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 "qapi-event.h"
26 #include "hw/virtio/virtio-access.h"
28 #define VIRTIO_NET_VM_VERSION 11
30 #define MAC_TABLE_ENTRIES 64
31 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */
34 * Calculate the number of bytes up to and including the given 'field' of
37 #define endof(container, field) \
38 (offsetof(container, field) + sizeof(((container *)0)->field))
40 typedef struct VirtIOFeature
{
45 static VirtIOFeature feature_sizes
[] = {
46 {.flags
= 1 << VIRTIO_NET_F_MAC
,
47 .end
= endof(struct virtio_net_config
, mac
)},
48 {.flags
= 1 << VIRTIO_NET_F_STATUS
,
49 .end
= endof(struct virtio_net_config
, status
)},
50 {.flags
= 1 << VIRTIO_NET_F_MQ
,
51 .end
= endof(struct virtio_net_config
, max_virtqueue_pairs
)},
55 static VirtIONetQueue
*virtio_net_get_subqueue(NetClientState
*nc
)
57 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
59 return &n
->vqs
[nc
->queue_index
];
62 static int vq2q(int queue_index
)
64 return queue_index
/ 2;
68 * - we could suppress RX interrupt if we were so inclined.
71 static void virtio_net_get_config(VirtIODevice
*vdev
, uint8_t *config
)
73 VirtIONet
*n
= VIRTIO_NET(vdev
);
74 struct virtio_net_config netcfg
;
76 virtio_stw_p(vdev
, &netcfg
.status
, n
->status
);
77 virtio_stw_p(vdev
, &netcfg
.max_virtqueue_pairs
, n
->max_queues
);
78 memcpy(netcfg
.mac
, n
->mac
, ETH_ALEN
);
79 memcpy(config
, &netcfg
, n
->config_size
);
82 static void virtio_net_set_config(VirtIODevice
*vdev
, const uint8_t *config
)
84 VirtIONet
*n
= VIRTIO_NET(vdev
);
85 struct virtio_net_config netcfg
= {};
87 memcpy(&netcfg
, config
, n
->config_size
);
89 if (!virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
) &&
90 !virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
) &&
91 memcmp(netcfg
.mac
, n
->mac
, ETH_ALEN
)) {
92 memcpy(n
->mac
, netcfg
.mac
, ETH_ALEN
);
93 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->mac
);
97 static bool virtio_net_started(VirtIONet
*n
, uint8_t status
)
99 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
100 return (status
& VIRTIO_CONFIG_S_DRIVER_OK
) &&
101 (n
->status
& VIRTIO_NET_S_LINK_UP
) && vdev
->vm_running
;
104 static void virtio_net_announce_timer(void *opaque
)
106 VirtIONet
*n
= opaque
;
107 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
109 n
->announce_counter
--;
110 n
->status
|= VIRTIO_NET_S_ANNOUNCE
;
111 virtio_notify_config(vdev
);
114 static void virtio_net_vhost_status(VirtIONet
*n
, uint8_t status
)
116 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
117 NetClientState
*nc
= qemu_get_queue(n
->nic
);
118 int queues
= n
->multiqueue
? n
->max_queues
: 1;
120 if (!get_vhost_net(nc
->peer
)) {
124 if ((virtio_net_started(n
, status
) && !nc
->peer
->link_down
) ==
125 !!n
->vhost_started
) {
128 if (!n
->vhost_started
) {
131 /* Any packets outstanding? Purge them to avoid touching rings
132 * when vhost is running.
134 for (i
= 0; i
< queues
; i
++) {
135 NetClientState
*qnc
= qemu_get_subqueue(n
->nic
, i
);
137 /* Purge both directions: TX and RX. */
138 qemu_net_queue_purge(qnc
->peer
->incoming_queue
, qnc
);
139 qemu_net_queue_purge(qnc
->incoming_queue
, qnc
->peer
);
142 n
->vhost_started
= 1;
143 r
= vhost_net_start(vdev
, n
->nic
->ncs
, queues
);
145 error_report("unable to start vhost net: %d: "
146 "falling back on userspace virtio", -r
);
147 n
->vhost_started
= 0;
150 vhost_net_stop(vdev
, n
->nic
->ncs
, queues
);
151 n
->vhost_started
= 0;
155 static void virtio_net_set_status(struct VirtIODevice
*vdev
, uint8_t status
)
157 VirtIONet
*n
= VIRTIO_NET(vdev
);
160 uint8_t queue_status
;
162 virtio_net_vhost_status(n
, status
);
164 for (i
= 0; i
< n
->max_queues
; i
++) {
167 if ((!n
->multiqueue
&& i
!= 0) || i
>= n
->curr_queues
) {
170 queue_status
= status
;
173 if (!q
->tx_waiting
) {
177 if (virtio_net_started(n
, queue_status
) && !n
->vhost_started
) {
179 timer_mod(q
->tx_timer
,
180 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) + n
->tx_timeout
);
182 qemu_bh_schedule(q
->tx_bh
);
186 timer_del(q
->tx_timer
);
188 qemu_bh_cancel(q
->tx_bh
);
194 static void virtio_net_set_link_status(NetClientState
*nc
)
196 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
197 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
198 uint16_t old_status
= n
->status
;
201 n
->status
&= ~VIRTIO_NET_S_LINK_UP
;
203 n
->status
|= VIRTIO_NET_S_LINK_UP
;
205 if (n
->status
!= old_status
)
206 virtio_notify_config(vdev
);
208 virtio_net_set_status(vdev
, vdev
->status
);
211 static void rxfilter_notify(NetClientState
*nc
)
213 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
215 if (nc
->rxfilter_notify_enabled
) {
216 gchar
*path
= object_get_canonical_path(OBJECT(n
->qdev
));
217 qapi_event_send_nic_rx_filter_changed(!!n
->netclient_name
,
218 n
->netclient_name
, path
, &error_abort
);
221 /* disable event notification to avoid events flooding */
222 nc
->rxfilter_notify_enabled
= 0;
226 static intList
*get_vlan_table(VirtIONet
*n
)
228 intList
*list
, *entry
;
232 for (i
= 0; i
< MAX_VLAN
>> 5; i
++) {
233 for (j
= 0; n
->vlans
[i
] && j
<= 0x1f; j
++) {
234 if (n
->vlans
[i
] & (1U << j
)) {
235 entry
= g_malloc0(sizeof(*entry
));
236 entry
->value
= (i
<< 5) + j
;
246 static RxFilterInfo
*virtio_net_query_rxfilter(NetClientState
*nc
)
248 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
249 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
251 strList
*str_list
, *entry
;
254 info
= g_malloc0(sizeof(*info
));
255 info
->name
= g_strdup(nc
->name
);
256 info
->promiscuous
= n
->promisc
;
259 info
->unicast
= RX_STATE_NONE
;
260 } else if (n
->alluni
) {
261 info
->unicast
= RX_STATE_ALL
;
263 info
->unicast
= RX_STATE_NORMAL
;
267 info
->multicast
= RX_STATE_NONE
