(VirtIONet
*n
, int multiqueue
)
1222 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1223 int i
, max
= multiqueue
? n
->max_queues
: 1;
1225 n
->multiqueue
= multiqueue
;
1227 for (i
= 2; i
<= n
->max_queues
* 2 + 1; i
++) {
1228 virtio_del_queue(vdev
, i
);
1231 for (i
= 1; i
< max
; i
++) {
1232 n
->vqs
[i
].rx_vq
= virtio_add_queue(vdev
, 256, virtio_net_handle_rx
);
1233 if (n
->vqs
[i
].tx_timer
) {
1235 virtio_add_queue(vdev
, 256, virtio_net_handle_tx_timer
);
1236 n
->vqs
[i
].tx_timer
= qemu_new_timer_ns(vm_clock
,
1237 virtio_net_tx_timer
,
1241 virtio_add_queue(vdev
, 256, virtio_net_handle_tx_bh
);
1242 n
->vqs
[i
].tx_bh
= qemu_bh_new(virtio_net_tx_bh
, &n
->vqs
[i
]);
1245 n
->vqs
[i
].tx_waiting
= 0;
1249 /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack
1250 * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid
1253 n
->ctrl_vq
= virtio_add_queue(vdev
, 64, virtio_net_handle_ctrl
);
1255 virtio_net_set_queues(n
);
1258 static void virtio_net_save(QEMUFile
*f
, void *opaque
)
1261 VirtIONet
*n
= opaque
;
1262 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1264 /* At this point, backend must be stopped, otherwise
1265 * it might keep writing to memory. */
1266 assert(!n
->vhost_started
);
1267 virtio_save(vdev
, f
);
1269 qemu_put_buffer(f
, n
->mac
, ETH_ALEN
);
1270 qemu_put_be32(f
, n
->vqs
[0].tx_waiting
);
1271 qemu_put_be32(f
, n
->mergeable_rx_bufs
);
1272 qemu_put_be16(f
, n
->status
);
1273 qemu_put_byte(f
, n
->promisc
);
1274 qemu_put_byte(f
, n
->allmulti
);
1275 qemu_put_be32(f
, n
->mac_table
.in_use
);
1276 qemu_put_buffer(f
, n
->mac_table
.macs
, n
->mac_table
.in_use
* ETH_ALEN
);
1277 qemu_put_buffer(f
, (uint8_t *)n
->vlans
, MAX_VLAN
>> 3);
1278 qemu_put_be32(f
, n
->has_vnet_hdr
);
1279 qemu_put_byte(f
, n
->mac_table
.multi_overflow
);
1280 qemu_put_byte(f
, n
->mac_table
.uni_overflow
);
1281 qemu_put_byte(f
, n
->alluni
);
1282 qemu_put_byte(f
, n
->nomulti
);
1283 qemu_put_byte(f
, n
->nouni
);
1284 qemu_put_byte(f
, n
->nobcast
);
1285 qemu_put_byte(f
, n
->has_ufo
);
1286 if (n
->max_queues
> 1) {
1287 qemu_put_be16(f
, n
->max_queues
);
1288 qemu_put_be16(f
, n
->curr_queues
);
1289 for (i
= 1; i
< n
->curr_queues
; i
++) {
1290 qemu_put_be32(f
, n
->vqs
[i
].tx_waiting
);
1294 if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
) & vdev
->guest_features
) {
1295 qemu_put_be64(f
, n
->curr_guest_offloads
);
1299 static int virtio_net_load(QEMUFile
*f
, void *opaque
, int version_id
)
1301 VirtIONet
*n
= opaque
;
1302 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1303 int ret
, i
, link_down
;
1305 if (version_id
< 2 || version_id
> VIRTIO_NET_VM_VERSION
)
1308 ret
= virtio_load(vdev
, f
);
1313 qemu_get_buffer(f
, n
->mac
, ETH_ALEN
);
1314 n
->vqs
[0].tx_waiting
= qemu_get_be32(f
);
1316 virtio_net_set_mrg_rx_bufs(n
, qemu_get_be32(f
));
1318 if (version_id
>= 3)
1319 n
->status
= qemu_get_be16(f
);
1321 if (version_id
>= 4) {
1322 if (version_id
< 8) {
1323 n
->promisc
= qemu_get_be32(f
);
1324 n
->allmulti
= qemu_get_be32(f
);
1326 n
->promisc
= qemu_get_byte(f
);
1327 n
->allmulti
= qemu_get_byte(f
);
1331 if (version_id
>= 5) {
1332 n
->mac_table
.in_use
= qemu_get_be32(f
);
1333 /* MAC_TABLE_ENTRIES may be different from the saved image */
1334 if (n
->mac_table
.in_use
<= MAC_TABLE_ENTRIES
) {
1335 qemu_get_buffer(f
, n
->mac_table
.macs
,
1336 n
->mac_table
.in_use
* ETH_ALEN
);
1337 } else if (n
->mac_table
.in_use
) {
1338 uint8_t *buf
= g_malloc0(n
->mac_table
.in_use
