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.
14 #include "qemu/osdep.h"
15 #include "qemu/atomic.h"
17 #include "qemu/main-loop.h"
18 #include "qemu/module.h"
19 #include "hw/virtio/virtio.h"
21 #include "net/checksum.h"
23 #include "qemu/error-report.h"
24 #include "qemu/timer.h"
25 #include "qemu/option.h"
26 #include "qemu/option_int.h"
27 #include "qemu/config-file.h"
28 #include "qapi/qmp/qdict.h"
29 #include "hw/virtio/virtio-net.h"
30 #include "net/vhost_net.h"
31 #include "net/announce.h"
32 #include "hw/virtio/virtio-bus.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-events-net.h"
35 #include "hw/qdev-properties.h"
36 #include "qapi/qapi-types-migration.h"
37 #include "qapi/qapi-events-migration.h"
38 #include "hw/virtio/virtio-access.h"
39 #include "migration/misc.h"
40 #include "standard-headers/linux/ethtool.h"
41 #include "sysemu/sysemu.h"
43 #include "monitor/qdev.h"
44 #include "hw/pci/pci.h"
46 #define VIRTIO_NET_VM_VERSION 11
48 #define MAC_TABLE_ENTRIES 64
49 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */
51 /* previously fixed value */
52 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
53 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
55 /* for now, only allow larger queues; with virtio-1, guest can downsize */
56 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
57 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
59 #define VIRTIO_NET_IP4_ADDR_SIZE 8 /* ipv4 saddr + daddr */
61 #define VIRTIO_NET_TCP_FLAG 0x3F
62 #define VIRTIO_NET_TCP_HDR_LENGTH 0xF000
64 /* IPv4 max payload, 16 bits in the header */
65 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
66 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
68 /* header length value in ip header without option */
69 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
71 #define VIRTIO_NET_IP6_ADDR_SIZE 32 /* ipv6 saddr + daddr */
72 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
74 /* Purge coalesced packets timer interval, This value affects the performance
75 a lot, and should be tuned carefully, '300000'(300us) is the recommended
76 value to pass the WHQL test, '50000' can gain 2x netperf throughput with
78 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
80 /* temporary until standard header include it */
81 #if !defined(VIRTIO_NET_HDR_F_RSC_INFO)
83 #define VIRTIO_NET_HDR_F_RSC_INFO 4 /* rsc_ext data in csum_ fields */
84 #define VIRTIO_NET_F_RSC_EXT 61
86 static inline __virtio16
*virtio_net_rsc_ext_num_packets(
87 struct virtio_net_hdr
*hdr
)
89 return &hdr
->csum_start
;
92 static inline __virtio16
*virtio_net_rsc_ext_num_dupacks(
93 struct virtio_net_hdr
*hdr
)
95 return &hdr
->csum_offset
;
100 static VirtIOFeature feature_sizes
[] = {
101 {.flags
= 1ULL << VIRTIO_NET_F_MAC
,
102 .end
= endof(struct virtio_net_config
, mac
)},
103 {.flags
= 1ULL << VIRTIO_NET_F_STATUS
,
104 .end
= endof(struct virtio_net_config
, status
)},
105 {.flags
= 1ULL << VIRTIO_NET_F_MQ
,
106 .end
= endof(struct virtio_net_config
, max_virtqueue_pairs
)},
107 {.flags
= 1ULL << VIRTIO_NET_F_MTU
,
108 .end
= endof(struct virtio_net_config
, mtu
)},
109 {.flags
= 1ULL << VIRTIO_NET_F_SPEED_DUPLEX
,
110 .end
= endof(struct virtio_net_config
, duplex
)},
114 static VirtIONetQueue
*virtio_net_get_subqueue(NetClientState
*nc
)
116 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
118 return &n
->vqs
[nc
->queue_index
];
121 static int vq2q(int queue_index
)
123 return queue_index
/ 2;
127 * - we could suppress RX interrupt if we were so inclined.
130 static void virtio_net_get_config(VirtIODevice
*vdev
, uint8_t *config
)
132 VirtIONet
*n
= VIRTIO_NET(vdev
);
133 struct virtio_net_config netcfg
;
135 virtio_stw_p(vdev
, &netcfg
.status
, n
->status
);
136 virtio_stw_p(vdev
, &netcfg
.max_virtqueue_pairs
, n
->max_queues
);
137 virtio_stw_p(vdev
, &netcfg
.mtu
, n
->net_conf
.mtu
);
138 memcpy(netcfg
.mac
, n
->mac
, ETH_ALEN
);
139 virtio_stl_p(vdev
, &netcfg
.speed
, n
->net_conf
.speed
);
140 netcfg
.duplex
= n
->net_conf
.duplex
;
141 memcpy(config
, &netcfg
, n
->config_size
);
144 static void virtio_net_set_config(VirtIODevice
*vdev
, const uint8_t *config
)
146 VirtIONet
*n
= VIRTIO_NET(vdev
);
147 struct virtio_net_config netcfg
= {};
149 memcpy(&netcfg
, config
, n
->config_size
);
151 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
) &&
152 !virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
) &&
153 memcmp(netcfg
.mac
, n
->mac
, ETH_ALEN
)) {
154 memcpy(n
->mac
, netcfg
.mac
, ETH_ALEN
);
155 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->mac
);
159 static bool virtio_net_started(VirtIONet
*n
, uint8_t status
)
161 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
162 return (status
& VIRTIO_CONFIG_S_DRIVER_OK
) &&
163 (n
->status
& VIRTIO_NET_S_LINK_UP
) && vdev
->vm_running
;
166 static void virtio_net_announce_notify(VirtIONet
*net
)
168 VirtIODevice
*vdev
= VIRTIO_DEVICE(net
);
169 trace_virtio_net_announce_notify();
171 net
->status
|= VIRTIO_NET_S_ANNOUNCE
;
172 virtio_notify_config(vdev
);
175 static void virtio_net_announce_timer(void *opaque
)
177 VirtIONet
*n
= opaque
;
178 trace_virtio_net_announce_timer(n
->announce_timer
.round
);
180 n
->announce_timer
.round
--;
181 virtio_net_announce_notify(n
);
184 static void virtio_net_announce(NetClientState
*nc
)
186 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
187 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
190 * Make sure the virtio migration announcement timer isn't running
191 * If it is, let it trigger announcement so that we do not cause
194 if (n
->announce_timer
.round
) {
198 if (virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
) &&
199 virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
)) {
200 virtio_net_announce_notify(n
);
204 static void virtio_net_vhost_status(VirtIONet
*n
, uint8_t status
)
206 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
207 NetClientState
*nc
= qemu_get_queue(n
->nic
);
208 int queues
= n
->multiqueue
? n
->max_queues
: 1;
210 if (!get_vhost_net(nc
->peer
)) {
214 if ((virtio_net_started(n
, status
) && !nc
->peer
->link_down
) ==
215 !!n
->vhost_started
) {
218 if (!n
->vhost_started
) {
221 if (n
->needs_vnet_hdr_swap
) {
222 error_report("backend does not support %s vnet headers; "
223 "falling back on userspace virtio",
224 virtio_is_big_endian(vdev
) ? "BE" : "LE");
228 /* Any packets outstanding? Purge them to avoid touching rings
229 * when vhost is running.
231 for (i
= 0; i
< queues
; i
++) {
232 NetClientState
*qnc
= qemu_get_subqueue(n
->nic
, i
);
234 /* Purge both directions: TX and RX. */
235 qemu_net_queue_purge(qnc
->peer
->incoming_queue
, qnc
);
236 qemu_net_queue_purge(qnc
->incoming_queue
, qnc
->peer
);
239 if (virtio_has_feature(vdev
->guest_features
, VIRTIO_NET_F_MTU
)) {
240 r
= vhost_net_set_mtu(get_vhost_net(nc
->peer
), n
->net_conf
.mtu
);
242 error_report("%uBytes MTU not supported by the backend",
249 n
->vhost_started
= 1;
250 r
= vhost_net_start(vdev
, n
->nic
->ncs
, queues
);
252 error_report("unable to start vhost net: %d: "
253 "falling back on userspace virtio", -r
);
254 n
->vhost_started
= 0;
257 vhost_net_stop(vdev
, n
->nic
->ncs
, queues
);
258 n
->vhost_started
= 0;
262 static int virtio_net_set_vnet_endian_one(VirtIODevice
*vdev
,
263 NetClientState
*peer
,
266 if (virtio_is_big_endian(vdev
)) {
267 return qemu_set_vnet_be(peer
, enable
);
269 return qemu_set_vnet_le(peer
, enable
);
273 static bool virtio_net_set_vnet_endian(VirtIODevice
*vdev
, NetClientState
*ncs
,
274 int queues
, bool enable
)
278 for (i
= 0; i
< queues
; i
++) {
279 if (virtio_net_set_vnet_endian_one(vdev
, ncs
[i
].peer
, enable
) < 0 &&
282 virtio_net_set_vnet_endian_one(vdev
, ncs
[i
].peer
, false);
292 static void virtio_net_vnet_endian_status(VirtIONet
*n
, uint8_t status
)
294 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
295 int queues
= n
->multiqueue
? n
->max_queues
: 1;
297 if (virtio_net_started(n
, status
)) {
298 /* Before using the device, we tell the network backend about the
299 * endianness to use when parsing vnet headers. If the backend
300 * can't do it, we fallback onto fixing the headers in the core
303 n
->needs_vnet_hdr_swap
= virtio_net_set_vnet_endian(vdev
, n
->nic
->ncs
,
305 } else if (virtio_net_started(n
, vdev
->status
)) {
306 /* After using the device, we need to reset the network backend to
307 * the default (guest native endianness), otherwise the guest may
308 * lose network connectivity if it is rebooted into a different
311 virtio_net_set_vnet_endian(vdev
, n
->nic
->ncs
, queues
, false);
315 static void virtio_net_drop_tx_queue_data(VirtIODevice
*vdev
, VirtQueue
*vq
)
317 unsigned int dropped
= virtqueue_drop_all(vq
);
319 virtio_notify(vdev
, vq
);
323 static void virtio_net_set_status(struct VirtIODevice
*vdev
, uint8_t status
)
325 VirtIONet
*n
= VIRTIO_NET(vdev
);
328 uint8_t queue_status
;
330 virtio_net_vnet_endian_status(n
, status
);
331 virtio_net_vhost_status(n
, status
);
333 for (i
= 0; i
< n
->max_queues
; i
++) {
334 NetClientState
*ncs
= qemu_get_subqueue(n
->nic
, i
);
338 if ((!n
->multiqueue
&& i
!= 0) || i
>= n
->curr_queues
) {
341 queue_status
= status
;
344 virtio_net_started(n
, queue_status
) && !n
->vhost_started
;
347 qemu_flush_queued_packets(ncs
);
350 if (!q
->tx_waiting
) {
356 timer_mod(q
->tx_timer
,
357 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) + n
->tx_timeout
);
359 qemu_bh_schedule(q
->tx_bh
);
363 timer_del(q
->tx_timer
);
365 qemu_bh_cancel(q
->tx_bh
);
367 if ((n
->status
& VIRTIO_NET_S_LINK_UP
) == 0 &&
368 (queue_status
& VIRTIO_CONFIG_S_DRIVER_OK
) &&
370 /* if tx is waiting we are likely have some packets in tx queue
371 * and disabled notification */
373 virtio_queue_set_notification(q
->tx_vq
, 1);
374 virtio_net_drop_tx_queue_data(vdev
, q
->tx_vq
);
380 static void virtio_net_set_link_status(NetClientState
*nc
)
382 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
383 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
384 uint16_t old_status
= n
->status
;
387 n
->status
&= ~VIRTIO_NET_S_LINK_UP
;
389 n
->status
|= VIRTIO_NET_S_LINK_UP
;
391 if (n
->status
!= old_status
)
392 virtio_notify_config(vdev
);
394 virtio_net_set_status(vdev
, vdev
->status
);
397 static void rxfilter_notify(NetClientState
*nc
)
399 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
401 if (nc
->rxfilter_notify_enabled
) {
402 gchar
*path
= object_get_canonical_path(OBJECT(n
->qdev
));
403 qapi_event_send_nic_rx_filter_changed(!!n
->netclient_name
,
404 n
->netclient_name
, path
);
407 /* disable event notification to avoid events flooding */
408 nc
->rxfilter_notify_enabled
= 0;
412 static intList
*get_vlan_table(VirtIONet
*n
)
414 intList
*list
, *entry
;
418 for (i
= 0; i
< MAX_VLAN
>> 5; i
++) {
419 for (j
= 0; n
->vlans
[i
] && j
<= 0x1f; j
++) {
420 if (n
->vlans
[i
] & (1U << j
)) {
421 entry
= g_malloc0(sizeof(*entry
));
422 entry
->value
= (i
<< 5) + j
;
432 static RxFilterInfo
*virtio_net_query_rxfilter(NetClientState
*nc
)
434 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
435 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
437 strList
*str_list
, *entry
;
440 info
= g_malloc0(sizeof(*info
));
441 info
->name
= g_strdup(nc
->name
);
442 info
->promiscuous
= n
->promisc
;
445 info
->unicast
= RX_STATE_NONE
;
446 } else if (n
->alluni
) {
447 info
->unicast
= RX_STATE_ALL
;
449 info
->unicast
= RX_STATE_NORMAL
;
453 info
->multicast
= RX_STATE_NONE
;
454 } else if (n
->allmulti
) {
455 info
->multicast
= RX_STATE_ALL
;
457 info
->multicast
= RX_STATE_NORMAL
;
460 info
->broadcast_allowed
= n
->nobcast
;
461 info
->multicast_overflow
= n
->mac_table
.multi_overflow
;
462 info
->unicast_overflow
= n
->mac_table
.uni_overflow
;
464 info
->main_mac
= qemu_mac_strdup_printf(n
->mac
);
467 for (i
= 0; i
< n
->mac_table
.first_multi
; i
++) {
468 entry
= g_malloc0(sizeof(*entry
));
469 entry
->value
= qemu_mac_strdup_printf(n
->mac_table
.macs
+ i
* ETH_ALEN
);
470 entry
->next
= str_list
;
473 info
->unicast_table
= str_list
;
476 for (i
= n
->mac_table
.first_multi
; i
< n
->mac_table
.in_use
; i
++) {
477 entry
= g_malloc0(sizeof(*entry
));
478 entry
->value
= qemu_mac_strdup_printf(n
->mac_table
.macs
+ i
* ETH_ALEN
);
479 entry
->next
= str_list
;
482 info
->multicast_table
= str_list
;
483 info
->vlan_table
= get_vlan_table(n
);
485 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_VLAN
)) {
486 info
->vlan
= RX_STATE_ALL
;
487 } else if (!info
->vlan_table
) {
488 info
->vlan
= RX_STATE_NONE
;
490 info
->vlan
= RX_STATE_NORMAL
;
493 /* enable event notification after query */
494 nc
->rxfilter_notify_enabled
= 1;
499 static void virtio_net_reset(VirtIODevice
*vdev
)
501 VirtIONet
*n
= VIRTIO_NET(vdev
);
504 /* Reset back to compatibility mode */
511 /* multiqueue is disabled by default */
513 timer_del(n
->announce_timer
.tm
);
514 n
->announce_timer
.round
= 0;
515 n
->status
&= ~VIRTIO_NET_S_ANNOUNCE
;
517 /* Flush any MAC and VLAN filter table state */
518 n
->mac_table
.in_use
= 0;
519 n
->mac_table
.first_multi
= 0;
520 n
->mac_table
.multi_overflow
= 0;
521 n
->mac_table
.uni_overflow
= 0;
522 memset(n
->mac_table
.macs
, 0, MAC_TABLE_ENTRIES
* ETH_ALEN
);
523 memcpy(&n
->mac
[0], &n
->nic
->conf
->macaddr
, sizeof(n
->mac
));
524 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->mac
);
525 memset(n
->vlans
, 0, MAX_VLAN
>> 3);
527 /* Flush any async TX */
528 for (i
= 0; i
< n
->max_queues
; i
++) {
529 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, i
);
532 qemu_flush_or_purge_queued_packets(nc
->peer
, true);
533 assert(!virtio_net_get_subqueue(nc
)->async_tx
.elem
);
538 static void peer_test_vnet_hdr(VirtIONet
*n
)
540 NetClientState
*nc
= qemu_get_queue(n
->nic
);
545 n
->has_vnet_hdr
= qemu_has_vnet_hdr(nc
->peer
);
548 static int peer_has_vnet_hdr(VirtIONet
*n
)
550 return n
->has_vnet_hdr
;
553 static int peer_has_ufo(VirtIONet
*n
)
555 if (!peer_has_vnet_hdr(n
))
558 n
->has_ufo
= qemu_has_ufo(qemu_get_queue(n
->nic
)->peer
);
563 static void virtio_net_set_mrg_rx_bufs(VirtIONet
*n
, int mergeable_rx_bufs
,
569 n
->mergeable_rx_bufs
= mergeable_rx_bufs
;
572 n
->guest_hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
574 n
->guest_hdr_len
= n
->mergeable_rx_bufs
?