;
268 } else if (n
->allmulti
) {
269 info
->multicast
= RX_STATE_ALL
;
271 info
->multicast
= RX_STATE_NORMAL
;
274 info
->broadcast_allowed
= n
->nobcast
;
275 info
->multicast_overflow
= n
->mac_table
.multi_overflow
;
276 info
->unicast_overflow
= n
->mac_table
.uni_overflow
;
278 info
->main_mac
= qemu_mac_strdup_printf(n
->mac
);
281 for (i
= 0; i
< n
->mac_table
.first_multi
; i
++) {
282 entry
= g_malloc0(sizeof(*entry
));
283 entry
->value
= qemu_mac_strdup_printf(n
->mac_table
.macs
+ i
* ETH_ALEN
);
284 entry
->next
= str_list
;
287 info
->unicast_table
= str_list
;
290 for (i
= n
->mac_table
.first_multi
; i
< n
->mac_table
.in_use
; i
++) {
291 entry
= g_malloc0(sizeof(*entry
));
292 entry
->value
= qemu_mac_strdup_printf(n
->mac_table
.macs
+ i
* ETH_ALEN
);
293 entry
->next
= str_list
;
296 info
->multicast_table
= str_list
;
297 info
->vlan_table
= get_vlan_table(n
);
299 if (!virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VLAN
)) {
300 info
->vlan
= RX_STATE_ALL
;
301 } else if (!info
->vlan_table
) {
302 info
->vlan
= RX_STATE_NONE
;
304 info
->vlan
= RX_STATE_NORMAL
;
307 /* enable event notification after query */
308 nc
->rxfilter_notify_enabled
= 1;
313 static void virtio_net_reset(VirtIODevice
*vdev
)
315 VirtIONet
*n
= VIRTIO_NET(vdev
);
317 /* Reset back to compatibility mode */
324 /* multiqueue is disabled by default */
326 timer_del(n
->announce_timer
);
327 n
->announce_counter
= 0;
328 n
->status
&= ~VIRTIO_NET_S_ANNOUNCE
;
330 /* Flush any MAC and VLAN filter table state */
331 n
->mac_table
.in_use
= 0;
332 n
->mac_table
.first_multi
= 0;
333 n
->mac_table
.multi_overflow
= 0;
334 n
->mac_table
.uni_overflow
= 0;
335 memset(n
->mac_table
.macs
, 0, MAC_TABLE_ENTRIES
* ETH_ALEN
);
336 memcpy(&n
->mac
[0], &n
->nic
->conf
->macaddr
, sizeof(n
->mac
));
337 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->mac
);
338 memset(n
->vlans
, 0, MAX_VLAN
>> 3);
341 static void peer_test_vnet_hdr(VirtIONet
*n
)
343 NetClientState
*nc
= qemu_get_queue(n
->nic
);
348 n
->has_vnet_hdr
= qemu_has_vnet_hdr(nc
->peer
);
351 static int peer_has_vnet_hdr(VirtIONet
*n
)
353 return n
->has_vnet_hdr
;
356 static int peer_has_ufo(VirtIONet
*n
)
358 if (!peer_has_vnet_hdr(n
))
361 n
->has_ufo
= qemu_has_ufo(qemu_get_queue(n
->nic
)->peer
);
366 static void virtio_net_set_mrg_rx_bufs(VirtIONet
*n
, int mergeable_rx_bufs
,
372 n
->mergeable_rx_bufs
= mergeable_rx_bufs
;
375 n
->guest_hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
377 n
->guest_hdr_len
= n
->mergeable_rx_bufs
?
378 sizeof(struct virtio_net_hdr_mrg_rxbuf
) :
379 sizeof(struct virtio_net_hdr
);
382 for (i
= 0; i
< n
->max_queues
; i
++) {
383 nc
= qemu_get_subqueue(n
->nic
, i
);
385 if (peer_has_vnet_hdr(n
) &&
386 qemu_has_vnet_hdr_len(nc
->peer
, n
->guest_hdr_len
)) {
387 qemu_set_vnet_hdr_len(nc
->peer
, n
->guest_hdr_len
);
388 n
->host_hdr_len
= n
->guest_hdr_len
;
393 static int peer_attach(VirtIONet
*n
, int index
)
395 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, index
);
401 if (nc
->peer
->info
->type
!= NET_CLIENT_OPTIONS_KIND_TAP
) {
405 return tap_enable(nc
->peer
);
408 static int peer_detach(VirtIONet
*n
, int index
)
410 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, index
);
416 if (nc
->peer
->info
->type
!= NET_CLIENT_OPTIONS_KIND_TAP
) {
420 return tap_disable(nc
->peer
);
423 static void virtio_net_set_queues(VirtIONet
*n
)
428 for (i
= 0; i
< n
->max_queues
; i
++) {
429 if (i
< n
->curr_queues
) {
430 r
= peer_attach(n
, i
);
433 r
= peer_detach(n
, i
);
439 static void virtio_net_set_multiqueue(VirtIONet
*n
, int multiqueue
);
441 static uint64_t virtio_net_get_features(VirtIODevice
*vdev
, uint64_t features
)
443 VirtIONet
*n
= VIRTIO_NET(vdev
);
444 NetClientState
*nc
= qemu_get_queue(n
->nic
);
446 /* Firstly sync all virtio-net possible supported features */
447 features
|= n
->host_features
;
449 virtio_add_feature(&features
, VIRTIO_NET_F_MAC
);
451 if (!peer_has_vnet_hdr(n
)) {
452 virtio_clear_feature(&features
, VIRTIO_NET_F_CSUM
);
453 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_TSO4
);
454 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_TSO6
);
455 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_ECN
);
457 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_CSUM
);
458 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_TSO4
);
459 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_TSO6
);
460 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_ECN
);
463 if (!peer_has_vnet_hdr(n
) || !peer_has_ufo(n
)) {
464 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_UFO
);
465 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_UFO
);
468 if (!get_vhost_net(nc
->peer
)) {
471 return vhost_net_get_features(get_vhost_net(nc
->peer
), features
);
474 static uint64_t virtio_net_bad_features(VirtIODevice
*vdev
)
476 uint64_t features
= 0;
478 /* Linux kernel 2.6.25. It understood MAC (as everyone must),
480 virtio_add_feature(&features
, VIRTIO_NET_F_MAC
);
481 virtio_add_feature(&features
, VIRTIO_NET_F_CSUM
);
482 virtio_add_feature(&features
, VIRTIO_NET_F_HOST_TSO4
);
483 virtio_add_feature(&features
, VIRTIO_NET_F_HOST_TSO6
);
484 virtio_add_feature(&features
, VIRTIO_NET_F_HOST_ECN
);
489 static void virtio_net_apply_guest_offloads(VirtIONet
*n
)
491 qemu_set_offload(qemu_get_queue(n
->nic
)->peer
,
492 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_CSUM
)),
493 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_TSO4
)),
494 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_TSO6
)),
495 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_ECN
)),
496 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_UFO
)));
499 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features