);
1339 qemu_get_buffer(f
, buf
, n
->mac_table
.in_use
* ETH_ALEN
);
1341 n
->mac_table
.multi_overflow
= n
->mac_table
.uni_overflow
= 1;
1342 n
->mac_table
.in_use
= 0;
1346 if (version_id
>= 6)
1347 qemu_get_buffer(f
, (uint8_t *)n
->vlans
, MAX_VLAN
>> 3);
1349 if (version_id
>= 7) {
1350 if (qemu_get_be32(f
) && !peer_has_vnet_hdr(n
)) {
1351 error_report("virtio-net: saved image requires vnet_hdr=on");
1356 if (version_id
>= 9) {
1357 n
->mac_table
.multi_overflow
= qemu_get_byte(f
);
1358 n
->mac_table
.uni_overflow
= qemu_get_byte(f
);
1361 if (version_id
>= 10) {
1362 n
->alluni
= qemu_get_byte(f
);
1363 n
->nomulti
= qemu_get_byte(f
);
1364 n
->nouni
= qemu_get_byte(f
);
1365 n
->nobcast
= qemu_get_byte(f
);
1368 if (version_id
>= 11) {
1369 if (qemu_get_byte(f
) && !peer_has_ufo(n
)) {
1370 error_report("virtio-net: saved image requires TUN_F_UFO support");
1375 if (n
->max_queues
> 1) {
1376 if (n
->max_queues
!= qemu_get_be16(f
)) {
1377 error_report("virtio-net: different max_queues ");
1381 n
->curr_queues
= qemu_get_be16(f
);
1382 for (i
= 1; i
< n
->curr_queues
; i
++) {
1383 n
->vqs
[i
].tx_waiting
= qemu_get_be32(f
);
1387 if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
) & vdev
->guest_features
) {
1388 n
->curr_guest_offloads
= qemu_get_be64(f
);
1390 n
->curr_guest_offloads
= virtio_net_supported_guest_offloads(n
);
1393 if (peer_has_vnet_hdr(n
)) {
1394 virtio_net_apply_guest_offloads(n
);
1397 virtio_net_set_queues(n
);
1399 /* Find the first multicast entry in the saved MAC filter */
1400 for (i
= 0; i
< n
->mac_table
.in_use
; i
++) {
1401 if (n
->mac_table
.macs
[i
* ETH_ALEN
] & 1) {
1405 n
->mac_table
.first_multi
= i
;
1407 /* nc.link_down can't be migrated, so infer link_down according
1408 * to link status bit in n->status */
1409 link_down
= (n
->status
& VIRTIO_NET_S_LINK_UP
) == 0;
1410 for (i
= 0; i
< n
->max_queues
; i
++) {
1411 qemu_get_subqueue(n
->nic
, i
)->link_down
= link_down
;
1417 static void virtio_net_cleanup(NetClientState
*nc
)
1419 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
1424 static NetClientInfo net_virtio_info
= {
1425 .type
= NET_CLIENT_OPTIONS_KIND_NIC
,
1426 .size
= sizeof(NICState
),
1427 .can_receive
= virtio_net_can_receive
,
1428 .receive
= virtio_net_receive
,
1429 .cleanup
= virtio_net_cleanup
,
1430 .link_status_changed
= virtio_net_set_link_status
,
1431 .query_rx_filter
= virtio_net_query_rxfilter
,
1434 static bool virtio_net_guest_notifier_pending(VirtIODevice
*vdev
, int idx
)
1436 VirtIONet
*n
= VIRTIO_NET(vdev
);
1437 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, vq2q(idx
));
1438 assert(n
->vhost_started
);
1439 return vhost_net_virtqueue_pending(tap_get_vhost_net(nc
->peer
), idx
);
1442 static void virtio_net_guest_notifier_mask(VirtIODevice
*vdev
, int idx
,
1445 VirtIONet
*n
= VIRTIO_NET(vdev
);
1446 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, vq2q(idx
));
1447 assert(n
->vhost_started
);
1448 vhost_net_virtqueue_mask(tap_get_vhost_net(nc
->peer
),
1452 void virtio_net_set_config_size(VirtIONet
*n
, uint32_t host_features
)
1454 int i
, config_size
= 0;
1455 host_features
|= (1 << VIRTIO_NET_F_MAC
);
1456 for (i
= 0; feature_sizes
[i
].flags
!= 0; i
++) {
1457 if (host_features
& feature_sizes
[i
].flags
) {
1458 config_size
= MAX(feature_sizes
[i
].end
, config_size
);
1461 n
->config_size
= config_size
;
1464 void virtio_net_set_netclient_name(VirtIONet
*n
, const char *name
,
1468 * The name can be NULL, the netclient name will be type.x.