575 sizeof(struct virtio_net_hdr_mrg_rxbuf
) :
576 sizeof(struct virtio_net_hdr
);
579 for (i
= 0; i
< n
->max_queues
; i
++) {
580 nc
= qemu_get_subqueue(n
->nic
, i
);
582 if (peer_has_vnet_hdr(n
) &&
583 qemu_has_vnet_hdr_len(nc
->peer
, n
->guest_hdr_len
)) {
584 qemu_set_vnet_hdr_len(nc
->peer
, n
->guest_hdr_len
);
585 n
->host_hdr_len
= n
->guest_hdr_len
;
590 static int virtio_net_max_tx_queue_size(VirtIONet
*n
)
592 NetClientState
*peer
= n
->nic_conf
.peers
.ncs
[0];
595 * Backends other than vhost-user don't support max queue size.
598 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
;
601 if (peer
->info
->type
!= NET_CLIENT_DRIVER_VHOST_USER
) {
602 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
;
605 return VIRTQUEUE_MAX_SIZE
;
608 static int peer_attach(VirtIONet
*n
, int index
)
610 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, index
);
616 if (nc
->peer
->info
->type
== NET_CLIENT_DRIVER_VHOST_USER
) {
617 vhost_set_vring_enable(nc
->peer
, 1);
620 if (nc
->peer
->info
->type
!= NET_CLIENT_DRIVER_TAP
) {
624 if (n
->max_queues
== 1) {
628 return tap_enable(nc
->peer
);
631 static int peer_detach(VirtIONet
*n
, int index
)
633 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, index
);
639 if (nc
->peer
->info
->type
== NET_CLIENT_DRIVER_VHOST_USER
) {
640 vhost_set_vring_enable(nc
->peer
, 0);
643 if (nc
->peer
->info
->type
!= NET_CLIENT_DRIVER_TAP
) {
647 return tap_disable(nc
->peer
);
650 static void virtio_net_set_queues(VirtIONet
*n
)
655 if (n
->nic
->peer_deleted
) {
659 for (i
= 0; i
< n
->max_queues
; i
++) {
660 if (i
< n
->curr_queues
) {
661 r
= peer_attach(n
, i
);
664 r
= peer_detach(n
, i
);
670 static void virtio_net_set_multiqueue(VirtIONet
*n
, int multiqueue
);
672 static uint64_t virtio_net_get_features(VirtIODevice
*vdev
, uint64_t features
,
675 VirtIONet
*n
= VIRTIO_NET(vdev
);
676 NetClientState
*nc
= qemu_get_queue(n
->nic
);
678 /* Firstly sync all virtio-net possible supported features */
679 features
|= n
->host_features
;
681 virtio_add_feature(&features
, VIRTIO_NET_F_MAC
);
683 if (!peer_has_vnet_hdr(n
)) {
684 virtio_clear_feature(&features
, VIRTIO_NET_F_CSUM
);
685 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_TSO4
);
686 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_TSO6
);
687 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_ECN
);
689 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_CSUM
);
690 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_TSO4
);
691 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_TSO6
);
692 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_ECN
);
695 if (!peer_has_vnet_hdr(n
) || !peer_has_ufo(n
)) {
696 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_UFO
);
697 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_UFO
);
700 if (!get_vhost_net(nc
->peer
)) {
704 features
= vhost_net_get_features(get_vhost_net(nc
->peer
), features
);
705 vdev
->backend_features
= features
;
707 if (n
->mtu_bypass_backend
&&
708 (n
->host_features
& 1ULL << VIRTIO_NET_F_MTU
)) {
709 features
|= (1ULL << VIRTIO_NET_F_MTU
);
715 static uint64_t virtio_net_bad_features(VirtIODevice
*vdev
)
717 uint64_t features
= 0;
719 /* Linux kernel 2.6.25. It understood MAC (as everyone must),
721 virtio_add_feature(&features
, VIRTIO_NET_F_MAC
);
722 virtio_add_feature(&features
, VIRTIO_NET_F_CSUM
);
723 virtio_add_feature(&features
, VIRTIO_NET_F_HOST_TSO4
);
724 virtio_add_feature(&features
, VIRTIO_NET_F_HOST_TSO6
);
725 virtio_add_feature(&features
, VIRTIO_NET_F_HOST_ECN
);
730 static void virtio_net_apply_guest_offloads(VirtIONet
*n
)
732 qemu_set_offload(qemu_get_queue(n
->nic
)->peer
,
733 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_CSUM
)),
734 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_TSO4
)),
735 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_TSO6
)),
736 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_ECN
)),
737 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_UFO
)));
740 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features
)
742 static const uint64_t guest_offloads_mask
=
743 (1ULL << VIRTIO_NET_F_GUEST_CSUM
) |
744 (1ULL << VIRTIO_NET_F_GUEST_TSO4
) |
745 (1ULL << VIRTIO_NET_F_GUEST_TSO6
) |
746 (1ULL << VIRTIO_NET_F_GUEST_ECN
) |
747 (1ULL << VIRTIO_NET_F_GUEST_UFO
);
749 return guest_offloads_mask
& features
;
752 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet
*n
)
754 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
755 return virtio_net_guest_offloads_by_features(vdev
->guest_features
);
758 static void failover_add_primary(VirtIONet
*n
, Error
**errp
)
762 n
->primary_device_opts
= qemu_opts_find(qemu_find_opts("device"),
763 n
->primary_device_id
);
764 if (n
->primary_device_opts
) {
765 n
->primary_dev
= qdev_device_add(n
->primary_device_opts
, &err
);
767 qemu_opts_del(n
->primary_device_opts
);
769 if (n
->primary_dev
) {
770 n
->primary_bus
= n
->primary_dev
->parent_bus
;
772 qdev_unplug(n
->primary_dev
, &err
);
773 qdev_set_id(n
->primary_dev
, "");
778 error_setg(errp
, "Primary device not found");
779 error_append_hint(errp
, "Virtio-net failover will not work. Make "
780 "sure primary device has parameter"
781 " failover_pair_id=<virtio-net-id>\n");
784 error_propagate(errp
, err
);
788 static int is_my_primary(void *opaque
, QemuOpts
*opts
, Error
**errp
)
790 VirtIONet
*n
= opaque
;
793 const char *standby_id
= qemu_opt_get(opts
, "failover_pair_id");
795 if (standby_id
!= NULL
&& (g_strcmp0(standby_id
, n
->netclient_name
) == 0)) {
796 n
->primary_device_id
= g_strdup(opts
->id
);
803 static DeviceState
*virtio_net_find_primary(VirtIONet
*n
, Error
**errp
)
805 DeviceState
*dev
= NULL
;
808 if (qemu_opts_foreach(qemu_find_opts("device"),
809 is_my_primary
, n
, &err
)) {
811 error_propagate(errp
, err
);
814 if (n
->primary_device_id
) {
815 dev
= qdev_find_recursive(sysbus_get_default(),
816 n
->primary_device_id
);
818 error_setg(errp
, "Primary device id not found");
827 static DeviceState
*virtio_connect_failover_devices(VirtIONet
*n
,
831 DeviceState
*prim_dev
= NULL
;
834 prim_dev
= virtio_net_find_primary(n
, &err
);
836 n
->primary_device_id
= g_strdup(prim_dev
->id
);
837 n
->primary_device_opts
= prim_dev
->opts
;
840 error_propagate(errp
, err
);
847 static void virtio_net_set_features(VirtIODevice
*vdev
, uint64_t features
)
849 VirtIONet
*n
= VIRTIO_NET(vdev
);
853 if (n
->mtu_bypass_backend
&&
854 !virtio_has_feature(vdev
->backend_features
, VIRTIO_NET_F_MTU
)) {
855 features
&= ~(1ULL << VIRTIO_NET_F_MTU
);
858 virtio_net_set_multiqueue(n
,
859 virtio_has_feature(features
, VIRTIO_NET_F_MQ
));
861 virtio_net_set_mrg_rx_bufs(n
,
862 virtio_has_feature(features
,
863 VIRTIO_NET_F_MRG_RXBUF
),
864 virtio_has_feature(features
,
865 VIRTIO_F_VERSION_1
));
867 n
->rsc4_enabled
= virtio_has_feature(features
, VIRTIO_NET_F_RSC_EXT
) &&
868 virtio_has_feature(features
, VIRTIO_NET_F_GUEST_TSO4
);
869 n
->rsc6_enabled
= virtio_has_feature(features
, VIRTIO_NET_F_RSC_EXT
) &&
870 virtio_has_feature(features
, VIRTIO_NET_F_GUEST_TSO6
);
872 if (n
->has_vnet_hdr
) {
873 n
->curr_guest_offloads
=
874 virtio_net_guest_offloads_by_features(features
);
875 virtio_net_apply_guest_offloads(n
);
878 for (i
= 0; i
< n
->max_queues
; i
++) {
879 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, i
);
881 if (!get_vhost_net(nc
->peer
)) {
884 vhost_net_ack_features(get_vhost_net(nc
->peer
), features
);
887 if (virtio_has_feature(features
, VIRTIO_NET_F_CTRL_VLAN
)) {
888 memset(n
->vlans
, 0, MAX_VLAN
>> 3);
890 memset(n
->vlans
, 0xff, MAX_VLAN
>> 3);
893 if (virtio_has_feature(features
, VIRTIO_NET_F_STANDBY
)) {
894 qapi_event_send_failover_negotiated(n
->netclient_name
);
895 atomic_set(&n
->primary_should_be_hidden
, false);
896 failover_add_primary(n
, &err
);
898 n
->primary_dev
= virtio_connect_failover_devices(n
, n
->qdev
, &err
);
902 failover_add_primary(n
, &err
);
912 warn_report_err(err
);
916 static int virtio_net_handle_rx_mode(VirtIONet
*n
, uint8_t cmd
,
917 struct iovec
*iov
, unsigned int iov_cnt
)
921 NetClientState
*nc
= qemu_get_queue(n
->nic
);
923 s
= iov_to_buf(iov
, iov_cnt
, 0, &on
, sizeof(on
));
924 if (s
!= sizeof(on
)) {
925 return VIRTIO_NET_ERR
;
928 if (cmd
== VIRTIO_NET_CTRL_RX_PROMISC
) {
930 } else if (cmd
== VIRTIO_NET_CTRL_RX_ALLMULTI
) {
932 } else if (cmd
== VIRTIO_NET_CTRL_RX_ALLUNI
) {
934 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOMULTI
) {
936 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOUNI
) {
938 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOBCAST
) {
941 return VIRTIO_NET_ERR
;
946 return VIRTIO_NET_OK
;
949 static int virtio_net_handle_offloads(VirtIONet
*n
, uint8_t cmd