)
501 static const uint64_t guest_offloads_mask
=
502 (1ULL << VIRTIO_NET_F_GUEST_CSUM
) |
503 (1ULL << VIRTIO_NET_F_GUEST_TSO4
) |
504 (1ULL << VIRTIO_NET_F_GUEST_TSO6
) |
505 (1ULL << VIRTIO_NET_F_GUEST_ECN
) |
506 (1ULL << VIRTIO_NET_F_GUEST_UFO
);
508 return guest_offloads_mask
& features
;
511 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet
*n
)
513 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
514 return virtio_net_guest_offloads_by_features(vdev
->guest_features
);
517 static void virtio_net_set_features(VirtIODevice
*vdev
, uint64_t features
)
519 VirtIONet
*n
= VIRTIO_NET(vdev
);
522 virtio_net_set_multiqueue(n
,
523 __virtio_has_feature(features
, VIRTIO_NET_F_MQ
));
525 virtio_net_set_mrg_rx_bufs(n
,
526 __virtio_has_feature(features
,
527 VIRTIO_NET_F_MRG_RXBUF
),
528 __virtio_has_feature(features
,
529 VIRTIO_F_VERSION_1
));
531 if (n
->has_vnet_hdr
) {
532 n
->curr_guest_offloads
=
533 virtio_net_guest_offloads_by_features(features
);
534 virtio_net_apply_guest_offloads(n
);
537 for (i
= 0; i
< n
->max_queues
; i
++) {
538 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, i
);
540 if (!get_vhost_net(nc
->peer
)) {
543 vhost_net_ack_features(get_vhost_net(nc
->peer
), features
);
546 if (__virtio_has_feature(features
, VIRTIO_NET_F_CTRL_VLAN
)) {
547 memset(n
->vlans
, 0, MAX_VLAN
>> 3);
549 memset(n
->vlans
, 0xff, MAX_VLAN
>> 3);
553 static int virtio_net_handle_rx_mode(VirtIONet
*n
, uint8_t cmd
,
554 struct iovec
*iov
, unsigned int iov_cnt
)
558 NetClientState
*nc
= qemu_get_queue(n
->nic
);
560 s
= iov_to_buf(iov
, iov_cnt
, 0, &on
, sizeof(on
));
561 if (s
!= sizeof(on
)) {
562 return VIRTIO_NET_ERR
;
565 if (cmd
== VIRTIO_NET_CTRL_RX_PROMISC
) {
567 } else if (cmd
== VIRTIO_NET_CTRL_RX_ALLMULTI
) {
569 } else if (cmd
== VIRTIO_NET_CTRL_RX_ALLUNI
) {
571 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOMULTI
) {
573 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOUNI
) {
575 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOBCAST
) {
578 return VIRTIO_NET_ERR
;
583 return VIRTIO_NET_OK
;
586 static int virtio_net_handle_offloads(VirtIONet
*n
, uint8_t cmd
,
587 struct iovec
*iov
, unsigned int iov_cnt
)
589 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
593 if (!virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
)) {
594 return VIRTIO_NET_ERR
;
597 s
= iov_to_buf(iov
, iov_cnt
, 0, &offloads
, sizeof(offloads
));
598 if (s
!= sizeof(offloads
)) {
599 return VIRTIO_NET_ERR
;
602 if (cmd
== VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET
) {
603 uint64_t supported_offloads
;
605 if (!n
->has_vnet_hdr
) {
606 return VIRTIO_NET_ERR
;
609 supported_offloads
= virtio_net_supported_guest_offloads(n
);
610 if (offloads
& ~supported_offloads
) {
611 return VIRTIO_NET_ERR
;
614 n
->curr_guest_offloads
= offloads
;
615 virtio_net_apply_guest_offloads(n
);
617 return VIRTIO_NET_OK
;
619 return VIRTIO_NET_ERR
;
623 static int virtio_net_handle_mac(VirtIONet
*n
, uint8_t cmd
,
624 struct iovec
*iov
, unsigned int iov_cnt
)
626 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
627 struct virtio_net_ctrl_mac mac_data
;
629 NetClientState
*nc
= qemu_get_queue(n
->nic
);
631 if (cmd
== VIRTIO_NET_CTRL_MAC_ADDR_SET
) {
632 if (iov_size(iov
, iov_cnt
) != sizeof(n
->mac
)) {
633 return VIRTIO_NET_ERR
;
635 s
= iov_to_buf(iov
, iov_cnt
, 0, &n
->mac
, sizeof(n
->mac
));
636 assert(s
== sizeof(n
->mac
));
637 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->mac
);
640 return VIRTIO_NET_OK
;
643 if (cmd
!= VIRTIO_NET_CTRL_MAC_TABLE_SET
) {
644 return VIRTIO_NET_ERR
;
649 uint8_t uni_overflow
= 0;
650 uint8_t multi_overflow
= 0;
651 uint8_t *macs
= g_malloc0(MAC_TABLE_ENTRIES
* ETH_ALEN
);
653 s
= iov_to_buf(iov
, iov_cnt
, 0, &mac_data
.entries
,
654 sizeof(mac_data
.entries
));
655 mac_data
.entries
= virtio_ldl_p(vdev
, &mac_data
.entries
);
656 if (s
!= sizeof(mac_data
.entries
)) {
659 iov_discard_front(&iov
, &iov_cnt
, s
);
661 if (mac_data
.entries
* ETH_ALEN
> iov_size(iov
, iov_cnt
)) {
665 if (mac_data
.entries
<= MAC_TABLE_ENTRIES
) {
666 s
= iov_to_buf(iov
, iov_cnt
, 0, macs
,
667 mac_data
.entries
* ETH_ALEN
);
668 if (s
!= mac_data
.entries
* ETH_ALEN
) {
671 in_use
+= mac_data
.entries
;
676 iov_discard_front(&iov
, &iov_cnt
, mac_data
.entries
* ETH_ALEN
);
678 first_multi
= in_use
;
680 s
= iov_to_buf(iov
, iov_cnt
, 0, &mac_data
.entries
,
681 sizeof(mac_data
.entries
));
682 mac_data
.entries
= virtio_ldl_p(vdev
, &mac_data
.entries
);
683 if (s
!= sizeof(mac_data
.entries
)) {
687 iov_discard_front(&iov
, &iov_cnt
, s
);
689 if (mac_data
.entries
* ETH_ALEN
!= iov_size(iov
, iov_cnt
)) {
693 if (mac_data
.entries
<= MAC_TABLE_ENTRIES
- in_use
) {
694 s
= iov_to_buf(iov
, iov_cnt
, 0, &macs
[in_use
* ETH_ALEN
],
695 mac_data
.entries
* ETH_ALEN
);
696 if (s
!= mac_data
.entries
* ETH_ALEN
) {
699 in_use
+= mac_data
.entries
;
704 n
->mac_table
.in_use
= in_use
;
705 n
->mac_table
.first_multi
= first_multi
;
706 n
->mac_table
.uni_overflow
= uni_overflow
;
707 n
->mac_table
.multi_overflow
= multi_overflow
;
708 memcpy(n
->mac_table
.macs
, macs
, MAC_TABLE_ENTRIES
* ETH_ALEN
);
712 return VIRTIO_NET_OK
;
716 return VIRTIO_NET_ERR
;
719 static int virtio_net_handle_vlan_table(VirtIONet
*n
, uint8_t cmd
,
720 struct iovec
*iov
, unsigned int iov_cnt
)
722 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
725 NetClientState
*nc
= qemu_get_queue(n
->nic
);
727 s
= iov_to_buf(iov
, iov_cnt
, 0, &vid
, sizeof(vid
));
728 vid
= virtio_lduw_p(vdev
, &vid
);
729 if (s
!= sizeof(vid
)) {
730 return VIRTIO_NET_ERR
;
734 return VIRTIO_NET_ERR
;
736 if (cmd
== VIRTIO_NET_CTRL_VLAN_ADD
)
737 n
->vlans
[vid
>> 5] |= (1U << (vid
& 0x1f));
738 else if (cmd
== VIRTIO_NET_CTRL_VLAN_DEL