1470 assert(type
!= NULL
);
1472 if (n
->netclient_name
) {
1473 g_free(n
->netclient_name
);
1474 n
->netclient_name
= NULL
;
1476 if (n
->netclient_type
) {
1477 g_free(n
->netclient_type
);
1478 n
->netclient_type
= NULL
;
1482 n
->netclient_name
= g_strdup(name
);
1484 n
->netclient_type
= g_strdup(type
);
1487 static int virtio_net_device_init(VirtIODevice
*vdev
)
1491 DeviceState
*qdev
= DEVICE(vdev
);
1492 VirtIONet
*n
= VIRTIO_NET(vdev
);
1495 virtio_init(VIRTIO_DEVICE(n
), "virtio-net", VIRTIO_ID_NET
,
1498 n
->max_queues
= MAX(n
->nic_conf
.queues
, 1);
1499 n
->vqs
= g_malloc0(sizeof(VirtIONetQueue
) * n
->max_queues
);
1500 n
->vqs
[0].rx_vq
= virtio_add_queue(vdev
, 256, virtio_net_handle_rx
);
1503 n
->tx_timeout
= n
->net_conf
.txtimer
;
1505 if (n
->net_conf
.tx
&& strcmp(n
->net_conf
.tx
, "timer")
1506 && strcmp(n
->net_conf
.tx
, "bh")) {
1507 error_report("virtio-net: "
1508 "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1510 error_report("Defaulting to \"bh\"");
1513 if (n
->net_conf
.tx
&& !strcmp(n
->net_conf
.tx
, "timer")) {
1514 n
->vqs
[0].tx_vq
= virtio_add_queue(vdev
, 256,
1515 virtio_net_handle_tx_timer
);
1516 n
->vqs
[0].tx_timer
= qemu_new_timer_ns(vm_clock
, virtio_net_tx_timer
,
1519 n
->vqs
[0].tx_vq
= virtio_add_queue(vdev
, 256,
1520 virtio_net_handle_tx_bh
);
1521 n
->vqs
[0].tx_bh
= qemu_bh_new(virtio_net_tx_bh
, &n
->vqs
[0]);
1523 n
->ctrl_vq
= virtio_add_queue(vdev
, 64, virtio_net_handle_ctrl
);
1524 qemu_macaddr_default_if_unset(&n
->nic_conf
.macaddr
);
1525 memcpy(&n
->mac
[0], &n
->nic_conf
.macaddr
, sizeof(n
->mac
));
1526 n
->status
= VIRTIO_NET_S_LINK_UP
;
1528 if (n
->netclient_type
) {
1530 * Happen when virtio_net_set_netclient_name has been called.