,
950 struct iovec
*iov
, unsigned int iov_cnt
)
952 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
956 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
)) {
957 return VIRTIO_NET_ERR
;
960 s
= iov_to_buf(iov
, iov_cnt
, 0, &offloads
, sizeof(offloads
));
961 if (s
!= sizeof(offloads
)) {
962 return VIRTIO_NET_ERR
;
965 if (cmd
== VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET
) {
966 uint64_t supported_offloads
;
968 offloads
= virtio_ldq_p(vdev
, &offloads
);
970 if (!n
->has_vnet_hdr
) {
971 return VIRTIO_NET_ERR
;
974 n
->rsc4_enabled
= virtio_has_feature(offloads
, VIRTIO_NET_F_RSC_EXT
) &&
975 virtio_has_feature(offloads
, VIRTIO_NET_F_GUEST_TSO4
);
976 n
->rsc6_enabled
= virtio_has_feature(offloads
, VIRTIO_NET_F_RSC_EXT
) &&
977 virtio_has_feature(offloads
, VIRTIO_NET_F_GUEST_TSO6
);
978 virtio_clear_feature(&offloads
, VIRTIO_NET_F_RSC_EXT
);
980 supported_offloads
= virtio_net_supported_guest_offloads(n
);
981 if (offloads
& ~supported_offloads
) {
982 return VIRTIO_NET_ERR
;
985 n
->curr_guest_offloads
= offloads
;
986 virtio_net_apply_guest_offloads(n
);
988 return VIRTIO_NET_OK
;
990 return VIRTIO_NET_ERR
;
994 static int virtio_net_handle_mac(VirtIONet
*n
, uint8_t cmd
,
995 struct iovec
*iov
, unsigned int iov_cnt
)
997 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
998 struct virtio_net_ctrl_mac mac_data
;
1000 NetClientState
*nc
= qemu_get_queue(n
->nic
);
1002 if (cmd
== VIRTIO_NET_CTRL_MAC_ADDR_SET
) {
1003 if (iov_size(iov
, iov_cnt
) != sizeof(n
->mac
)) {
1004 return VIRTIO_NET_ERR
;
1006 s
= iov_to_buf(iov
, iov_cnt
, 0, &n
->mac
, sizeof(n
->mac
));
1007 assert(s
== sizeof(n
->mac
));
1008 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->mac
);
1009 rxfilter_notify(nc
);
1011 return VIRTIO_NET_OK
;
1014 if (cmd
!= VIRTIO_NET_CTRL_MAC_TABLE_SET
) {
1015 return VIRTIO_NET_ERR
;
1019 int first_multi
= 0;
1020 uint8_t uni_overflow
= 0;
1021 uint8_t multi_overflow
= 0;
1022 uint8_t *macs
= g_malloc0(MAC_TABLE_ENTRIES
* ETH_ALEN
);
1024 s
= iov_to_buf(iov
, iov_cnt
, 0, &mac_data
.entries
,
1025 sizeof(mac_data
.entries
));
1026 mac_data
.entries
= virtio_ldl_p(vdev
, &mac_data
.entries
);
1027 if (s
!= sizeof(mac_data
.entries
)) {
1030 iov_discard_front(&iov
, &iov_cnt
, s
);
1032 if (mac_data
.entries
* ETH_ALEN
> iov_size(iov
, iov_cnt
)) {
1036 if (mac_data
.entries
<= MAC_TABLE_ENTRIES
) {
1037 s
= iov_to_buf(iov
, iov_cnt
, 0, macs
,
1038 mac_data
.entries
* ETH_ALEN
);
1039 if (s
!= mac_data
.entries
* ETH_ALEN
) {
1042 in_use
+= mac_data
.entries
;
1047 iov_discard_front(&iov
, &iov_cnt
, mac_data
.entries
* ETH_ALEN
);
1049 first_multi
= in_use
;
1051 s
= iov_to_buf(iov
, iov_cnt
, 0, &mac_data
.entries
,
1052 sizeof(mac_data
.entries
));
1053 mac_data
.entries
= virtio_ldl_p(vdev
, &mac_data
.entries
);
1054 if (s
!= sizeof(mac_data
.entries
)) {
1058 iov_discard_front(&iov
, &iov_cnt
, s
);
1060 if (mac_data
.entries
* ETH_ALEN
!= iov_size(iov
, iov_cnt
)) {
1064 if (mac_data
.entries
<= MAC_TABLE_ENTRIES
- in_use
) {
1065 s
= iov_to_buf(iov
, iov_cnt
, 0, &macs
[in_use
* ETH_ALEN
],
1066 mac_data
.entries
* ETH_ALEN
);
1067 if (s
!= mac_data
.entries
* ETH_ALEN
) {
1070 in_use
+= mac_data
.entries
;
1075 n
->mac_table
.in_use
= in_use
;
1076 n
->mac_table
.first_multi
= first_multi
;
1077 n
->mac_table
.uni_overflow
= uni_overflow
;
1078 n
->mac_table
.multi_overflow
= multi_overflow
;
1079 memcpy(n
->mac_table
.macs
, macs
, MAC_TABLE_ENTRIES
* ETH_ALEN
);
1081 rxfilter_notify(nc
);
1083 return VIRTIO_NET_OK
;
1087 return VIRTIO_NET_ERR
;
1090 static int virtio_net_handle_vlan_table(VirtIONet
*n
, uint8_t cmd
,
1091 struct iovec
*iov
, unsigned int iov_cnt
)
1093 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1096 NetClientState
*nc
= qemu_get_queue(n
->nic
);
1098 s
= iov_to_buf(iov
, iov_cnt
, 0, &vid
, sizeof(vid
));
1099 vid
= virtio_lduw_p(vdev
, &vid
);
1100 if (s
!= sizeof(vid
)) {
1101 return VIRTIO_NET_ERR
;
1104 if (vid
>= MAX_VLAN
)
1105 return VIRTIO_NET_ERR
;
1107 if (cmd
== VIRTIO_NET_CTRL_VLAN_ADD
)
1108 n
->vlans
[vid
>> 5] |= (1U << (vid
& 0x1f));
1109 else if (cmd
== VIRTIO_NET_CTRL_VLAN_DEL
)
1110 n
->vlans
[vid
>> 5] &= ~(1U << (vid
& 0x1f));
1112 return VIRTIO_NET_ERR
;
1114 rxfilter_notify(nc
);
1116 return VIRTIO_NET_OK
;
1119 static int virtio_net_handle_announce(VirtIONet
*n
, uint8_t cmd
,
1120 struct iovec
*iov
, unsigned int iov_cnt
)
1122 trace_virtio_net_handle_announce(n
->announce_timer
.round
);
1123 if (cmd
== VIRTIO_NET_CTRL_ANNOUNCE_ACK
&&
1124 n
->status
& VIRTIO_NET_S_ANNOUNCE
) {
1125 n
->status
&= ~VIRTIO_NET_S_ANNOUNCE
;
1126 if (n
->announce_timer
.round
) {
1127 qemu_announce_timer_step(&n
->announce_timer
);
1129 return VIRTIO_NET_OK
;
1131 return VIRTIO_NET_ERR
;
1135 static int virtio_net_handle_mq(VirtIONet
*n
, uint8_t cmd
,
1136 struct iovec
*iov
, unsigned int iov_cnt
)
1138 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1139 struct virtio_net_ctrl_mq mq
;
1143 s
= iov_to_buf(iov
, iov_cnt
, 0, &mq
, sizeof(mq
));
1144 if (s
!= sizeof(mq
)) {
1145 return VIRTIO_NET_ERR
;
1148 if (cmd
!= VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
) {
1149 return VIRTIO_NET_ERR
;
1152 queues
= virtio_lduw_p(vdev
, &mq
.virtqueue_pairs
);
1154 if (queues
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
1155 queues
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
1156 queues
> n
->max_queues
||
1158 return VIRTIO_NET_ERR
;
1161 n
->curr_queues
= queues
;
1162 /* stop the backend before changing the number of queues to avoid handling a
1164 virtio_net_set_status(vdev
, vdev
->status
);
1165 virtio_net_set_queues(n
);
1167 return VIRTIO_NET_OK
;
1170 static void virtio_net_handle_ctrl(VirtIODevice
*vdev
, VirtQueue
*vq
)
1172 VirtIONet
*n
= VIRTIO_NET(vdev
);
1173 struct virtio_net_ctrl_hdr ctrl
;
1174 virtio_net_ctrl_ack status
= VIRTIO_NET_ERR
;
1175 VirtQueueElement
*elem
;
1177 struct iovec
*iov
, *iov2
;
1178 unsigned int iov_cnt
;
1181 elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
));
1185 if (iov_size(elem
->in_sg
, elem
->in_num
) < sizeof(status
) ||
1186 iov_size(elem
->out_sg
, elem
->out_num
) < sizeof(ctrl
)) {
1187 virtio_error(vdev
, "virtio-net ctrl missing headers");
1188 virtqueue_detach_element(vq
, elem
, 0);
1193 iov_cnt
= elem
->out_num
;
1194 iov2
= iov
= g_memdup(elem
->out_sg
, sizeof(struct iovec
) * elem
->out_num
);
1195 s
= iov_to_buf(iov
, iov_cnt
, 0, &ctrl
, sizeof(ctrl
));
1196 iov_discard_front(&iov
, &iov_cnt
, sizeof(ctrl
));
1197 if (s
!= sizeof(ctrl
)) {
1198 status
= VIRTIO_NET_ERR
;
1199 } else if (ctrl
.class == VIRTIO_NET_CTRL_RX
) {
1200 status
= virtio_net_handle_rx_mode(n
, ctrl
.cmd
, iov
, iov_cnt
);
1201 } else if (ctrl
.class == VIRTIO_NET_CTRL_MAC
) {
1202 status
= virtio_net_handle_mac(n
, ctrl
.cmd
, iov
, iov_cnt
);
1203 } else if (ctrl
.class == VIRTIO_NET_CTRL_VLAN
) {
1204 status
= virtio_net_handle_vlan_table(n
, ctrl
.cmd
, iov
, iov_cnt
);
1205 } else if (ctrl
.class == VIRTIO_NET_CTRL_ANNOUNCE
) {
1206 status
= virtio_net_handle_announce(n
, ctrl
.cmd
, iov
, iov_cnt
);
1207 } else if (ctrl
.class == VIRTIO_NET_CTRL_MQ
) {
1208 status
= virtio_net_handle_mq(n
, ctrl
.cmd
, iov
, iov_cnt
);
1209 } else if (ctrl
.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS
) {
1210 status
= virtio_net_handle_offloads(n
, ctrl
.cmd
, iov
, iov_cnt
);
1213 s
= iov_from_buf(elem
->in_sg
, elem
->in_num
, 0, &status
, sizeof(status
));
1214 assert(s
== sizeof(status
));
1216 virtqueue_push(vq
, elem
, sizeof(status
));
1217 virtio_notify(vdev
, vq
);
1225 static void virtio_net_handle_rx(VirtIODevice
*vdev
, VirtQueue
*vq
)
1227 VirtIONet
*n
= VIRTIO_NET(vdev
);
1228 int queue_index
= vq2q(virtio_get_queue_index(vq
));
1230 qemu_flush_queued_packets(qemu_get_subqueue(n
->nic
, queue_index
));
1233 static int virtio_net_can_receive(NetClientState
*nc
)
1235 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
1236 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1237 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
1239 if (!vdev
->vm_running
) {
1243 if (nc
->queue_index
>= n
->curr_queues
) {
1247 if (!virtio_queue_ready(q
->rx_vq
) ||
1248 !(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
1255 static int virtio_net_has_buffers(VirtIONetQueue
*q
, int bufsize
)
1257 VirtIONet
*n
= q
->n
;
1258 if (virtio_queue_empty(q
->rx_vq
) ||
1259 (n
->mergeable_rx_bufs
&&
1260 !virtqueue_avail_bytes(q
->rx_vq
, bufsize
, 0))) {
1261 virtio_queue_set_notification(q
->rx_vq
, 1);
1263 /* To avoid a race condition where the guest has made some buffers
1264 * available after the above check but before notification was
1265 * enabled, check for available buffers again.