)
739 n
->vlans
[vid
>> 5] &= ~(1U << (vid
& 0x1f));
741 return VIRTIO_NET_ERR
;
745 return VIRTIO_NET_OK
;
748 static int virtio_net_handle_announce(VirtIONet
*n
, uint8_t cmd
,
749 struct iovec
*iov
, unsigned int iov_cnt
)
751 if (cmd
== VIRTIO_NET_CTRL_ANNOUNCE_ACK
&&
752 n
->status
& VIRTIO_NET_S_ANNOUNCE
) {
753 n
->status
&= ~VIRTIO_NET_S_ANNOUNCE
;
754 if (n
->announce_counter
) {
755 timer_mod(n
->announce_timer
,
756 qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
757 self_announce_delay(n
->announce_counter
));
759 return VIRTIO_NET_OK
;
761 return VIRTIO_NET_ERR
;
765 static int virtio_net_handle_mq(VirtIONet
*n
, uint8_t cmd
,
766 struct iovec
*iov
, unsigned int iov_cnt
)
768 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
769 struct virtio_net_ctrl_mq mq
;
773 s
= iov_to_buf(iov
, iov_cnt
, 0, &mq
, sizeof(mq
));
774 if (s
!= sizeof(mq
)) {
775 return VIRTIO_NET_ERR
;
778 if (cmd
!= VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
) {
779 return VIRTIO_NET_ERR
;
782 queues
= virtio_lduw_p(vdev
, &mq
.virtqueue_pairs
);
784 if (queues
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
785 queues
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
786 queues
> n
->max_queues
||
788 return VIRTIO_NET_ERR
;
791 n
->curr_queues
= queues
;
792 /* stop the backend before changing the number of queues to avoid handling a
794 virtio_net_set_status(vdev
, vdev
->status
);
795 virtio_net_set_queues(n
);
797 return VIRTIO_NET_OK
;
799 static void virtio_net_handle_ctrl(VirtIODevice
*vdev
, VirtQueue
*vq
)
801 VirtIONet
*n
= VIRTIO_NET(vdev
);
802 struct virtio_net_ctrl_hdr ctrl
;
803 virtio_net_ctrl_ack status
= VIRTIO_NET_ERR
;
804 VirtQueueElement elem
;
806 struct iovec
*iov
, *iov2
;
807 unsigned int iov_cnt
;
809 while (virtqueue_pop(vq
, &elem
)) {
810 if (iov_size(elem
.in_sg
, elem
.in_num
) < sizeof(status
) ||
811 iov_size(elem
.out_sg
, elem
.out_num
) < sizeof(ctrl
)) {
812 error_report("virtio-net ctrl missing headers");
816 iov_cnt
= elem
.out_num
;
817 iov2
= iov
= g_memdup(elem
.out_sg
, sizeof(struct iovec
) * elem
.out_num
);
818 s
= iov_to_buf(iov
, iov_cnt
, 0, &ctrl
, sizeof(ctrl
));
819 iov_discard_front(&iov
, &iov_cnt
, sizeof(ctrl
));
820 if (s
!= sizeof(ctrl
)) {
821 status
= VIRTIO_NET_ERR
;
822 } else if (ctrl
.class == VIRTIO_NET_CTRL_RX
) {
823 status
= virtio_net_handle_rx_mode(n
, ctrl
.cmd
, iov
, iov_cnt
);
824 } else if (ctrl
.class == VIRTIO_NET_CTRL_MAC
) {
825 status
= virtio_net_handle_mac(n
, ctrl
.cmd
, iov
, iov_cnt
);
826 } else if (ctrl
.class == VIRTIO_NET_CTRL_VLAN
) {
827 status
= virtio_net_handle_vlan_table(n
, ctrl
.cmd
, iov
, iov_cnt
);
828 } else if (ctrl
.class == VIRTIO_NET_CTRL_ANNOUNCE
) {
829 status
= virtio_net_handle_announce(n
, ctrl
.cmd
, iov
, iov_cnt
);
830 } else if (ctrl
.class == VIRTIO_NET_CTRL_MQ
) {
831 status
= virtio_net_handle_mq(n
, ctrl
.cmd
, iov
, iov_cnt
);
832 } else if (ctrl
.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS
) {
833 status
= virtio_net_handle_offloads(n
, ctrl
.cmd
, iov
, iov_cnt
);
836 s
= iov_from_buf(elem
.in_sg
, elem
.in_num
, 0, &status
, sizeof(status
));
837 assert(s
== sizeof(status
));
839 virtqueue_push(vq
, &elem
, sizeof(status
));
840 virtio_notify(vdev
, vq
);
847 static void virtio_net_handle_rx(VirtIODevice
*vdev
, VirtQueue
*vq
)
849 VirtIONet
*n
= VIRTIO_NET(vdev
);
850 int queue_index
= vq2q(virtio_get_queue_index(vq
));
852 qemu_flush_queued_packets(qemu_get_subqueue(n
->nic
, queue_index
));
855 static int virtio_net_can_receive(NetClientState
*nc
)
857 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
858 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
859 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
861 if (!vdev
->vm_running
) {
865 if (nc
->queue_index
>= n
->curr_queues
) {
869 if (!virtio_queue_ready(q
->rx_vq
) ||
870 !(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
877 static int virtio_net_has_buffers(VirtIONetQueue
*q
, int bufsize
)
880 if (virtio_queue_empty(q
->rx_vq
) ||
881 (n
->mergeable_rx_bufs
&&
882 !virtqueue_avail_bytes(q
->rx_vq
, bufsize
, 0))) {
883 virtio_queue_set_notification(q
->rx_vq
, 1);
885 /* To avoid a race condition where the guest has made some buffers
886 * available after the above check but before notification was
887 * enabled, check for available buffers again.
889 if (virtio_queue_empty(q
->rx_vq
) ||
890 (n
->mergeable_rx_bufs
&&
891 !virtqueue_avail_bytes(q
->rx_vq
, bufsize
, 0))) {
896 virtio_queue_set_notification(q
->rx_vq
, 0);
900 static void virtio_net_hdr_swap(VirtIODevice
*vdev
, struct virtio_net_hdr
*hdr
)
902 virtio_tswap16s(vdev
, &hdr
->hdr_len
);
903 virtio_tswap16s(vdev
, &hdr
->gso_size
);
904 virtio_tswap16s(vdev
, &hdr
->csum_start
);
905 virtio_tswap16s(vdev
, &hdr
->csum_offset
);
908 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
909 * it never finds out that the packets don't have valid checksums. This
910 * causes dhclient to get upset. Fedora's carried a patch for ages to
911 * fix this with Xen but it hasn't appeared in an upstream release of
914 * To avoid breaking existing guests, we catch udp packets and add
915 * checksums. This is terrible but it's better than hacking the guest
918 * N.B. if we introduce a zero-copy API, this operation is no longer free so
919 * we should provide a mechanism to disable it to avoid polluting the host
922 static void work_around_broken_dhclient(struct virtio_net_hdr
*hdr
,
923 uint8_t *buf
, size_t size
)
925 if ((hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) && /* missing csum */
926 (size
> 27 && size
< 1500) && /* normal sized MTU */
927 (buf
[12] == 0x08 && buf
[13] == 0x00) && /* ethertype == IPv4 */
928 (buf
[23] == 17) && /* ip.protocol == UDP */
929 (buf