1532 n
->nic
= qemu_new_nic(&net_virtio_info
, &n
->nic_conf
,
1533 n
->netclient_type
, n
->netclient_name
, n
);
1535 n
->nic
= qemu_new_nic(&net_virtio_info
, &n
->nic_conf
,
1536 object_get_typename(OBJECT(qdev
)), qdev
->id
, n
);
1539 peer_test_vnet_hdr(n
);
1540 if (peer_has_vnet_hdr(n
)) {
1541 for (i
= 0; i
< n
->max_queues
; i
++) {
1542 tap_using_vnet_hdr(qemu_get_subqueue(n
->nic
, i
)->peer
, true);
1544 n
->host_hdr_len
= sizeof(struct virtio_net_hdr
);
1546 n
->host_hdr_len
= 0;
1549 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->nic_conf
.macaddr
.a
);
1551 n
->vqs
[0].tx_waiting
= 0;
1552 n
->tx_burst
= n
->net_conf
.txburst
;
1553 virtio_net_set_mrg_rx_bufs(n
, 0);
1554 n
->promisc
= 1; /* for compatibility */
1556 n
->mac_table
.macs
= g_malloc0(MAC_TABLE_ENTRIES
* ETH_ALEN
);
1558 n
->vlans
= g_malloc0(MAX_VLAN
>> 3);
1560 nc
= qemu_get_queue(n
->nic
);
1561 nc
->rxfilter_notify_enabled
= 1;
1564 register_savevm(qdev
, "virtio-net", -1, VIRTIO_NET_VM_VERSION
,
1565 virtio_net_save
, virtio_net_load
, n
);
1567 add_boot_device_path(n
->nic_conf
.bootindex
, qdev
, "/ethernet-phy@0");
1571 static int virtio_net_device_exit(DeviceState
*qdev
)
1573 VirtIONet
*n
= VIRTIO_NET(qdev
);
1574 VirtIODevice
*vdev
= VIRTIO_DEVICE(qdev
);
1577 /* This will stop vhost backend if appropriate. */
1578 virtio_net_set_status(vdev
, 0);
1580 unregister_savevm(qdev
, "virtio-net", n
);
1582 if (n
->netclient_name
) {
1583 g_free(n
->netclient_name
);
1584 n
->netclient_name
= NULL
;
1586 if (n
->netclient_type
) {
1587 g_free(n
->netclient_type
);
1588 n
->netclient_type
= NULL
;
1591 g_free(n
->mac_table
.macs
);
1594 for (i
= 0; i
< n
->max_queues
; i
++) {
1595 VirtIONetQueue
*q
= &n
->vqs
[i
];
1596 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, i
);
1598 qemu_purge_queued_packets(nc
);
1601 qemu_del_timer(q
->tx_timer
);
1602 qemu_free_timer(q
->tx_timer
);
1604 qemu_bh_delete(q
->tx_bh
);
1609 qemu_del_nic(n
->nic
);
1610 virtio_cleanup(vdev
);
1615 static void virtio_net_instance_init(Object
*obj
)
1617 VirtIONet
*n
= VIRTIO_NET(obj
);
1620 * The default config_size is sizeof(struct virtio_net_config).
1621 * Can be overriden with virtio_net_set_config_size.
1623 n
->config_size
= sizeof(struct virtio_net_config
);
1626 static Property virtio_net_properties
[] = {
1627 DEFINE_NIC_PROPERTIES(VirtIONet
, nic_conf
),
1628 DEFINE_PROP_UINT32("x-txtimer", VirtIONet
, net_conf
.txtimer
,
1630 DEFINE_PROP_INT32("x-txburst", VirtIONet
, net_conf
.txburst
, TX_BURST
),
1631 DEFINE_PROP_STRING("tx", VirtIONet
, net_conf
.tx
),
1632 DEFINE_PROP_END_OF_LIST(),
1635 static void virtio_net_class_init(ObjectClass
*klass
, void *data
)
1637 DeviceClass
*dc
= DEVICE_CLASS(klass
);
1638 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
1639 dc
->exit
= virtio_net_device_exit
;
1640 dc
->props
= virtio_net_properties
;
1641 set_bit(DEVICE_CATEGORY_NETWORK
, dc
->categories
);
1642 vdc
->init
= virtio_net_device_init
;
1643 vdc
->get_config
= virtio_net_get_config
;
1644 vdc
->set_config
= virtio_net_set_config
;
1645 vdc
->get_features
= virtio_net_get_features
;
1646 vdc
->set_features
= virtio_net_set_features
;
1647 vdc
->bad_features
= virtio_net_bad_features
;
1648 vdc
->reset
= virtio_net_reset
;
1649 vdc
->set_status
= virtio_net_set_status
;
1650 vdc
->guest_notifier_mask
= virtio_net_guest_notifier_mask
;
1651 vdc
->guest_notifier_pending
= virtio_net_guest_notifier_pending
;
1654 static const TypeInfo virtio_net_info
= {
1655 .name
= TYPE_VIRTIO_NET
,
1656 .parent
= TYPE_VIRTIO_DEVICE
,
1657 .instance_size
= sizeof(VirtIONet
),
1658 .instance_init
= virtio_net_instance_init
,
1659 .class_init
= virtio_net_class_init
,
1662 static void virtio_register_types(void)
1664 type_register_static(&virtio_net_info
);
1667 type_init(virtio_register_types
)