1267 if (virtio_queue_empty(q
->rx_vq
) ||
1268 (n
->mergeable_rx_bufs
&&
1269 !virtqueue_avail_bytes(q
->rx_vq
, bufsize
, 0))) {
1274 virtio_queue_set_notification(q
->rx_vq
, 0);
1278 static void virtio_net_hdr_swap(VirtIODevice
*vdev
, struct virtio_net_hdr
*hdr
)
1280 virtio_tswap16s(vdev
, &hdr
->hdr_len
);
1281 virtio_tswap16s(vdev
, &hdr
->gso_size
);
1282 virtio_tswap16s(vdev
, &hdr
->csum_start
);
1283 virtio_tswap16s(vdev
, &hdr
->csum_offset
);
1286 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1287 * it never finds out that the packets don't have valid checksums. This
1288 * causes dhclient to get upset. Fedora's carried a patch for ages to
1289 * fix this with Xen but it hasn't appeared in an upstream release of
1292 * To avoid breaking existing guests, we catch udp packets and add
1293 * checksums. This is terrible but it's better than hacking the guest
1296 * N.B. if we introduce a zero-copy API, this operation is no longer free so
1297 * we should provide a mechanism to disable it to avoid polluting the host
1300 static void work_around_broken_dhclient(struct virtio_net_hdr
*hdr
,
1301 uint8_t *buf
, size_t size
)
1303 if ((hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) && /* missing csum */
1304 (size
> 27 && size
< 1500) && /* normal sized MTU */
1305 (buf
[12] == 0x08 && buf
[13] == 0x00) && /* ethertype == IPv4 */
1306 (buf
[23] == 17) && /* ip.protocol == UDP */
1307 (buf
[34] == 0 && buf
[35] == 67)) { /* udp.srcport == bootps */
1308 net_checksum_calculate(buf
, size
);
1309 hdr
->flags
&= ~VIRTIO_NET_HDR_F_NEEDS_CSUM
;
1313 static void receive_header(VirtIONet
*n
, const struct iovec
*iov
, int iov_cnt
,
1314 const void *buf
, size_t size
)
1316 if (n
->has_vnet_hdr
) {
1317 /* FIXME this cast is evil */
1318 void *wbuf
= (void *)buf
;
1319 work_around_broken_dhclient(wbuf
, wbuf
+ n
->host_hdr_len
,
1320 size
- n
->host_hdr_len
);
1322 if (n
->needs_vnet_hdr_swap
) {
1323 virtio_net_hdr_swap(VIRTIO_DEVICE(n
), wbuf
);
1325 iov_from_buf(iov
, iov_cnt
, 0, buf
, sizeof(struct virtio_net_hdr
));
1327 struct virtio_net_hdr hdr
= {
1329 .gso_type
= VIRTIO_NET_HDR_GSO_NONE
1331 iov_from_buf(iov
, iov_cnt
, 0, &hdr
, sizeof hdr
);
1335 static int receive_filter(VirtIONet
*n
, const uint8_t *buf
, int size
)
1337 static const uint8_t bcast
[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1338 static const uint8_t vlan
[] = {0x81, 0x00};
1339 uint8_t *ptr
= (uint8_t *)buf
;
1345 ptr
+= n
->host_hdr_len
;
1347 if (!memcmp(&ptr
[12], vlan
, sizeof(vlan
))) {
1348 int vid
= lduw_be_p(ptr
+ 14) & 0xfff;
1349 if (!(n
->vlans
[vid
>> 5] & (1U << (vid
& 0x1f))))
1353 if (ptr
[0] & 1) { // multicast
1354 if (!memcmp(ptr
, bcast
, sizeof(bcast
))) {
1356 } else if (n
->nomulti
) {
1358 } else if (n
->allmulti
|| n
->mac_table
.multi_overflow
) {
1362 for (i
= n
->mac_table
.first_multi
; i
< n
->mac_table
.in_use
; i
++) {
1363 if (!memcmp(ptr
, &n
->mac_table
.macs
[i
* ETH_ALEN
], ETH_ALEN
)) {
1370 } else if (n
->alluni
|| n
->mac_table
.uni_overflow
) {
1372 } else if (!memcmp(ptr
, n
->mac
, ETH_ALEN
)) {
1376 for (i
= 0; i
< n
->mac_table
.first_multi
; i
++) {
1377 if (!memcmp(ptr
, &n
->mac_table
.macs
[i
* ETH_ALEN
], ETH_ALEN
)) {
1386 static ssize_t
virtio_net_receive_rcu(NetClientState
*nc
, const uint8_t *buf
,
1389 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
1390 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
1391 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1392 struct iovec mhdr_sg
[VIRTQUEUE_MAX_SIZE
];
1393 struct virtio_net_hdr_mrg_rxbuf mhdr
;
1394 unsigned mhdr_cnt
= 0;
1395 size_t offset
, i
, guest_offset
;
1397 if (!virtio_net_can_receive(nc
)) {
1401 /* hdr_len refers to the header we supply to the guest */
1402 if (!virtio_net_has_buffers(q
, size
+ n
->guest_hdr_len
- n
->host_hdr_len
)) {
1406 if (!receive_filter(n
, buf
, size
))
1411 while (offset
< size
) {
1412 VirtQueueElement
*elem
;
1414 const struct iovec
*sg
;
1418 elem
= virtqueue_pop(q
->rx_vq
, sizeof(VirtQueueElement
));
1421 virtio_error(vdev
, "virtio-net unexpected empty queue: "
1422 "i %zd mergeable %d offset %zd, size %zd, "
1423 "guest hdr len %zd, host hdr len %zd "
1424 "guest features 0x%" PRIx64
,
1425 i
, n
->mergeable_rx_bufs
, offset
, size
,
1426 n
->guest_hdr_len
, n
->host_hdr_len
,
1427 vdev
->guest_features
);
1432 if (elem
->in_num
< 1) {
1434 "virtio-net receive queue contains no in buffers");
1435 virtqueue_detach_element(q
->rx_vq
, elem
, 0);
1442 assert(offset
== 0);
1443 if (n
->mergeable_rx_bufs
) {
1444 mhdr_cnt
= iov_copy(mhdr_sg
, ARRAY_SIZE(mhdr_sg
),
1446 offsetof(typeof(mhdr
), num_buffers
),
1447 sizeof(mhdr
.num_buffers
));
1450 receive_header(n
, sg
, elem
->in_num
, buf
, size
);
1451 offset
= n
->host_hdr_len
;
1452 total
+= n
->guest_hdr_len
;
1453 guest_offset
= n
->guest_hdr_len
;
1458 /* copy in packet. ugh */
1459 len
= iov_from_buf(sg
, elem
->in_num
, guest_offset
,
1460 buf
+ offset
, size
- offset
);
1463 /* If buffers can't be merged, at this point we
1464 * must have consumed the complete packet.
1465 * Otherwise, drop it. */
1466 if (!n
->mergeable_rx_bufs
&& offset
< size
) {
1467 virtqueue_unpop(q
->rx_vq
, elem
, total
);
1472 /* signal other side */
1473 virtqueue_fill(q
->rx_vq
, elem
, total
, i
++);
1478 virtio_stw_p(vdev
, &mhdr
.num_buffers
, i
);
1479 iov_from_buf(mhdr_sg
, mhdr_cnt
,
1481 &mhdr
.num_buffers
, sizeof mhdr
.num_buffers
);
1484 virtqueue_flush(q
->rx_vq
, i
);
1485 virtio_notify(vdev
, q
->rx_vq
);
1490 static ssize_t
virtio_net_do_receive(NetClientState
*nc
, const uint8_t *buf
,
1493 RCU_READ_LOCK_GUARD();
1495 return virtio_net_receive_rcu(nc
, buf
, size
);
1498 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain
*chain
,
1500 VirtioNetRscUnit
*unit
)
1503 struct ip_header
*ip
;
1505 ip
= (struct ip_header
*)(buf
+ chain
->n
->guest_hdr_len
1506 + sizeof(struct eth_header
));
1507 unit
->ip
= (void *)ip
;
1508 ip_hdrlen
= (ip
->ip_ver_len
& 0xF) << 2;
1509 unit
->ip_plen
= &ip
->ip_len
;
1510 unit
->tcp
= (struct tcp_header
*)(((uint8_t *)unit
->ip
) + ip_hdrlen
);
1511 unit
->tcp_hdrlen
= (htons(unit
->tcp
->th_offset_flags
) & 0xF000) >> 10;
1512 unit
->payload
= htons(*unit
->ip_plen
) - ip_hdrlen
- unit
->tcp_hdrlen
;
1515 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain
*chain
,
1517 VirtioNetRscUnit
*unit
)
1519 struct ip6_header
*ip6
;
1521 ip6
= (struct ip6_header
*)(buf
+ chain
->n
->guest_hdr_len
1522 + sizeof(struct eth_header
));
1524 unit
->ip_plen
= &(ip6
->ip6_ctlun
.ip6_un1
.ip6_un1_plen
);
1525 unit
->tcp
= (struct tcp_header
*)(((uint8_t *)unit
->ip
)\
1526 + sizeof(struct ip6_header
));
1527 unit
->tcp_hdrlen
= (htons(unit
->tcp
->th_offset_flags
) & 0xF000) >> 10;
1529 /* There is a difference between payload lenght in ipv4 and v6,
1530 ip header is excluded in ipv6 */
1531 unit
->payload
= htons(*unit
->ip_plen
) - unit
->tcp_hdrlen
;
1534 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain
*chain
,
1535 VirtioNetRscSeg
*seg
)
1538 struct virtio_net_hdr
*h
;
1540 h
= (struct virtio_net_hdr
*)seg
->buf
;
1542 h
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
1544 if (seg
->is_coalesced
) {
1545 *virtio_net_rsc_ext_num_packets(h
) = seg
->packets
;
1546 *virtio_net_rsc_ext_num_dupacks(h
) = seg
->dup_ack
;
1547 h
->flags
= VIRTIO_NET_HDR_F_RSC_INFO
;
1548 if (chain
->proto
== ETH_P_IP
) {
1549 h
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
1551 h
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
1555 ret
= virtio_net_do_receive(seg
->nc
, seg
->buf
, seg
->size
);
1556 QTAILQ_REMOVE(&chain
->buffers
, seg
, next
);
1563 static void virtio_net_rsc_purge(void *opq
)
1565 VirtioNetRscSeg
*seg
, *rn
;
1566 VirtioNetRscChain
*chain
= (VirtioNetRscChain
*)opq
;
1568 QTAILQ_FOREACH_SAFE(seg
, &chain
->buffers
, next
, rn
) {
1569 if (virtio_net_rsc_drain_seg(chain
, seg
) == 0) {
1570 chain
->stat
.purge_failed
++;
1575 chain
->stat
.timer
++;
1576 if (!QTAILQ_EMPTY(&chain
->buffers
)) {
1577 timer_mod(chain
->drain_timer
,
1578 qemu_clock_get_ns(QEMU_CLOCK_HOST
) + chain
->n
->rsc_timeout
);
1582 static void virtio_net_rsc_cleanup(VirtIONet
*n
)
1584 VirtioNetRscChain
*chain
, *rn_chain
;
1585 VirtioNetRscSeg
*seg
, *rn_seg
;
1587 QTAILQ_FOREACH_SAFE(chain
, &n
->rsc_chains
, next
, rn_chain
) {
1588 QTAILQ_FOREACH_SAFE(seg
, &chain
->buffers
, next
, rn_seg
) {
1589 QTAILQ_REMOVE(&chain
->buffers
, seg
, next
);
1594 timer_del(chain
->drain_timer
);
1595 timer_free(chain
->drain_timer
);
1596 QTAILQ_REMOVE(&n
->rsc_chains
, chain
, next
);
1601 static void virtio_net_rsc_cache_buf(VirtioNetRscChain
*chain
,
1603 const uint8_t *buf
, size_t size
)
1606 VirtioNetRscSeg
*seg
;
1608 hdr_len
= chain
->n
->guest_hdr_len
;
1609 seg
= g_malloc(sizeof(VirtioNetRscSeg
));
1610 seg
->buf
= g_malloc(hdr_len
+ sizeof(struct eth_header
)
1611 + sizeof(struct ip6_header
) + VIRTIO_NET_MAX_TCP_PAYLOAD
);
1612 memcpy(seg
->buf
, buf
, size
);
1616 seg
->is_coalesced
= 0;
1619 QTAILQ_INSERT_TAIL(&chain
->buffers
, seg
, next
);
1620 chain
->stat
.cache
++;
1622 switch (chain
->proto
) {
1624 virtio_net_rsc_extract_unit4(chain
, seg
->buf
, &seg
->unit
);
1627 virtio_net_rsc_extract_unit6(chain
, seg
->buf
, &seg
->unit
);
1630 g_assert_not_reached();
1634 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain
*chain
,
1635 VirtioNetRscSeg
*seg
,
1637 struct tcp_header
*n_tcp
,
1638 struct tcp_header
*o_tcp
)
1640 uint32_t nack
, oack
;
1641 uint16_t nwin
, owin
;
1643 nack
= htonl(n_tcp
->th_ack
);
1644 nwin
= htons(n_tcp
->th_win
);
1645 oack
= htonl(o_tcp
->th_ack
);
1646 owin
= htons(o_tcp
->th_win
);
1648 if ((nack
- oack
) >= VIRTIO_NET_MAX_TCP_PAYLOAD
) {
1649 chain
->stat
.ack_out_of_win
++;
1651 } else if (nack
== oack
) {
1652 /* duplicated ack or window probe */
1654 /* duplicated ack, add dup ack count due to whql test up to 1 */
1655 chain
->stat
.dup_ack
++;
1658 /* Coalesce window update */
1659 o_tcp
->th_win
= n_tcp
->th_win
;
1660 chain
->stat
.win_update
++;
1661 return RSC_COALESCE
;
1664 /* pure ack, go to 'C', finalize*/
1665 chain
->stat
.pure_ack
++;
1670 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain
*chain
,
1671 VirtioNetRscSeg
*seg
,
1673 VirtioNetRscUnit
*n_unit
)
1677 uint32_t nseq
, oseq
;
1678 VirtioNetRscUnit
*o_unit
;
1680 o_unit
= &seg
->unit
;
1681 o_ip_len
= htons(*o_unit
->ip_plen
);
1682 nseq
= htonl(n_unit
->tcp
->th_seq
);
1683 oseq
= htonl(o_unit
->tcp
->th_seq
);
1685 /* out of order or retransmitted. */
1686 if ((nseq
- oseq
) > VIRTIO_NET_MAX_TCP_PAYLOAD
) {
1687 chain
->stat
.data_out_of_win
++;
1691 data
= ((uint8_t *)n_unit
->tcp
) + n_unit
->tcp_hdrlen
;
1693 if ((o_unit
->payload
== 0) && n_unit
->payload
) {
1694 /* From no payload to payload, normal case, not a dup ack or etc */
1695 chain
->stat
.data_after_pure_ack
++;
1698 return virtio_net_rsc_handle_ack(chain
, seg
, buf
,
1699 n_unit
->tcp
, o_unit
->tcp
);
1701 } else if ((nseq
- oseq
) != o_unit
->payload
) {
1702 /* Not a consistent packet, out of order */
1703 chain
->stat
.data_out_of_order
++;
1707 if ((o_ip_len
+ n_unit
->payload
) > chain
->max_payload
) {
1708 chain
->stat
.over_size
++;
1712 /* Here comes the right data, the payload length in v4/v6 is different,
1713 so use the field value to update and record the new data len */
1714 o_unit
->payload
+= n_unit
->payload
; /* update new data len */
1716 /* update field in ip header */
1717 *o_unit
->ip_plen
= htons(o_ip_len
+ n_unit
->payload
);
1719 /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
1720 for windows guest, while this may change the behavior for linux
1721 guest (only if it uses RSC feature). */
1722 o_unit
->tcp
->th_offset_flags
= n_unit
->tcp
->th_offset_flags
;
1724 o_unit
->tcp
->th_ack
= n_unit
->tcp
->th_ack
;
1725 o_unit
->tcp
->th_win
= n_unit
->tcp
->th_win
;
1727 memmove(seg
->buf
+ seg
->size
, data
, n_unit
->payload
);
1728 seg
->size