[34] == 0 && buf
[35] == 67)) { /* udp.srcport == bootps */
930 net_checksum_calculate(buf
, size
);
931 hdr
->flags
&= ~VIRTIO_NET_HDR_F_NEEDS_CSUM
;
935 static void receive_header(VirtIONet
*n
, const struct iovec
*iov
, int iov_cnt
,
936 const void *buf
, size_t size
)
938 if (n
->has_vnet_hdr
) {
939 /* FIXME this cast is evil */
940 void *wbuf
= (void *)buf
;
941 work_around_broken_dhclient(wbuf
, wbuf
+ n
->host_hdr_len
,
942 size
- n
->host_hdr_len
);
943 virtio_net_hdr_swap(VIRTIO_DEVICE(n
), wbuf
);
944 iov_from_buf(iov
, iov_cnt
, 0, buf
, sizeof(struct virtio_net_hdr
));
946 struct virtio_net_hdr hdr
= {
948 .gso_type
= VIRTIO_NET_HDR_GSO_NONE
950 iov_from_buf(iov
, iov_cnt
, 0, &hdr
, sizeof hdr
);
954 static int receive_filter(VirtIONet
*n
, const uint8_t *buf
, int size
)
956 static const uint8_t bcast
[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
957 static const uint8_t vlan
[] = {0x81, 0x00};
958 uint8_t *ptr
= (uint8_t *)buf
;
964 ptr
+= n
->host_hdr_len
;
966 if (!memcmp(&ptr
[12], vlan
, sizeof(vlan
))) {
967 int vid
= be16_to_cpup((uint16_t *)(ptr
+ 14)) & 0xfff;
968 if (!(n
->vlans
[vid
>> 5] & (1U << (vid
& 0x1f))))
972 if (ptr
[0] & 1) { // multicast
973 if (!memcmp(ptr
, bcast
, sizeof(bcast
))) {
975 } else if (n
->nomulti
) {
977 } else if (n
->allmulti
|| n
->mac_table
.multi_overflow
) {
981 for (i
= n
->mac_table
.first_multi
; i
< n
->mac_table
.in_use
; i
++) {
982 if (!memcmp(ptr
, &n
->mac_table
.macs
[i
* ETH_ALEN
], ETH_ALEN
)) {
989 } else if (n
->alluni
|| n
->mac_table
.uni_overflow
) {
991 } else if (!memcmp(ptr
, n
->mac
, ETH_ALEN
)) {
995 for (i
= 0; i
< n
->mac_table
.first_multi
; i
++) {
996 if (!memcmp(ptr
, &n
->mac_table
.macs
[i
* ETH_ALEN
], ETH_ALEN
)) {
1005 static ssize_t
virtio_net_receive(NetClientState
*nc
, const uint8_t *buf
, size_t size
)
1007 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
1008 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
1009 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1010 struct iovec mhdr_sg
[VIRTQUEUE_MAX_SIZE
];
1011 struct virtio_net_hdr_mrg_rxbuf mhdr
;
1012 unsigned mhdr_cnt
= 0;
1013 size_t offset
, i
, guest_offset
;
1015 if (!virtio_net_can_receive(nc
)) {
1019 /* hdr_len refers to the header we supply to the guest */
1020 if (!virtio_net_has_buffers(q
, size
+ n
->guest_hdr_len
- n
->host_hdr_len
)) {
1024 if (!receive_filter(n
, buf
, size
))
1029 while (offset
< size
) {
1030 VirtQueueElement elem
;
1032 const struct iovec
*sg
= elem
.in_sg
;
1036 if (virtqueue_pop(q
->rx_vq
, &elem
) == 0) {
1039 error_report("virtio-net unexpected empty queue: "
1040 "i %zd mergeable %d offset %zd, size %zd, "
1041 "guest hdr len %zd, host hdr len %zd "
1042 "guest features 0x%" PRIx64
,
1043 i
, n
->mergeable_rx_bufs
, offset
, size
,
1044 n
->guest_hdr_len
, n
->host_hdr_len
,
1045 vdev
->guest_features
);
1049 if (elem
.in_num
< 1) {
1050 error_report("virtio-net receive queue contains no in buffers");
1055 assert(offset
== 0);
1056 if (n
->mergeable_rx_bufs
) {
1057 mhdr_cnt
= iov_copy(mhdr_sg
, ARRAY_SIZE(mhdr_sg
),
1059 offsetof(typeof(mhdr
), num_buffers
),
1060 sizeof(mhdr
.num_buffers
));
1063 receive_header(n
, sg
, elem
.in_num
, buf
, size
);
1064 offset
= n
->host_hdr_len
;
1065 total
+= n
->guest_hdr_len
;
1066 guest_offset
= n
->guest_hdr_len
;
1071 /* copy in packet. ugh */
1072 len
= iov_from_buf(sg
, elem
.in_num
, guest_offset
,
1073 buf
+ offset
, size
- offset
);
1076 /* If buffers can't be merged, at this point we
1077 * must have consumed the complete packet.
1078 * Otherwise, drop it. */
1079 if (!n
->mergeable_rx_bufs
&& offset
< size
) {
1081 error_report("virtio-net truncated non-mergeable packet: "
1082 "i %zd mergeable %d offset %zd, size %zd, "
1083 "guest hdr len %zd, host hdr len %zd",
1084 i
, n
->mergeable_rx_bufs
,
1085 offset
, size
, n
->guest_hdr_len
, n
->host_hdr_len
);
1090 /* signal other side */
1091 virtqueue_fill(q
->rx_vq
, &elem
, total
, i
++);
1095 virtio_stw_p(vdev
, &mhdr
.num_buffers
, i
);
1096 iov_from_buf(mhdr_sg
, mhdr_cnt
,
1098 &mhdr
.num_buffers
, sizeof mhdr
.num_buffers
);
1101 virtqueue_flush(q
->rx_vq
, i
);
1102 virtio_notify(vdev
, q
->rx_vq
);
1107 static int32_t virtio_net_flush_tx(VirtIONetQueue
*q
);
1109 static void virtio_net_tx_complete(NetClientState
*nc
, ssize_t len
)
1111 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
1112 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
1113 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1115 virtqueue_push(q
->tx_vq
, &q
->async_tx
.elem
, 0);
1116 virtio_notify(vdev
, q
->tx_vq
);
1118 q
->async_tx
.elem
.out_num
= q
->async_tx
.len
= 0;
1120 virtio_queue_set_notification(q
->tx_vq
, 1);
1121 virtio_net_flush_tx(q
);
1125 static int32_t virtio_net_flush_tx(VirtIONetQueue
*q
)
1127 VirtIONet
*n
= q
->n
;
1128 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1129 VirtQueueElement elem
;
1130 int32_t num_packets
= 0;
1131 int queue_index
= vq2q(virtio_get_queue_index(q
->tx_vq
));
1132 if (!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
1136 if (q
->async_tx
.elem
.out_num
) {
1137 virtio_queue_set_notification(q
->tx_vq
, 0);
1141 while (virtqueue_pop(q
->tx_vq
, &elem
)) {
1143 unsigned int out_num
= elem
.out_num
;
1144 struct iovec
*out_sg
= &elem
.out_sg
[0];
1145 struct iovec sg
[VIRTQUEUE_MAX_SIZE
];
1148 error_report("virtio-net header not in first element");
1152 if (n
->has_vnet_hdr
) {
1153 if (out_sg
[0].iov_len
< n
->guest_hdr_len
) {
1154 error_report("virtio-net header incorrect");
1157 virtio_net_hdr_swap(vdev
, (void *) out_sg
[0].iov_base
);
1161 * If host wants to see the guest header as is, we can
1162 * pass it on unchanged. Otherwise, copy just the parts
1163 * that host is interested in.
1165 assert(n
->host_hdr_len
<= n
->guest_hdr_len
);
1166 if (n
->host_hdr_len
!= n
->guest_hdr_len
) {
1167 unsigned sg_num
= iov_copy(sg
, ARRAY_SIZE(sg
),
1169 0, n
->host_hdr_len
);
1170 sg_num
+= iov_copy(sg
+ sg_num
, ARRAY_SIZE(sg
) - sg_num
,
1172 n
->guest_hdr_len
, -1);
1177 len
= n
->guest_hdr_len
;
1179 ret
= qemu_sendv_packet_async(qemu_get_subqueue(n
->nic
, queue_index
),
1180 out_sg
, out_num
, virtio_net_tx_complete
);
1182 virtio_queue_set_notification(q
->tx_vq
, 0);
1183 q
->async_tx
.elem
= elem
;
1184 q
->async_tx
.len
= len
;
1190 virtqueue_push(q
->tx_vq
, &elem
, 0);
1191 virtio_notify(vdev
, q
->tx_vq
);
1193 if (++num_packets
>= n
->tx_burst
) {
1200 static void virtio_net_handle_tx_timer(VirtIODevice
*vdev
, VirtQueue
*vq
)
1202 VirtIONet
*n
= VIRTIO_NET(vdev
);
1203 VirtIONetQueue
*q
= &n
->vqs
[vq2q(virtio_get_queue_index(vq
))];
1205 /* This happens when device was stopped but VCPU wasn't. */
1206 if (!vdev
->vm_running
) {
1211 if (q
->tx_waiting
) {
1212 virtio_queue_set_notification(vq
, 1);
1213 timer_del(q
->tx_timer
);
1215 virtio_net_flush_tx(q
);
1217 timer_mod(q
->tx_timer
,
1218 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) + n
->tx_timeout
);
1220 virtio_queue_set_notification(vq
, 0);
1224 static void virtio_net_handle_tx_bh(VirtIODevice
*vdev
, VirtQueue
*vq
)
1226 VirtIONet
*n
= VIRTIO_NET(vdev
);
1227 VirtIONetQueue
*q
= &n
->vqs
[vq2q(virtio_get_queue_index(vq
))];
1229 if (unlikely(q
->tx_waiting
)) {
1233 /* This happens when device was stopped but VCPU wasn't. */
1234 if (!vdev
->vm_running
) {
1237 virtio_queue_set_notification(vq
, 0);
1238 qemu_bh_schedule(q
->tx_bh
);
1241 static void virtio_net_tx_timer(void *opaque
)
1243 VirtIONetQueue
*q
= opaque
;
1244 VirtIONet
*n
= q
->n
;
1245 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1246 /* This happens when device was stopped but BH wasn't. */
1247 if (!vdev
->vm_running
) {
1248 /* Make sure tx waiting is set, so we'll run when restarted. */
1249 assert(q
->tx_waiting
);
1255 /* Just in case the driver is not ready on more */
1256 if (!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
1260 virtio_queue_set_notification(q
->tx_vq
, 1);
1261 virtio_net_flush_tx(q
);
1264 static void virtio_net_tx_bh(void *opaque
)
1266 VirtIONetQueue
*q
= opaque
;
1267 VirtIONet
*n
= q
->n
;
1268 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1271 /* This happens when device was stopped but BH wasn't. */
1272 if (!vdev
->vm_running
) {
1273 /* Make sure tx waiting is set, so we'll run when restarted. */
1274 assert(q
->tx_waiting
);
1280 /* Just in case the driver is not ready on more */
1281 if (unlikely(!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
))) {
1285 ret
= virtio_net_flush_tx(q
);
1286 if (ret
== -EBUSY
) {
1287 return; /* Notification re-enable handled by tx_complete */
1290 /* If we flush a full burst of packets, assume there are
1291 * more coming and immediately reschedule */
1292 if (ret
>= n
->tx_burst
) {
1293 qemu_bh_schedule(q
->tx_bh
);
1298 /* If less than a full burst, re-enable notification and flush
1299 * anything that may have come in while we weren't looking. If
1300 * we find something, assume the guest is still active and reschedule */
1301 virtio_queue_set_notification(q
->tx_vq
, 1);
1302 if (virtio_net_flush_tx(q
) > 0) {
1303 virtio_queue_set_notification(q
->tx_vq
, 0);
1304 qemu_bh_schedule(q
->tx_bh
);
1309 static void virtio_net_set_multiqueue(VirtIONet
*n
, int multiqueue
)
1311 n
->multiqueue
= multiqueue
;
1313 virtio_net_set_queues(n
);
1316 static void virtio_net_save(QEMUFile
*f
, void *opaque
)
1318 VirtIONet
*n
= opaque
;
1319 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1321 /* At this point, backend must be stopped, otherwise
1322 * it might keep writing to memory. */
1323 assert(!n
->vhost_started
);
1324 virtio_save(vdev
, f
);
1327 static void virtio_net_save_device(VirtIODevice
*vdev
, QEMUFile
*f
)
1329 VirtIONet
*n
= VIRTIO_NET(vdev
);
1332 qemu_put_buffer(f
, n
->mac
, ETH_ALEN
);
1333 qemu_put_be32(f
, n
->vqs
[0].tx_waiting
);
1334 qemu_put_be32(f
, n
->mergeable_rx_bufs
);
1335 qemu_put_be16(f
, n
->status
);
1336 qemu_put_byte(f
, n
->promisc
);
1337 qemu_put_byte(f
, n
->allmulti
);
1338 qemu_put_be32(f
, n
->mac_table
.in_use
);
1339 qemu_put_buffer(f
, n
->mac_table
.macs
, n
->mac_table
.in_use
* ETH_ALEN
);
1340 qemu_put_buffer(f
, (uint8_t *)n
->vlans
, MAX_VLAN
>> 3);
1341 qemu_put_be32(f
, n
->has_vnet_hdr
);
1342 qemu_put_byte(f
, n
->mac_table
.multi_overflow
);
1343 qemu_put_byte(f
, n
->mac_table
.uni_overflow
);
1344 qemu_put_byte(f
, n
->alluni
);
1345 qemu_put_byte(f
, n
->nomulti
);
1346 qemu_put_byte(f
, n
->nouni
);
1347 qemu_put_byte(f
, n
->nobcast
);
1348 qemu_put_byte(f
, n
->has_ufo
);
1349 if (n
->max_queues
> 1) {
1350 qemu_put_be16(f
, n
->max_queues
);
1351 qemu_put_be16(f
, n
->curr_queues
);
1352 for (i
= 1; i
< n
->curr_queues
; i
++) {
1353 qemu_put_be32(f
, n
->vqs
[i
].tx_waiting
);
1357 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
)) {
1358 qemu_put_be64(f
, n
->curr_guest_offloads
);
1362 static int virtio_net_load(QEMUFile
*f
, void *opaque
, int version_id
)
1364 VirtIONet
*n
= opaque
;
1365 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1367 if (version_id
< 2 || version_id
> VIRTIO_NET_VM_VERSION
)
1370 return virtio_load(vdev
, f
, version_id
);
1373 static int virtio_net_load_device(VirtIODevice
*vdev
, QEMUFile
*f
,
1376 VirtIONet
*n
= VIRTIO_NET(vdev
);
1379 qemu_get_buffer(f
, n
->mac
, ETH_ALEN
);
1380 n
->vqs
[0].tx_waiting
= qemu_get_be32(f
);
1382 virtio_net_set_mrg_rx_bufs(n
, qemu_get_be32(f
),
1383 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
));
1385 if (version_id
>= 3)
1386 n
->status
= qemu_get_be16(f
);
1388 if (version_id
>= 4) {
1389 if (version_id
< 8) {
1390 n
->promisc
= qemu_get_be32(f
);
1391 n
->allmulti
= qemu_get_be32(f
);
1393 n
->promisc
= qemu_get_byte(f
);
1394 n
->allmulti
= qemu_get_byte(f
);
1398 if (version_id
>= 5) {
1399 n
->mac_table
.in_use
= qemu_get_be32(f
);
1400 /* MAC_TABLE_ENTRIES may be different from the saved image */
1401 if (n
->mac_table
.in_use
<= MAC_TABLE_ENTRIES
) {
1402 qemu_get_buffer(f
, n
->mac_table
.macs
,
1403 n
->mac_table
.in_use
* ETH_ALEN
);
1407 /* Overflow detected - can happen if source has a larger MAC table.
1408 * We simply set overflow flag so there's no need to maintain the
1409 * table of addresses, discard them all.
1410 * Note: 64 bit math to avoid integer overflow.