+= n_unit
->payload
;
1730 chain
->stat
.coalesced
++;
1731 return RSC_COALESCE
;
1735 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain
*chain
,
1736 VirtioNetRscSeg
*seg
,
1737 const uint8_t *buf
, size_t size
,
1738 VirtioNetRscUnit
*unit
)
1740 struct ip_header
*ip1
, *ip2
;
1742 ip1
= (struct ip_header
*)(unit
->ip
);
1743 ip2
= (struct ip_header
*)(seg
->unit
.ip
);
1744 if ((ip1
->ip_src
^ ip2
->ip_src
) || (ip1
->ip_dst
^ ip2
->ip_dst
)
1745 || (unit
->tcp
->th_sport
^ seg
->unit
.tcp
->th_sport
)
1746 || (unit
->tcp
->th_dport
^ seg
->unit
.tcp
->th_dport
)) {
1747 chain
->stat
.no_match
++;
1748 return RSC_NO_MATCH
;
1751 return virtio_net_rsc_coalesce_data(chain
, seg
, buf
, unit
);
1754 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain
*chain
,
1755 VirtioNetRscSeg
*seg
,
1756 const uint8_t *buf
, size_t size
,
1757 VirtioNetRscUnit
*unit
)
1759 struct ip6_header
*ip1
, *ip2
;
1761 ip1
= (struct ip6_header
*)(unit
->ip
);
1762 ip2
= (struct ip6_header
*)(seg
->unit
.ip
);
1763 if (memcmp(&ip1
->ip6_src
, &ip2
->ip6_src
, sizeof(struct in6_address
))
1764 || memcmp(&ip1
->ip6_dst
, &ip2
->ip6_dst
, sizeof(struct in6_address
))
1765 || (unit
->tcp
->th_sport
^ seg
->unit
.tcp
->th_sport
)
1766 || (unit
->tcp
->th_dport
^ seg
->unit
.tcp
->th_dport
)) {
1767 chain
->stat
.no_match
++;
1768 return RSC_NO_MATCH
;
1771 return virtio_net_rsc_coalesce_data(chain
, seg
, buf
, unit
);
1774 /* Packets with 'SYN' should bypass, other flag should be sent after drain
1775 * to prevent out of order */
1776 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain
*chain
,
1777 struct tcp_header
*tcp
)
1782 tcp_flag
= htons(tcp
->th_offset_flags
);
1783 tcp_hdr
= (tcp_flag
& VIRTIO_NET_TCP_HDR_LENGTH
) >> 10;
1784 tcp_flag
&= VIRTIO_NET_TCP_FLAG
;
1785 tcp_flag
= htons(tcp
->th_offset_flags
) & 0x3F;
1786 if (tcp_flag
& TH_SYN
) {
1787 chain
->stat
.tcp_syn
++;
1791 if (tcp_flag
& (TH_FIN
| TH_URG
| TH_RST
| TH_ECE
| TH_CWR
)) {
1792 chain
->stat
.tcp_ctrl_drain
++;
1796 if (tcp_hdr
> sizeof(struct tcp_header
)) {
1797 chain
->stat
.tcp_all_opt
++;
1801 return RSC_CANDIDATE
;
1804 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain
*chain
,
1806 const uint8_t *buf
, size_t size
,
1807 VirtioNetRscUnit
*unit
)
1810 VirtioNetRscSeg
*seg
, *nseg
;
1812 if (QTAILQ_EMPTY(&chain
->buffers
)) {
1813 chain
->stat
.empty_cache
++;
1814 virtio_net_rsc_cache_buf(chain
, nc
, buf
, size
);
1815 timer_mod(chain
->drain_timer
,
1816 qemu_clock_get_ns(QEMU_CLOCK_HOST
) + chain
->n
->rsc_timeout
);
1820 QTAILQ_FOREACH_SAFE(seg
, &chain
->buffers
, next
, nseg
) {
1821 if (chain
->proto
== ETH_P_IP
) {
1822 ret
= virtio_net_rsc_coalesce4(chain
, seg
, buf
, size
, unit
);
1824 ret
= virtio_net_rsc_coalesce6(chain
, seg
, buf
, size
, unit
);
1827 if (ret
== RSC_FINAL
) {
1828 if (virtio_net_rsc_drain_seg(chain
, seg
) == 0) {
1830 chain
->stat
.final_failed
++;
1834 /* Send current packet */
1835 return virtio_net_do_receive(nc
, buf
, size
);
1836 } else if (ret
== RSC_NO_MATCH
) {
1839 /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
1840 seg
->is_coalesced
= 1;
1845 chain
->stat
.no_match_cache
++;
1846 virtio_net_rsc_cache_buf(chain
, nc
, buf
, size
);
1850 /* Drain a connection data, this is to avoid out of order segments */
1851 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain
*chain
,
1853 const uint8_t *buf
, size_t size
,
1854 uint16_t ip_start
, uint16_t ip_size
,
1857 VirtioNetRscSeg
*seg
, *nseg
;
1858 uint32_t ppair1
, ppair2
;
1860 ppair1
= *(uint32_t *)(buf
+ tcp_port
);
1861 QTAILQ_FOREACH_SAFE(seg
, &chain
->buffers
, next
, nseg
) {
1862 ppair2
= *(uint32_t *)(seg
->buf
+ tcp_port
);
1863 if (memcmp(buf
+ ip_start
, seg
->buf
+ ip_start
, ip_size
)
1864 || (ppair1
!= ppair2
)) {
1867 if (virtio_net_rsc_drain_seg(chain
, seg
) == 0) {
1868 chain
->stat
.drain_failed
++;
1874 return virtio_net_do_receive(nc
, buf
, size
);
1877 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain
*chain
,
1878 struct ip_header
*ip
,
1879 const uint8_t *buf
, size_t size
)
1883 /* Not an ipv4 packet */
1884 if (((ip
->ip_ver_len
& 0xF0) >> 4) != IP_HEADER_VERSION_4
) {
1885 chain
->stat
.ip_option
++;
1889 /* Don't handle packets with ip option */
1890 if ((ip
->ip_ver_len
& 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH
) {
1891 chain
->stat
.ip_option
++;
1895 if (ip
->ip_p
!= IPPROTO_TCP
) {
1896 chain
->stat
.bypass_not_tcp
++;
1900 /* Don't handle packets with ip fragment */
1901 if (!(htons(ip
->ip_off
) & IP_DF
)) {
1902 chain
->stat
.ip_frag
++;
1906 /* Don't handle packets with ecn flag */
1907 if (IPTOS_ECN(ip
->ip_tos
)) {
1908 chain
->stat
.ip_ecn
++;
1912 ip_len
= htons(ip
->ip_len
);
1913 if (ip_len
< (sizeof(struct ip_header
) + sizeof(struct tcp_header
))
1914 || ip_len
> (size
- chain
->n
->guest_hdr_len
-
1915 sizeof(struct eth_header
))) {
1916 chain
->stat
.ip_hacked
++;
1920 return RSC_CANDIDATE
;
1923 static size_t virtio_net_rsc_receive4(VirtioNetRscChain
*chain
,
1925 const uint8_t *buf
, size_t size
)
1929 VirtioNetRscUnit unit
;
1931 hdr_len
= ((VirtIONet
*)(chain
->n
))->guest_hdr_len
;
1933 if (size
< (hdr_len
+ sizeof(struct eth_header
) + sizeof(struct ip_header
)
1934 + sizeof(struct tcp_header
))) {
1935 chain
->stat
.bypass_not_tcp
++;
1936 return virtio_net_do_receive(nc
, buf
, size
);
1939 virtio_net_rsc_extract_unit4(chain
, buf
, &unit
);
1940 if (virtio_net_rsc_sanity_check4(chain
, unit
.ip
, buf
, size
)
1942 return virtio_net_do_receive(nc
, buf
, size
);
1945 ret
= virtio_net_rsc_tcp_ctrl_check(chain
, unit
.tcp
);
1946 if (ret
== RSC_BYPASS
) {
1947 return virtio_net_do_receive(nc
, buf
, size
);
1948 } else if (ret
== RSC_FINAL
) {
1949 return virtio_net_rsc_drain_flow(chain
, nc
, buf
, size
,
1950 ((hdr_len
+ sizeof(struct eth_header
)) + 12),
1951 VIRTIO_NET_IP4_ADDR_SIZE
,
1952 hdr_len
+ sizeof(struct eth_header
) + sizeof(struct ip_header
));
1955 return virtio_net_rsc_do_coalesce(chain
, nc
, buf
, size
, &unit
);
1958 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain
*chain
,
1959 struct ip6_header
*ip6
,
1960 const uint8_t *buf
, size_t size
)
1964 if (((ip6
->ip6_ctlun
.ip6_un1
.ip6_un1_flow
& 0xF0) >> 4)
1965 != IP_HEADER_VERSION_6
) {
1969 /* Both option and protocol is checked in this */
1970 if (ip6
->ip6_ctlun
.ip6_un1
.ip6_un1_nxt
!= IPPROTO_TCP
) {
1971 chain
->stat
.bypass_not_tcp
++;
1975 ip_len
= htons(ip6
->ip6_ctlun
.ip6_un1
.ip6_un1_plen
);
1976 if (ip_len
< sizeof(struct tcp_header
) ||
1977 ip_len
> (size
- chain
->n
->guest_hdr_len
- sizeof(struct eth_header
)
1978 - sizeof(struct ip6_header
))) {
1979 chain
->stat
.ip_hacked
++;
1983 /* Don't handle packets with ecn flag */
1984 if (IP6_ECN(ip6
->ip6_ctlun
.ip6_un3
.ip6_un3_ecn
)) {
1985 chain
->stat
.ip_ecn
++;
1989 return RSC_CANDIDATE
;
1992 static size_t virtio_net_rsc_receive6(void *opq
, NetClientState
*nc
,
1993 const uint8_t *buf
, size_t size
)
1997 VirtioNetRscChain
*chain
;
1998 VirtioNetRscUnit unit
;
2000 chain
= (VirtioNetRscChain
*)opq
;
2001 hdr_len
= ((VirtIONet
*)(chain
->n
))->guest_hdr_len
;
2003 if (size
< (hdr_len
+ sizeof(struct eth_header
) + sizeof(struct ip6_header
)
2004 + sizeof(tcp_header
))) {
2005 return virtio_net_do_receive(nc
, buf
, size
);
2008 virtio_net_rsc_extract_unit6(chain
, buf
, &unit
);
2009 if (RSC_CANDIDATE
!= virtio_net_rsc_sanity_check6(chain
,
2010 unit
.ip
, buf
, size
)) {
2011 return virtio_net_do_receive(nc
, buf
, size
);
2014 ret
= virtio_net_rsc_tcp_ctrl_check(chain
, unit
.tcp
);
2015 if (ret
== RSC_BYPASS
) {
2016 return virtio_net_do_receive(nc
, buf
, size
);
2017 } else if (ret
== RSC_FINAL
) {
2018 return virtio_net_rsc_drain_flow(chain
, nc
, buf
, size
,
2019 ((hdr_len
+ sizeof(struct eth_header
)) + 8),
2020 VIRTIO_NET_IP6_ADDR_SIZE
,
2021 hdr_len
+ sizeof(struct eth_header
)
2022 + sizeof(struct ip6_header
));
2025 return virtio_net_rsc_do_coalesce(chain
, nc
, buf
, size
, &unit
);
2028 static VirtioNetRscChain
*virtio_net_rsc_lookup_chain(VirtIONet
*n
,
2032 VirtioNetRscChain
*chain
;
2034 if ((proto
!= (uint16_t)ETH_P_IP
) && (proto
!= (uint16_t)ETH_P_IPV6
)) {
2038 QTAILQ_FOREACH(chain
, &n
->rsc_chains
, next
) {
2039 if (chain
->proto
== proto
) {
2044 chain
= g_malloc(sizeof(*chain
));
2046 chain
->proto
= proto
;
2047 if (proto
== (uint16_t)ETH_P_IP
) {
2048 chain
->max_payload
= VIRTIO_NET_MAX_IP4_PAYLOAD
;
2049 chain
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
2051 chain
->max_payload
= VIRTIO_NET_MAX_IP6_PAYLOAD
;
2052 chain
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
2054 chain
->drain_timer
= timer_new_ns(QEMU_CLOCK_HOST
,
2055 virtio_net_rsc_purge
, chain
);
2056 memset(&chain
->stat
, 0, sizeof(chain
->stat
));
2058 QTAILQ_INIT(&chain
->buffers
);
2059 QTAILQ_INSERT_TAIL(&n
->rsc_chains
, chain
, next
);
2064 static ssize_t
virtio_net_rsc_receive(NetClientState
*nc
,
2069 VirtioNetRscChain
*chain
;
2070 struct eth_header
*eth
;
2073 n
= qemu_get_nic_opaque(nc
);
2074 if (size
< (n
->host_hdr_len
+ sizeof(struct eth_header
))) {
2075 return virtio_net_do_receive(nc
, buf
, size
);
2078 eth
= (struct eth_header
*)(buf
+ n
->guest_hdr_len
);
2079 proto
= htons(eth
->h_proto
);
2081 chain
= virtio_net_rsc_lookup_chain(n
, nc
, proto
);
2083 chain
->stat
.received
++;
2084 if (proto
== (uint16_t)ETH_P_IP
&& n
->rsc4_enabled
) {
2085 return virtio_net_rsc_receive4(chain
, nc
, buf
, size
);
2086 } else if (proto
== (uint16_t)ETH_P_IPV6
&& n
->rsc6_enabled
) {
2087 return virtio_net_rsc_receive6(chain
, nc
, buf
, size
);
2090 return virtio_net_do_receive(nc
, buf
, size
);
2093 static ssize_t
virtio_net_receive(NetClientState
*nc
, const uint8_t *buf
,
2096 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
2097 if ((n
->rsc4_enabled
|| n
->rsc6_enabled
)) {
2098 return virtio_net_rsc_receive(nc
, buf
, size
);
2100 return virtio_net_do_receive(nc
, buf
, size
);
2104 static int32_t virtio_net_flush_tx(VirtIONetQueue
*q
);
2106 static void virtio_net_tx_complete(NetClientState
*nc
, ssize_t len
)
2108 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
2109 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
2110 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2112 virtqueue_push(q
->tx_vq
, q
->async_tx
.elem
, 0);
2113 virtio_notify(vdev
, q
->tx_vq
);
2115 g_free(q
->async_tx
.elem
);
2116 q
->async_tx
.elem
= NULL
;
2118 virtio_queue_set_notification(q
->tx_vq
, 1);
2119 virtio_net_flush_tx(q
);
2123 static int32_t virtio_net_flush_tx(VirtIONetQueue
*q
)
2125 VirtIONet
*n
= q
->n
;
2126 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2127 VirtQueueElement
*elem
;
2128 int32_t num_packets
= 0;
2129 int queue_index
= vq2q(virtio_get_queue_index(q
->tx_vq
));
2130 if (!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
2134 if (q
->async_tx
.elem
) {
2135 virtio_queue_set_notification(q
->tx_vq
, 0);
2141 unsigned int out_num
;
2142 struct iovec sg
[VIRTQUEUE_MAX_SIZE
], sg2
[VIRTQUEUE_MAX_SIZE
+ 1], *out_sg
;
2143 struct virtio_net_hdr_mrg_rxbuf mhdr
;
2145 elem
= virtqueue_pop(q
->tx_vq
, sizeof(VirtQueueElement
));
2150 out_num
= elem
->out_num
;
2151 out_sg
= elem
->out_sg
;
2153 virtio_error(vdev
, "virtio-net header not in first element");
2154 virtqueue_detach_element(q
->tx_vq
, elem
, 0);
2159 if (n
->has_vnet_hdr
) {
2160 if (iov_to_buf(out_sg
, out_num
, 0, &mhdr
, n
->guest_hdr_len
) <
2162 virtio_error(vdev
, "virtio-net header incorrect");
2163 virtqueue_detach_element(q
->tx_vq
, elem
, 0);
2167 if (n
->needs_vnet_hdr_swap
) {
2168 virtio_net_hdr_swap(vdev
, (void *) &mhdr
);
2169 sg2
[0].iov_base
= &mhdr
;
2170 sg2
[0].iov_len
= n
->guest_hdr_len
;
2171 out_num
= iov_copy(&sg2
[1], ARRAY_SIZE(sg2
) - 1,
2173 n
->guest_hdr_len
, -1);
2174 if (out_num
== VIRTQUEUE_MAX_SIZE
) {
2182 * If host wants to see the guest header as is, we can
2183 * pass it on unchanged. Otherwise, copy just the parts
2184 * that host is interested in.
2186 assert(n
->host_hdr_len
<= n
->guest_hdr_len
);
2187 if (n
->host_hdr_len
!= n
->guest_hdr_len
) {
2188 unsigned sg_num
= iov_copy(sg
, ARRAY_SIZE(sg
),
2190 0, n
->host_hdr_len
);
2191 sg_num
+= iov_copy(sg
+ sg_num
, ARRAY_SIZE(sg
) - sg_num
,
2193 n
->guest_hdr_len
, -1);
2198 ret
= qemu_sendv_packet_async(qemu_get_subqueue(n
->nic
, queue_index
),
2199 out_sg
, out_num
, virtio_net_tx_complete
);
2201 virtio_queue_set_notification(q
->tx_vq
, 0);
2202 q
->async_tx
.elem
= elem
;
2207 virtqueue_push(q
->tx_vq
, elem
, 0);
2208 virtio_notify(vdev
, q
->tx_vq
);
2211 if (++num_packets
>= n
->tx_burst
) {
2218 static void virtio_net_handle_tx_timer(VirtIODevice
*vdev
, VirtQueue
*vq
)
2220 VirtIONet
*n
= VIRTIO_NET(vdev
);
2221 VirtIONetQueue
*q
= &n
->vqs
[vq2q(virtio_get_queue_index(vq
))];
2223 if (unlikely((n
->status
& VIRTIO_NET_S_LINK_UP
) == 0)) {
2224 virtio_net_drop_tx_queue_data(vdev
, vq
);
2228 /* This happens when device was stopped but VCPU wasn't. */
2229 if (!vdev
->vm_running
) {
2234 if (q
->tx_waiting
) {
2235 virtio_queue_set_notification(vq
, 1);
2236 timer_del(q
->tx_timer
);
2238 if (virtio_net_flush_tx(q
) == -EINVAL
) {
2242 timer_mod(q
->tx_timer
,
2243 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) + n
->tx_timeout
);
2245 virtio_queue_set_notification(vq
, 0);
2249 static void virtio_net_handle_tx_bh(VirtIODevice
*vdev
, VirtQueue
*vq
)
2251 VirtIONet
*n
= VIRTIO_NET(vdev
);
2252 VirtIONetQueue
*q
= &n
->vqs
[vq2q(virtio_get_queue_index(vq
))];
2254 if (unlikely((n
->status
& VIRTIO_NET_S_LINK_UP
) == 0)) {
2255 virtio_net_drop_tx_queue_data(vdev
, vq
);
2259 if (unlikely(q
->tx_waiting
)) {
2263 /* This happens when device was stopped but VCPU wasn't. */
2264 if (!vdev
->vm_running
) {
2267 virtio_queue_set_notification(vq
, 0);
2268 qemu_bh_schedule(q
->tx_bh
);
2271 static void virtio_net_tx_timer(void *opaque
)
2273 VirtIONetQueue
*q
= opaque
;
2274 VirtIONet
*n
= q
->n
;
2275 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2276 /* This happens when device was stopped but BH wasn't. */
2277 if (!vdev
->vm_running
) {
2278 /* Make sure tx waiting is set, so we'll run when restarted. */
2279 assert(q
->tx_waiting
);
2285 /* Just in case the driver is not ready on more */
2286 if (!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
2290 virtio_queue_set_notification(q
->tx_vq
, 1);
2291 virtio_net_flush_tx(q
);
2294 static void virtio_net_tx_bh(void *opaque
)
2296 VirtIONetQueue
*q
= opaque
;
2297 VirtIONet
*n
= q
->n
;
2298 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2301 /* This happens when device was stopped but BH wasn't. */
2302 if (!vdev
->vm_running
) {
2303 /* Make sure tx waiting is set, so we'll run when restarted. */
2304 assert(q
->tx_waiting
);
2310 /* Just in case the driver is not ready on more */
2311 if (unlikely(!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
))) {
2315 ret
= virtio_net_flush_tx(q
);
2316 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
2317 return; /* Notification re-enable handled by tx_complete or device
2321 /* If we flush a full burst of packets, assume there are
2322 * more coming and immediately reschedule */
2323 if (ret
>= n
->tx_burst
) {
2324 qemu_bh_schedule(q
->tx_bh
);
2329 /* If less than a full burst, re-enable notification and flush
2330 * anything that may have come in while we weren't looking. If
2331 * we find something, assume the guest is still active and reschedule */
2332 virtio_queue_set_notification(q
->tx_vq
, 1);
2333 ret
= virtio_net_flush_tx(q
);
2334 if (ret
== -EINVAL
) {
2336 } else if (ret
> 0) {
2337 virtio_queue_set_notification(q
->tx_vq
, 0);
2338 qemu_bh_schedule(q
->tx_bh
);
2343 static void virtio_net_add_queue(VirtIONet
*n
, int index
)
2345 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2347 n
->vqs
[index
].rx_vq
= virtio_add_queue(vdev
, n
->net_conf
.rx_queue_size
,
2348 virtio_net_handle_rx
);
2350 if (n
->net_conf
.tx
&& !strcmp(n
->net_conf
.tx
, "timer")) {
2351 n
->vqs
[index
].tx_vq
=
2352 virtio_add_queue(vdev
, n
->net_conf
.tx_queue_size
,
2353 virtio_net_handle_tx_timer
);
2354 n
->vqs
[index
].tx_timer
= timer_new_ns(QEMU_CLOCK_VIRTUAL
,
2355 virtio_net_tx_timer
,
2358 n
->vqs
[index
].tx_vq
=
2359 virtio_add_queue(vdev
, n
->net_conf
.tx_queue_size
,
2360 virtio_net_handle_tx_bh
);
2361 n
->vqs
[index
].tx_bh
= qemu_bh_new(virtio_net_tx_bh
, &n
->vqs
[index
]);
2364 n
->vqs
[index
].tx_waiting
= 0;
2365 n
->vqs
[index
].n
= n
;
2368 static void virtio_net_del_queue(VirtIONet
*n
, int index
)
2370 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2371 VirtIONetQueue
*q
= &n
->vqs
[index
];
2372 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, index
);
2374 qemu_purge_queued_packets(nc
);
2376 virtio_del_queue(vdev
, index
* 2);
2378 timer_del(q
->tx_timer
);
2379 timer_free(q
->tx_timer
);
2382 qemu_bh_delete(q
->tx_bh
);
2386 virtio_del_queue(vdev
, index
* 2 + 1);
2389 static void virtio_net_change_num_queues(VirtIONet
*n
, int new_max_queues
)
2391 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2392 int old_num_queues
= virtio_get_num_queues(vdev
);
2393 int new_num_queues
= new_max_queues
* 2 + 1;
2396 assert(old_num_queues
>= 3);
2397 assert(old_num_queues
% 2 == 1);
2399 if (old_num_queues
== new_num_queues
) {
2404 * We always need to remove and add ctrl vq if
2405 * old_num_queues != new_num_queues. Remove ctrl_vq first,
2406 * and then we only enter one of the following two loops.
2408 virtio_del_queue(vdev
, old_num_queues
- 1);
2410 for (i
= new_num_queues
- 1; i
< old_num_queues
- 1; i
+= 2) {
2411 /* new_num_queues < old_num_queues */
2412 virtio_net_del_queue(n
, i
/ 2);
2415 for (i
= old_num_queues
- 1; i
< new_num_queues
- 1; i
+= 2) {
2416 /* new_num_queues > old_num_queues */
2417 virtio_net_add_queue(n
, i
/ 2);
2420 /* add ctrl_vq last */
2421 n
->ctrl_vq
= virtio_add_queue(vdev
, 64, virtio_net_handle_ctrl
);
2424 static void virtio_net_set_multiqueue(VirtIONet
*n
, int multiqueue
)
2426 int max
= multiqueue
? n
->max_queues
: 1;
2428 n
->multiqueue
= multiqueue
;
2429 virtio_net_change_num_queues(n
, max
);
2431 virtio_net_set_queues(n
);
2434 static int virtio_net_post_load_device(void *opaque
, int version_id
)
2436 VirtIONet
*n
= opaque
;
2437 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2440 trace_virtio_net_post_load_device();
2441 virtio_net_set_mrg_rx_bufs(n
, n
->mergeable_rx_bufs
,
2442 virtio_vdev_has_feature(vdev
,
2443 VIRTIO_F_VERSION_1
));
2445 /* MAC_TABLE_ENTRIES may be different from the saved image */
2446 if (n
->mac_table
.in_use
> MAC_TABLE_ENTRIES
) {
2447 n
->mac_table
.in_use
= 0;
2450 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
)) {
2451 n
->curr_guest_offloads
= virtio_net_supported_guest_offloads(n
);
2455 * curr_guest_offloads will be later overwritten by the
2456 * virtio_set_features_nocheck call done from the virtio_load.
2457 * Here we make sure it is preserved and restored accordingly
2458 * in the virtio_net_post_load_virtio callback.
2460 n
->saved_guest_offloads
= n
->curr_guest_offloads
;
2462 virtio_net_set_queues(n
);
2464 /* Find the first multicast entry in the saved MAC filter */
2465 for (i
= 0; i
< n
->mac_table
.in_use
; i
++) {
2466 if (n
->mac_table
.macs
[i
* ETH_ALEN
] & 1) {
2470 n
->mac_table
.first_multi
= i
;
2472 /* nc.link_down can't be migrated, so infer link_down according
2473 * to link status bit in n->status */
2474 link_down
= (n
->status
& VIRTIO_NET_S_LINK_UP
) == 0;
2475 for (i
= 0; i
< n
->max_queues
; i
++) {
2476 qemu_get_subqueue(n
->nic
, i
)->link_down
= link_down
;
2479 if (virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
) &&
2480 virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
)) {
2481 qemu_announce_timer_reset(&n
->announce_timer
, migrate_announce_params(),
2483 virtio_net_announce_timer
, n
);
2484 if (n
->announce_timer
.round
) {
2485 timer_mod(n
->announce_timer
.tm
,
2486 qemu_clock_get_ms(n
->announce_timer
.type
));
2488 qemu_announce_timer_del(&n
->announce_timer
, false);
2495 static int virtio_net_post_load_virtio(VirtIODevice
*vdev
)
2497 VirtIONet
*n
= VIRTIO_NET(vdev
);
2499 * The actual needed state is now in saved_guest_offloads,
2500 * see virtio_net_post_load_device for detail.
2501 * Restore it back and apply the desired offloads.
2503 n
->curr_guest_offloads
= n
->saved_guest_offloads
;
2504 if (peer_has_vnet_hdr(n
)) {
2505 virtio_net_apply_guest_offloads(n
);
2511 /* tx_waiting field of a VirtIONetQueue */
2512 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting
= {
2513 .name
= "virtio-net-queue-tx_waiting",
2514 .fields
= (VMStateField
[]) {
2515 VMSTATE_UINT32(tx_waiting
, VirtIONetQueue
),
2516 VMSTATE_END_OF_LIST()
2520 static bool max_queues_gt_1(void *opaque
, int version_id
)
2522 return VIRTIO_NET(opaque
)->max_queues
> 1;
2525 static bool has_ctrl_guest_offloads(void *opaque
, int version_id
)
2527 return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque
),
2528 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
);
2531 static bool mac_table_fits(void *opaque
, int version_id
)
2533 return VIRTIO_NET(opaque
)->mac_table
.in_use
<= MAC_TABLE_ENTRIES
;
2536 static bool mac_table_doesnt_fit(void *opaque
, int version_id
)
2538 return !mac_table_fits(opaque
, version_id
);
2541 /* This temporary type is shared by all the WITH_TMP methods
2542 * although only some fields are used by each.
2544 struct VirtIONetMigTmp
{
2546 VirtIONetQueue
*vqs_1
;
2547 uint16_t curr_queues_1
;
2549 uint32_t has_vnet_hdr
;
2552 /* The 2nd and subsequent tx_waiting flags are loaded later than
2553 * the 1st entry in the queues and only if there's more than one
2554 * entry. We use the tmp mechanism to calculate a temporary
2555 * pointer and count and also validate the count.