1412 for (i
= 0; i
< (int64_t)n
->mac_table
.in_use
* ETH_ALEN
; ++i
) {
1415 n
->mac_table
.multi_overflow
= n
->mac_table
.uni_overflow
= 1;
1416 n
->mac_table
.in_use
= 0;
1420 if (version_id
>= 6)
1421 qemu_get_buffer(f
, (uint8_t *)n
->vlans
, MAX_VLAN
>> 3);
1423 if (version_id
>= 7) {
1424 if (qemu_get_be32(f
) && !peer_has_vnet_hdr(n
)) {
1425 error_report("virtio-net: saved image requires vnet_hdr=on");
1430 if (version_id
>= 9) {
1431 n
->mac_table
.multi_overflow
= qemu_get_byte(f
);
1432 n
->mac_table
.uni_overflow
= qemu_get_byte(f
);
1435 if (version_id
>= 10) {
1436 n
->alluni
= qemu_get_byte(f
);
1437 n
->nomulti
= qemu_get_byte(f
);
1438 n
->nouni
= qemu_get_byte(f
);
1439 n
->nobcast
= qemu_get_byte(f
);
1442 if (version_id
>= 11) {
1443 if (qemu_get_byte(f
) && !peer_has_ufo(n
)) {
1444 error_report("virtio-net: saved image requires TUN_F_UFO support");
1449 if (n
->max_queues
> 1) {
1450 if (n
->max_queues
!= qemu_get_be16(f
)) {
1451 error_report("virtio-net: different max_queues ");
1455 n
->curr_queues
= qemu_get_be16(f
);
1456 if (n
->curr_queues
> n
->max_queues
) {
1457 error_report("virtio-net: curr_queues %x > max_queues %x",
1458 n
->curr_queues
, n
->max_queues
);
1461 for (i
= 1; i
< n
->curr_queues
; i
++) {
1462 n
->vqs
[i
].tx_waiting
= qemu_get_be32(f
);
1466 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
)) {
1467 n
->curr_guest_offloads
= qemu_get_be64(f
);
1469 n
->curr_guest_offloads
= virtio_net_supported_guest_offloads(n
);
1472 if (peer_has_vnet_hdr(n
)) {
1473 virtio_net_apply_guest_offloads(n
);
1476 virtio_net_set_queues(n
);
1478 /* Find the first multicast entry in the saved MAC filter */
1479 for (i
= 0; i
< n
->mac_table
.in_use
; i
++) {
1480 if (n
->mac_table
.macs
[i
* ETH_ALEN
] & 1) {
1484 n
->mac_table
.first_multi
= i
;
1486 /* nc.link_down can't be migrated, so infer link_down according
1487 * to link status bit in n->status */
1488 link_down
= (n
->status
& VIRTIO_NET_S_LINK_UP
) == 0;
1489 for (i
= 0; i
< n
->max_queues
; i
++) {
1490 qemu_get_subqueue(n
->nic
, i
)->link_down
= link_down
;
1493 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
) &&
1494 virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
)) {
1495 n
->announce_counter
= SELF_ANNOUNCE_ROUNDS
;
1496 timer_mod(n
->announce_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
));
1502 static NetClientInfo net_virtio_info
= {
1503 .type
= NET_CLIENT_OPTIONS_KIND_NIC
,
1504 .size
= sizeof(NICState
),
1505 .can_receive
= virtio_net_can_receive
,
1506 .receive
= virtio_net_receive
,
1507 .link_status_changed
= virtio_net_set_link_status
,
1508 .query_rx_filter
= virtio_net_query_rxfilter
,
1511 static bool virtio_net_guest_notifier_pending(VirtIODevice
*vdev
, int idx
)
1513 VirtIONet
*n
= VIRTIO_NET(vdev
);
1514 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, vq2q(idx
));
1515 assert(n
->vhost_started
);
1516 return vhost_net_virtqueue_pending(get_vhost_net(nc
->peer
), idx
);
1519 static void virtio_net_guest_notifier_mask(VirtIODevice
*vdev
, int idx
,
1522 VirtIONet
*n
= VIRTIO_NET(vdev
);
1523 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, vq2q(idx
));
1524 assert(n
->vhost_started
);
1525 vhost_net_virtqueue_mask(get_vhost_net(nc
->peer
),
1529 static void virtio_net_set_config_size(VirtIONet
*n
, uint64_t host_features
)
1531 int i
, config_size
= 0;
1532 virtio_add_feature(&host_features
, VIRTIO_NET_F_MAC
);
1533 for (i
= 0; feature_sizes
[i
].flags
!= 0; i
++) {
1534 if (host_features
& feature_sizes
[i
].flags
) {
1535 config_size
= MAX(feature_sizes
[i
].end
, config_size
);
1538 n
->config_size
= config_size
;
1541 void virtio_net_set_netclient_name(VirtIONet
*n
, const char *name
,
1545 * The name can be NULL, the netclient name will be type.x.
1547 assert(type
!= NULL
);
1549 g_free(n
->netclient_name
);
1550 g_free(n
->netclient_type
);
1551 n
->netclient_name
= g_strdup(name
);
1552 n
->netclient_type
= g_strdup(type
);
1555 static void virtio_net_device_realize(DeviceState
*dev
, Error
**errp
)
1557 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
1558 VirtIONet
*n
= VIRTIO_NET(dev
);
1562 virtio_net_set_config_size(n
, n
->host_features
);
1563 virtio_init(vdev
, "virtio-net", VIRTIO_ID_NET
, n
->config_size
);
1565 n
->max_queues
= MAX(n
->nic_conf
.peers
.queues
, 1);
1566 if (n
->max_queues
* 2 + 1 > VIRTIO_QUEUE_MAX
) {
1567 error_setg(errp
, "Invalid number of queues (= %" PRIu32
"), "
1568 "must be a positive integer less than %d.",
1569 n
->max_queues
, (VIRTIO_QUEUE_MAX
- 1) / 2);
1570 virtio_cleanup(vdev
);
1573 n
->vqs
= g_malloc0(sizeof(VirtIONetQueue
) * n
->max_queues
);
1575 n
->tx_timeout
= n
->net_conf
.txtimer
;
1577 if (n
->net_conf
.tx
&& strcmp(n
->net_conf
.tx
, "timer")
1578 && strcmp(n
->net_conf
.tx
, "bh")) {
1579 error_report("virtio-net: "
1580 "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1582 error_report("Defaulting to \"bh\"");
1585 for (i
= 0; i
< n
->max_queues
; i
++) {
1586 n
->vqs
[i
].rx_vq
= virtio_add_queue(vdev
, 256, virtio_net_handle_rx
);
1587 if (n
->net_conf
.tx
&& !strcmp(n
->net_conf
.tx
, "timer")) {
1589 virtio_add_queue(vdev
, 256, virtio_net_handle_tx_timer
);
1590 n
->vqs
[i
].tx_timer
= timer_new_ns(QEMU_CLOCK_VIRTUAL
,
1591 virtio_net_tx_timer
,
1595 virtio_add_queue(vdev
, 256, virtio_net_handle_tx_bh
);
1596 n
->vqs
[i
].tx_bh
= qemu_bh_new(virtio_net_tx_bh
, &n
->vqs
[i
]);
1599 n
->vqs
[i
].tx_waiting
= 0;
1603 n
->ctrl_vq
= virtio_add_queue(vdev
, 64, virtio_net_handle_ctrl
);
1604 qemu_macaddr_default_if_unset(&n
->nic_conf
.macaddr
);
1605 memcpy(&n
->mac
[0], &n
->nic_conf
.macaddr
, sizeof(n
->mac
));
1606 n
->status
= VIRTIO_NET_S_LINK_UP
;
1607 n
->announce_timer
= timer_new_ms(QEMU_CLOCK_VIRTUAL
,
1608 virtio_net_announce_timer
, n
);
1610 if (n
->netclient_type
) {
1612 * Happen when virtio_net_set_netclient_name has been called.
1614 n
->nic
= qemu_new_nic(&net_virtio_info
, &n
->nic_conf
,
1615 n
->netclient_type
, n
->netclient_name
, n
);
1617 n
->nic
= qemu_new_nic(&net_virtio_info
, &n
->nic_conf
,
1618 object_get_typename(OBJECT(dev
)), dev
->id
, n
);
1621 peer_test_vnet_hdr(n
);
1622 if (peer_has_vnet_hdr(n
)) {
1623 for (i
= 0; i
< n
->max_queues
; i
++) {
1624 qemu_using_vnet_hdr(qemu_get_subqueue(n
->nic
, i
)->peer
, true);
1626 n
->host_hdr_len
= sizeof(struct virtio_net_hdr
);
1628 n
->host_hdr_len
= 0;
1631 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->nic_conf
.macaddr
.a
);
1633 n
->vqs
[0].tx_waiting
= 0;
1634 n
->tx_burst
= n
->net_conf
.txburst
;
1635 virtio_net_set_mrg_rx_bufs(n
, 0, 0);
1636 n
->promisc
= 1; /* for compatibility */
1638 n
->mac_table
.macs
= g_malloc0(MAC_TABLE_ENTRIES
* ETH_ALEN
);
1640 n
->vlans
= g_malloc0(MAX_VLAN
>> 3);
1642 nc
= qemu_get_queue(n
->nic
);
1643 nc
->rxfilter_notify_enabled
= 1;
1646 register_savevm(dev
, "virtio-net", -1, VIRTIO_NET_VM_VERSION
,
1647 virtio_net_save
, virtio_net_load
, n
);
1650 static void virtio_net_device_unrealize(DeviceState
*dev
, Error
**errp
)
1652 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
1653 VirtIONet
*n
= VIRTIO_NET(dev
);
1656 /* This will stop vhost backend if appropriate. */
1657 virtio_net_set_status(vdev
, 0);
1659 unregister_savevm(dev
, "virtio-net", n
);
1661 g_free(n
->netclient_name
);
1662 n
->netclient_name
= NULL
;
1663 g_free(n
->netclient_type
);
1664 n
->netclient_type
= NULL
;
1666 g_free(n
->mac_table
.macs
);
1669 for (i
= 0; i
< n
->max_queues
; i
++) {
1670 VirtIONetQueue
*q
= &n
->vqs
[i
];
1671 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, i
);
1673 qemu_purge_queued_packets(nc
);
1676 timer_del(q
->tx_timer
);
1677 timer_free(q
->tx_timer
);
1678 } else if (q
->tx_bh
) {
1679 qemu_bh_delete(q
->tx_bh
);
1683 timer_del(n
->announce_timer
);
1684 timer_free(n
->announce_timer
);
1686 qemu_del_nic(n
->nic
);
1687 virtio_cleanup(vdev
);
1690 static void virtio_net_instance_init(Object
*obj
)
1692 VirtIONet
*n
= VIRTIO_NET(obj
);
1695 * The default config_size is sizeof(struct virtio_net_config).
1696 * Can be overriden with virtio_net_set_config_size.
1698 n
->config_size
= sizeof(struct virtio_net_config
);
1699 device_add_bootindex_property(obj
, &n
->nic_conf
.bootindex
,
1700 "bootindex", "/ethernet-phy@0",
1704 static Property virtio_net_properties
[] = {
1705 DEFINE_PROP_BIT("any_layout", VirtIONet
, host_features
,
1706 VIRTIO_F_ANY_LAYOUT
, true),
1707 DEFINE_PROP_BIT("csum", VirtIONet
, host_features
, VIRTIO_NET_F_CSUM
, true),
1708 DEFINE_PROP_BIT("guest_csum", VirtIONet
, host_features
,
1709 VIRTIO_NET_F_GUEST_CSUM
, true),
1710 DEFINE_PROP_BIT("gso", VirtIONet
, host_features
, VIRTIO_NET_F_GSO
, true),
1711 DEFINE_PROP_BIT("guest_tso4", VirtIONet
, host_features
,
1712 VIRTIO_NET_F_GUEST_TSO4
, true),
1713 DEFINE_PROP_BIT("guest_tso6", VirtIONet
, host_features
,
1714 VIRTIO_NET_F_GUEST_TSO6
, true),
1715 DEFINE_PROP_BIT("guest_ecn", VirtIONet
, host_features
,
1716 VIRTIO_NET_F_GUEST_ECN
, true),
1717 DEFINE_PROP_BIT("guest_ufo", VirtIONet
, host_features
,
1718 VIRTIO_NET_F_GUEST_UFO
, true),
1719 DEFINE_PROP_BIT("guest_announce", VirtIONet
, host_features
,
1720 VIRTIO_NET_F_GUEST_ANNOUNCE
, true),
1721 DEFINE_PROP_BIT("host_tso4", VirtIONet
, host_features
,
1722 VIRTIO_NET_F_HOST_TSO4
, true),
1723 DEFINE_PROP_BIT("host_tso6", VirtIONet
, host_features
,
1724 VIRTIO_NET_F_HOST_TSO6
, true),
1725 DEFINE_PROP_BIT("host_ecn", VirtIONet
, host_features
,
1726 VIRTIO_NET_F_HOST_ECN
, true),
1727 DEFINE_PROP_BIT("host_ufo", VirtIONet
, host_features
,
1728 VIRTIO_NET_F_HOST_UFO
, true),
1729 DEFINE_PROP_BIT("mrg_rxbuf", VirtIONet
, host_features
,
1730 VIRTIO_NET_F_MRG_RXBUF
, true),
1731 DEFINE_PROP_BIT("status", VirtIONet
, host_features
,
1732 VIRTIO_NET_F_STATUS
, true),
1733 DEFINE_PROP_BIT("ctrl_vq", VirtIONet
, host_features
,
1734 VIRTIO_NET_F_CTRL_VQ
, true),
1735 DEFINE_PROP_BIT("ctrl_rx", VirtIONet
, host_features
,
1736 VIRTIO_NET_F_CTRL_RX
, true),
1737 DEFINE_PROP_BIT("ctrl_vlan", VirtIONet
, host_features
,
1738 VIRTIO_NET_F_CTRL_VLAN
, true),
1739 DEFINE_PROP_BIT("ctrl_rx_extra", VirtIONet
, host_features
,
1740 VIRTIO_NET_F_CTRL_RX_EXTRA
, true),
1741 DEFINE_PROP_BIT("ctrl_mac_addr", VirtIONet
, host_features
,
1742 VIRTIO_NET_F_CTRL_MAC_ADDR
, true),
1743 DEFINE_PROP_BIT("ctrl_guest_offloads", VirtIONet
, host_features
,
1744 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
, true),
1745 DEFINE_PROP_BIT("mq", VirtIONet
, host_features
, VIRTIO_NET_F_MQ
, false),
1746 DEFINE_NIC_PROPERTIES(VirtIONet
, nic_conf
),
1747 DEFINE_PROP_UINT32("x-txtimer", VirtIONet
, net_conf
.txtimer
,
1749 DEFINE_PROP_INT32("x-txburst", VirtIONet
, net_conf
.txburst
, TX_BURST
),
1750 DEFINE_PROP_STRING("tx", VirtIONet
, net_conf
.tx
),
1751 DEFINE_PROP_END_OF_LIST(),
1754 static void virtio_net_class_init(ObjectClass
*klass
, void *data
)
1756 DeviceClass
*dc
= DEVICE_CLASS(klass
);
1757 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
1759 dc
->props
= virtio_net_properties
;
1760 set_bit(DEVICE_CATEGORY_NETWORK
, dc
->categories
);
1761 vdc
->realize
= virtio_net_device_realize
;
1762 vdc
->unrealize
= virtio_net_device_unrealize
;
1763 vdc
->get_config
= virtio_net_get_config
;
1764 vdc
->set_config
= virtio_net_set_config
;
1765 vdc
->get_features
= virtio_net_get_features
;
1766 vdc
->set_features
= virtio_net_set_features
;
1767 vdc
->bad_features
= virtio_net_bad_features
;
1768 vdc
->reset
= virtio_net_reset
;
1769 vdc
->set_status
= virtio_net_set_status
;
1770 vdc
->guest_notifier_mask
= virtio_net_guest_notifier_mask
;
1771 vdc
->guest_notifier_pending
= virtio_net_guest_notifier_pending
;
1772 vdc
->load
= virtio_net_load_device
;
1773 vdc
->save
= virtio_net_save_device
;
1776 static const TypeInfo virtio_net_info
= {
1777 .name
= TYPE_VIRTIO_NET
,
1778 .parent
= TYPE_VIRTIO_DEVICE
,
1779 .instance_size
= sizeof(VirtIONet
),
1780 .instance_init
= virtio_net_instance_init
,
1781 .class_init
= virtio_net_class_init
,
1784 static void virtio_register_types(void)
1786 type_register_static(&virtio_net_info
);
1789 type_init(virtio_register_types
)