2558 static int virtio_net_tx_waiting_pre_save(void *opaque
)
2560 struct VirtIONetMigTmp
*tmp
= opaque
;
2562 tmp
->vqs_1
= tmp
->parent
->vqs
+ 1;
2563 tmp
->curr_queues_1
= tmp
->parent
->curr_queues
- 1;
2564 if (tmp
->parent
->curr_queues
== 0) {
2565 tmp
->curr_queues_1
= 0;
2571 static int virtio_net_tx_waiting_pre_load(void *opaque
)
2573 struct VirtIONetMigTmp
*tmp
= opaque
;
2575 /* Reuse the pointer setup from save */
2576 virtio_net_tx_waiting_pre_save(opaque
);
2578 if (tmp
->parent
->curr_queues
> tmp
->parent
->max_queues
) {
2579 error_report("virtio-net: curr_queues %x > max_queues %x",
2580 tmp
->parent
->curr_queues
, tmp
->parent
->max_queues
);
2585 return 0; /* all good */
2588 static const VMStateDescription vmstate_virtio_net_tx_waiting
= {
2589 .name
= "virtio-net-tx_waiting",
2590 .pre_load
= virtio_net_tx_waiting_pre_load
,
2591 .pre_save
= virtio_net_tx_waiting_pre_save
,
2592 .fields
= (VMStateField
[]) {
2593 VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1
, struct VirtIONetMigTmp
,
2595 vmstate_virtio_net_queue_tx_waiting
,
2596 struct VirtIONetQueue
),
2597 VMSTATE_END_OF_LIST()
2601 /* the 'has_ufo' flag is just tested; if the incoming stream has the
2602 * flag set we need to check that we have it
2604 static int virtio_net_ufo_post_load(void *opaque
, int version_id
)
2606 struct VirtIONetMigTmp
*tmp
= opaque
;
2608 if (tmp
->has_ufo
&& !peer_has_ufo(tmp
->parent
)) {
2609 error_report("virtio-net: saved image requires TUN_F_UFO support");
2616 static int virtio_net_ufo_pre_save(void *opaque
)
2618 struct VirtIONetMigTmp
*tmp
= opaque
;
2620 tmp
->has_ufo
= tmp
->parent
->has_ufo
;
2625 static const VMStateDescription vmstate_virtio_net_has_ufo
= {
2626 .name
= "virtio-net-ufo",
2627 .post_load
= virtio_net_ufo_post_load
,
2628 .pre_save
= virtio_net_ufo_pre_save
,
2629 .fields
= (VMStateField
[]) {
2630 VMSTATE_UINT8(has_ufo
, struct VirtIONetMigTmp
),
2631 VMSTATE_END_OF_LIST()
2635 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
2636 * flag set we need to check that we have it
2638 static int virtio_net_vnet_post_load(void *opaque
, int version_id
)
2640 struct VirtIONetMigTmp
*tmp
= opaque
;
2642 if (tmp
->has_vnet_hdr
&& !peer_has_vnet_hdr(tmp
->parent
)) {
2643 error_report("virtio-net: saved image requires vnet_hdr=on");
2650 static int virtio_net_vnet_pre_save(void *opaque
)
2652 struct VirtIONetMigTmp
*tmp
= opaque
;
2654 tmp
->has_vnet_hdr
= tmp
->parent
->has_vnet_hdr
;
2659 static const VMStateDescription vmstate_virtio_net_has_vnet
= {
2660 .name
= "virtio-net-vnet",
2661 .post_load
= virtio_net_vnet_post_load
,
2662 .pre_save
= virtio_net_vnet_pre_save
,
2663 .fields
= (VMStateField
[]) {
2664 VMSTATE_UINT32(has_vnet_hdr
, struct VirtIONetMigTmp
),
2665 VMSTATE_END_OF_LIST()
2669 static const VMStateDescription vmstate_virtio_net_device
= {
2670 .name
= "virtio-net-device",
2671 .version_id
= VIRTIO_NET_VM_VERSION
,
2672 .minimum_version_id
= VIRTIO_NET_VM_VERSION
,
2673 .post_load
= virtio_net_post_load_device
,
2674 .fields
= (VMStateField
[]) {
2675 VMSTATE_UINT8_ARRAY(mac
, VirtIONet
, ETH_ALEN
),
2676 VMSTATE_STRUCT_POINTER(vqs
, VirtIONet
,
2677 vmstate_virtio_net_queue_tx_waiting
,
2679 VMSTATE_UINT32(mergeable_rx_bufs
, VirtIONet
),
2680 VMSTATE_UINT16(status
, VirtIONet
),
2681 VMSTATE_UINT8(promisc
, VirtIONet
),
2682 VMSTATE_UINT8(allmulti
, VirtIONet
),
2683 VMSTATE_UINT32(mac_table
.in_use
, VirtIONet
),
2685 /* Guarded pair: If it fits we load it, else we throw it away
2686 * - can happen if source has a larger MAC table.; post-load
2687 * sets flags in this case.
2689 VMSTATE_VBUFFER_MULTIPLY(mac_table
.macs
, VirtIONet
,
2690 0, mac_table_fits
, mac_table
.in_use
,
2692 VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet
, mac_table_doesnt_fit
, 0,
2693 mac_table
.in_use
, ETH_ALEN
),
2695 /* Note: This is an array of uint32's that's always been saved as a
2696 * buffer; hold onto your endiannesses; it's actually used as a bitmap
2697 * but based on the uint.
2699 VMSTATE_BUFFER_POINTER_UNSAFE(vlans
, VirtIONet
, 0, MAX_VLAN
>> 3),
2700 VMSTATE_WITH_TMP(VirtIONet
, struct VirtIONetMigTmp
,
2701 vmstate_virtio_net_has_vnet
),
2702 VMSTATE_UINT8(mac_table
.multi_overflow
, VirtIONet
),
2703 VMSTATE_UINT8(mac_table
.uni_overflow
, VirtIONet
),
2704 VMSTATE_UINT8(alluni
, VirtIONet
),
2705 VMSTATE_UINT8(nomulti
, VirtIONet
),
2706 VMSTATE_UINT8(nouni
, VirtIONet
),
2707 VMSTATE_UINT8(nobcast
, VirtIONet
),
2708 VMSTATE_WITH_TMP(VirtIONet
, struct VirtIONetMigTmp
,
2709 vmstate_virtio_net_has_ufo
),
2710 VMSTATE_SINGLE_TEST(max_queues
, VirtIONet
, max_queues_gt_1
, 0,
2711 vmstate_info_uint16_equal
, uint16_t),
2712 VMSTATE_UINT16_TEST(curr_queues
, VirtIONet
, max_queues_gt_1
),
2713 VMSTATE_WITH_TMP(VirtIONet
, struct VirtIONetMigTmp
,
2714 vmstate_virtio_net_tx_waiting
),
2715 VMSTATE_UINT64_TEST(curr_guest_offloads
, VirtIONet
,
2716 has_ctrl_guest_offloads
),
2717 VMSTATE_END_OF_LIST()
2721 static NetClientInfo net_virtio_info
= {
2722 .type
= NET_CLIENT_DRIVER_NIC
,
2723 .size
= sizeof(NICState
),
2724 .can_receive
= virtio_net_can_receive
,
2725 .receive
= virtio_net_receive
,
2726 .link_status_changed
= virtio_net_set_link_status
,
2727 .query_rx_filter
= virtio_net_query_rxfilter
,
2728 .announce
= virtio_net_announce
,
2731 static bool virtio_net_guest_notifier_pending(VirtIODevice
*vdev
, int idx
)
2733 VirtIONet
*n
= VIRTIO_NET(vdev
);
2734 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, vq2q(idx
));
2735 assert(n
->vhost_started
);
2736 return vhost_net_virtqueue_pending(get_vhost_net(nc
->peer
), idx
);
2739 static void virtio_net_guest_notifier_mask(VirtIODevice
*vdev
, int idx
,
2742 VirtIONet
*n
= VIRTIO_NET(vdev
);
2743 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, vq2q(idx
));
2744 assert(n
->vhost_started
);
2745 vhost_net_virtqueue_mask(get_vhost_net(nc
->peer
),
2749 static void virtio_net_set_config_size(VirtIONet
*n
, uint64_t host_features
)
2751 virtio_add_feature(&host_features
, VIRTIO_NET_F_MAC
);
2753 n
->config_size
= virtio_feature_get_config_size(feature_sizes
,
2757 void virtio_net_set_netclient_name(VirtIONet
*n
, const char *name
,
2761 * The name can be NULL, the netclient name will be type.x.
2763 assert(type
!= NULL
);
2765 g_free(n
->netclient_name
);
2766 g_free(n
->netclient_type
);
2767 n
->netclient_name
= g_strdup(name
);
2768 n
->netclient_type
= g_strdup(type
);
2771 static bool failover_unplug_primary(VirtIONet
*n
)
2773 HotplugHandler
*hotplug_ctrl
;
2777 hotplug_ctrl
= qdev_get_hotplug_handler(n
->primary_dev
);
2779 pci_dev
= PCI_DEVICE(n
->primary_dev
);
2780 pci_dev
->partially_hotplugged
= true;
2781 hotplug_handler_unplug_request(hotplug_ctrl
, n
->primary_dev
, &err
);
2783 error_report_err(err
);
2792 static bool failover_replug_primary(VirtIONet
*n
, Error
**errp
)
2794 HotplugHandler
*hotplug_ctrl
;
2795 PCIDevice
*pdev
= PCI_DEVICE(n
->primary_dev
);
2797 if (!pdev
->partially_hotplugged
) {
2800 if (!n
->primary_device_opts
) {
2801 n
->primary_device_opts
= qemu_opts_from_qdict(
2802 qemu_find_opts("device"),
2803 n
->primary_device_dict
, errp
);
2805 if (n
->primary_device_opts
) {
2806 if (n
->primary_dev
) {
2807 n
->primary_bus
= n
->primary_dev
->parent_bus
;
2809 qdev_set_parent_bus(n
->primary_dev
, n
->primary_bus
);
2810 n
->primary_should_be_hidden
= false;
2811 qemu_opt_set_bool(n
->primary_device_opts
,
2812 "partially_hotplugged", true, errp
);
2813 hotplug_ctrl
= qdev_get_hotplug_handler(n
->primary_dev
);
2815 hotplug_handler_pre_plug(hotplug_ctrl
, n
->primary_dev
, errp
);
2816 hotplug_handler_plug(hotplug_ctrl
, n
->primary_dev
, errp
);
2818 if (!n
->primary_dev
) {
2819 error_setg(errp
, "virtio_net: couldn't find primary device");
2822 return *errp
!= NULL
;
2825 static void virtio_net_handle_migration_primary(VirtIONet
*n
,
2828 bool should_be_hidden
;
2831 should_be_hidden
= atomic_read(&n
->primary_should_be_hidden
);
2833 if (!n
->primary_dev
) {
2834 n
->primary_dev
= virtio_connect_failover_devices(n
, n
->qdev
, &err
);
2835 if (!n
->primary_dev
) {
2840 if (migration_in_setup(s
) && !should_be_hidden
&&
2842 if (failover_unplug_primary(n
)) {
2843 vmstate_unregister(n
->primary_dev
, qdev_get_vmsd(n
->primary_dev
),
2845 qapi_event_send_unplug_primary(n
->primary_device_id
);
2846 atomic_set(&n
->primary_should_be_hidden
, true);
2848 warn_report("couldn't unplug primary device");
2850 } else if (migration_has_failed(s
)) {
2851 /* We already unplugged the device let's plugged it back */
2852 if (!failover_replug_primary(n
, &err
)) {
2854 error_report_err(err
);
2860 static void virtio_net_migration_state_notifier(Notifier
*notifier
, void *data
)
2862 MigrationState
*s
= data
;
2863 VirtIONet
*n
= container_of(notifier
, VirtIONet
, migration_state
);
2864 virtio_net_handle_migration_primary(n
, s
);
2867 static int virtio_net_primary_should_be_hidden(DeviceListener
*listener
,
2868 QemuOpts
*device_opts
)
2870 VirtIONet
*n
= container_of(listener
, VirtIONet
, primary_listener
);
2874 n
->primary_device_dict
= qemu_opts_to_qdict(device_opts
,
2875 n
->primary_device_dict
);
2876 if (n
->primary_device_dict
) {
2877 g_free(n
->standby_id
);
2878 n
->standby_id
= g_strdup(qdict_get_try_str(n
->primary_device_dict
,
2879 "failover_pair_id"));
2881 if (device_opts
&& g_strcmp0(n
->standby_id
, n
->netclient_name
) == 0) {
2884 match_found
= false;
2886 g_free(n
->standby_id
);
2887 n
->primary_device_dict
= NULL
;
2891 n
->primary_device_opts
= device_opts
;
2893 /* primary_should_be_hidden is set during feature negotiation */
2894 hide
= atomic_read(&n
->primary_should_be_hidden
);
2896 if (n
->primary_device_dict
) {
2897 g_free(n
->primary_device_id
);
2898 n
->primary_device_id
= g_strdup(qdict_get_try_str(
2899 n
->primary_device_dict
, "id"));
2900 if (!n
->primary_device_id
) {
2901 warn_report("primary_device_id not set");
2906 if (match_found
&& hide
) {
2908 } else if (match_found
&& !hide
) {
2915 static void virtio_net_device_realize(DeviceState
*dev
, Error
**errp
)
2917 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
2918 VirtIONet
*n
= VIRTIO_NET(dev
);
2922 if (n
->net_conf
.mtu
) {
2923 n
->host_features
|= (1ULL << VIRTIO_NET_F_MTU
);
2926 if (n
->net_conf
.duplex_str
) {
2927 if (strncmp(n
->net_conf
.duplex_str
, "half", 5) == 0) {
2928 n
->net_conf
.duplex
= DUPLEX_HALF
;
2929 } else if (strncmp(n
->net_conf
.duplex_str
, "full", 5) == 0) {
2930 n
->net_conf
.duplex
= DUPLEX_FULL
;
2932 error_setg(errp
, "'duplex' must be 'half' or 'full'");
2934 n
->host_features
|= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX
);
2936 n
->net_conf
.duplex
= DUPLEX_UNKNOWN
;
2939 if (n
->net_conf
.speed
< SPEED_UNKNOWN
) {
2940 error_setg(errp
, "'speed' must be between 0 and INT_MAX");
2941 } else if (n
->net_conf
.speed
>= 0) {
2942 n
->host_features
|= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX
);
2946 n
->primary_listener
.should_be_hidden
=
2947 virtio_net_primary_should_be_hidden
;
2948 atomic_set(&n
->primary_should_be_hidden
, true);
2949 device_listener_register(&n
->primary_listener
);
2950 n
->migration_state
.notify
= virtio_net_migration_state_notifier
;
2951 add_migration_state_change_notifier(&n
->migration_state
);
2952 n
->host_features
|= (1ULL << VIRTIO_NET_F_STANDBY
);
2955 virtio_net_set_config_size(n
, n
->host_features
);
2956 virtio_init(vdev
, "virtio-net", VIRTIO_ID_NET
, n
->config_size
);
2959 * We set a lower limit on RX queue size to what it always was.
2960 * Guests that want a smaller ring can always resize it without
2961 * help from us (using virtio 1 and up).
2963 if (n
->net_conf
.rx_queue_size
< VIRTIO_NET_RX_QUEUE_MIN_SIZE
||
2964 n
->net_conf
.rx_queue_size
> VIRTQUEUE_MAX_SIZE
||
2965 !is_power_of_2(n
->net_conf
.rx_queue_size
)) {
2966 error_setg(errp
, "Invalid rx_queue_size (= %" PRIu16
"), "
2967 "must be a power of 2 between %d and %d.",
2968 n
->net_conf
.rx_queue_size
, VIRTIO_NET_RX_QUEUE_MIN_SIZE
,
2969 VIRTQUEUE_MAX_SIZE
);
2970 virtio_cleanup(vdev
);
2974 if (n
->net_conf
.tx_queue_size
< VIRTIO_NET_TX_QUEUE_MIN_SIZE
||
2975 n
->net_conf
.tx_queue_size
> VIRTQUEUE_MAX_SIZE
||
2976 !is_power_of_2(n
->net_conf
.tx_queue_size
)) {
2977 error_setg(errp
, "Invalid tx_queue_size (= %" PRIu16
"), "
2978 "must be a power of 2 between %d and %d",
2979 n
->net_conf
.tx_queue_size
, VIRTIO_NET_TX_QUEUE_MIN_SIZE
,
2980 VIRTQUEUE_MAX_SIZE
);
2981 virtio_cleanup(vdev
);
2985 n
->max_queues
= MAX(n
->nic_conf
.peers
.queues
, 1);
2986 if (n
->max_queues
* 2 + 1 > VIRTIO_QUEUE_MAX
) {
2987 error_setg(errp
, "Invalid number of queues (= %" PRIu32
"), "
2988 "must be a positive integer less than %d.",
2989 n
->max_queues
, (VIRTIO_QUEUE_MAX
- 1) / 2);
2990 virtio_cleanup(vdev
);
2993 n
->vqs
= g_malloc0(sizeof(VirtIONetQueue
) * n
->max_queues
);
2995 n
->tx_timeout
= n
->net_conf
.txtimer
;
2997 if (n
->net_conf
.tx
&& strcmp(n
->net_conf
.tx
, "timer")
2998 && strcmp(n
->net_conf
.tx
, "bh")) {
2999 warn_report("virtio-net: "
3000 "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
3002 error_printf("Defaulting to \"bh\"");
3005 n
->net_conf
.tx_queue_size
= MIN(virtio_net_max_tx_queue_size(n
),
3006 n
->net_conf
.tx_queue_size
);
3008 for (i
= 0; i
< n
->max_queues
; i
++) {
3009 virtio_net_add_queue(n
, i
);
3012 n
->ctrl_vq
= virtio_add_queue(vdev
, 64, virtio_net_handle_ctrl
);
3013 qemu_macaddr_default_if_unset(&n
->nic_conf
.macaddr
);
3014 memcpy(&n
->mac
[0], &n
->nic_conf
.macaddr
, sizeof(n
->mac
));
3015 n
->status
= VIRTIO_NET_S_LINK_UP
;
3016 qemu_announce_timer_reset(&n
->announce_timer
, migrate_announce_params(),
3018 virtio_net_announce_timer
, n
);
3019 n
->announce_timer
.round
= 0;
3021 if (n
->netclient_type
) {
3023 * Happen when virtio_net_set_netclient_name has been called.
3025 n
->nic
= qemu_new_nic(&net_virtio_info
, &n
->nic_conf
,
3026 n
->netclient_type
, n
->netclient_name
, n
);
3028 n
->nic
= qemu_new_nic(&net_virtio_info
, &n
->nic_conf
,
3029 object_get_typename(OBJECT(dev
)), dev
->id
, n
);
3032 peer_test_vnet_hdr(n
);
3033 if (peer_has_vnet_hdr(n
)) {
3034 for (i
= 0; i
< n
->max_queues
; i
++) {
3035 qemu_using_vnet_hdr(qemu_get_subqueue(n
->nic
, i
)->peer
, true);
3037 n
->host_hdr_len
= sizeof(struct virtio_net_hdr
);
3039 n
->host_hdr_len
= 0;
3042 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->nic_conf
.macaddr
.a
);
3044 n
->vqs
[0].tx_waiting
= 0;
3045 n
->tx_burst
= n
->net_conf
.txburst
;
3046 virtio_net_set_mrg_rx_bufs(n
, 0, 0);
3047 n
->promisc
= 1; /* for compatibility */
3049 n
->mac_table
.macs
= g_malloc0(MAC_TABLE_ENTRIES
* ETH_ALEN
);
3051 n
->vlans
= g_malloc0(MAX_VLAN
>> 3);
3053 nc
= qemu_get_queue(n
->nic
);
3054 nc
->rxfilter_notify_enabled
= 1;
3056 QTAILQ_INIT(&n
->rsc_chains
);
3060 static void virtio_net_device_unrealize(DeviceState
*dev
, Error
**errp
)
3062 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
3063 VirtIONet
*n
= VIRTIO_NET(dev
);
3066 /* This will stop vhost backend if appropriate. */
3067 virtio_net_set_status(vdev
, 0);
3069 g_free(n
->netclient_name
);
3070 n
->netclient_name
= NULL
;
3071 g_free(n
->netclient_type
);
3072 n
->netclient_type
= NULL
;
3074 g_free(n
->mac_table
.macs
);
3078 g_free(n
->primary_device_id
);
3079 g_free(n
->standby_id
);
3080 qobject_unref(n
->primary_device_dict
);
3081 n
->primary_device_dict
= NULL
;
3084 max_queues
= n
->multiqueue
? n
->max_queues
: 1;
3085 for (i
= 0; i
< max_queues
; i
++) {
3086 virtio_net_del_queue(n
, i
);
3089 qemu_announce_timer_del(&n
->announce_timer
, false);
3091 qemu_del_nic(n
->nic
);
3092 virtio_net_rsc_cleanup(n
);
3093 virtio_cleanup(vdev
);
3096 static void virtio_net_instance_init(Object
*obj
)
3098 VirtIONet
*n
= VIRTIO_NET(obj
);
3101 * The default config_size is sizeof(struct virtio_net_config).
3102 * Can be overriden with virtio_net_set_config_size.
3104 n
->config_size
= sizeof(struct virtio_net_config
);
3105 device_add_bootindex_property(obj
, &n
->nic_conf
.bootindex
,
3106 "bootindex", "/ethernet-phy@0",
3110 static int virtio_net_pre_save(void *opaque
)
3112 VirtIONet
*n
= opaque
;
3114 /* At this point, backend must be stopped, otherwise
3115 * it might keep writing to memory. */
3116 assert(!n
->vhost_started
);
3121 static bool primary_unplug_pending(void *opaque
)
3123 DeviceState
*dev
= opaque
;
3124 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
3125 VirtIONet
*n
= VIRTIO_NET(vdev
);
3127 return n
->primary_dev
? n
->primary_dev
->pending_deleted_event
: false;
3130 static bool dev_unplug_pending(void *opaque
)
3132 DeviceState
*dev
= opaque
;
3133 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(dev
);
3135 return vdc
->primary_unplug_pending(dev
);
3138 static const VMStateDescription vmstate_virtio_net
= {
3139 .name
= "virtio-net",
3140 .minimum_version_id
= VIRTIO_NET_VM_VERSION
,
3141 .version_id
= VIRTIO_NET_VM_VERSION
,
3142 .fields
= (VMStateField
[]) {
3143 VMSTATE_VIRTIO_DEVICE
,
3144 VMSTATE_END_OF_LIST()
3146 .pre_save
= virtio_net_pre_save
,
3147 .dev_unplug_pending
= dev_unplug_pending
,
3150 static Property virtio_net_properties
[] = {
3151 DEFINE_PROP_BIT64("csum", VirtIONet
, host_features
,
3152 VIRTIO_NET_F_CSUM
, true),
3153 DEFINE_PROP_BIT64("guest_csum", VirtIONet
, host_features
,
3154 VIRTIO_NET_F_GUEST_CSUM
, true),
3155 DEFINE_PROP_BIT64("gso", VirtIONet
, host_features
, VIRTIO_NET_F_GSO
, true),
3156 DEFINE_PROP_BIT64("guest_tso4", VirtIONet
, host_features
,
3157 VIRTIO_NET_F_GUEST_TSO4
, true),
3158 DEFINE_PROP_BIT64("guest_tso6", VirtIONet
, host_features
,
3159 VIRTIO_NET_F_GUEST_TSO6
, true),
3160 DEFINE_PROP_BIT64("guest_ecn", VirtIONet
, host_features
,
3161 VIRTIO_NET_F_GUEST_ECN
, true),
3162 DEFINE_PROP_BIT64("guest_ufo", VirtIONet
, host_features
,
3163 VIRTIO_NET_F_GUEST_UFO
, true),
3164 DEFINE_PROP_BIT64("guest_announce", VirtIONet
, host_features
,
3165 VIRTIO_NET_F_GUEST_ANNOUNCE
, true),
3166 DEFINE_PROP_BIT64("host_tso4", VirtIONet
, host_features
,
3167 VIRTIO_NET_F_HOST_TSO4
, true),
3168 DEFINE_PROP_BIT64("host_tso6", VirtIONet
, host_features
,
3169 VIRTIO_NET_F_HOST_TSO6
, true),
3170 DEFINE_PROP_BIT64("host_ecn", VirtIONet
, host_features
,
3171 VIRTIO_NET_F_HOST_ECN
, true),
3172 DEFINE_PROP_BIT64("host_ufo", VirtIONet
, host_features
,
3173 VIRTIO_NET_F_HOST_UFO
, true),
3174 DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet
, host_features
,
3175 VIRTIO_NET_F_MRG_RXBUF
, true),
3176 DEFINE_PROP_BIT64("status", VirtIONet
, host_features
,
3177 VIRTIO_NET_F_STATUS
, true),
3178 DEFINE_PROP_BIT64("ctrl_vq", VirtIONet
, host_features
,
3179 VIRTIO_NET_F_CTRL_VQ
, true),
3180 DEFINE_PROP_BIT64("ctrl_rx", VirtIONet
, host_features
,
3181 VIRTIO_NET_F_CTRL_RX
, true),
3182 DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet
, host_features
,
3183 VIRTIO_NET_F_CTRL_VLAN
, true),
3184 DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet
, host_features
,
3185 VIRTIO_NET_F_CTRL_RX_EXTRA
, true),
3186 DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet
, host_features
,
3187 VIRTIO_NET_F_CTRL_MAC_ADDR
, true),
3188 DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet
, host_features
,
3189 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
, true),
3190 DEFINE_PROP_BIT64("mq", VirtIONet
, host_features
, VIRTIO_NET_F_MQ
, false),
3191 DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet
, host_features
,
3192 VIRTIO_NET_F_RSC_EXT
, false),
3193 DEFINE_PROP_UINT32("rsc_interval", VirtIONet
, rsc_timeout
,
3194 VIRTIO_NET_RSC_DEFAULT_INTERVAL
),
3195 DEFINE_NIC_PROPERTIES(VirtIONet
, nic_conf
),
3196 DEFINE_PROP_UINT32("x-txtimer", VirtIONet
, net_conf
.txtimer
,
3198 DEFINE_PROP_INT32("x-txburst", VirtIONet
, net_conf
.txburst
, TX_BURST
),
3199 DEFINE_PROP_STRING("tx", VirtIONet
, net_conf
.tx
),
3200 DEFINE_PROP_UINT16("rx_queue_size", VirtIONet
, net_conf
.rx_queue_size
,
3201 VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
),
3202 DEFINE_PROP_UINT16("tx_queue_size", VirtIONet
, net_conf
.tx_queue_size
,
3203 VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
),
3204 DEFINE_PROP_UINT16("host_mtu", VirtIONet
, net_conf
.mtu
, 0),
3205 DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet
, mtu_bypass_backend
,
3207 DEFINE_PROP_INT32("speed", VirtIONet
, net_conf
.speed
, SPEED_UNKNOWN
),
3208 DEFINE_PROP_STRING("duplex", VirtIONet
, net_conf
.duplex_str
),
3209 DEFINE_PROP_BOOL("failover", VirtIONet
, failover
, false),
3210 DEFINE_PROP_END_OF_LIST(),
3213 static void virtio_net_class_init(ObjectClass
*klass
, void *data
)
3215 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3216 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
3218 dc
->props
= virtio_net_properties
;
3219 dc
->vmsd
= &vmstate_virtio_net
;
3220 set_bit(DEVICE_CATEGORY_NETWORK
, dc
->categories
);
3221 vdc
->realize
= virtio_net_device_realize
;
3222 vdc
->unrealize
= virtio_net_device_unrealize
;
3223 vdc
->get_config
= virtio_net_get_config
;
3224 vdc
->set_config
= virtio_net_set_config
;
3225 vdc
->get_features
= virtio_net_get_features
;
3226 vdc
->set_features
= virtio_net_set_features
;
3227 vdc
->bad_features
= virtio_net_bad_features
;
3228 vdc
->reset
= virtio_net_reset
;
3229 vdc
->set_status
= virtio_net_set_status
;
3230 vdc
->guest_notifier_mask
= virtio_net_guest_notifier_mask
;
3231 vdc
->guest_notifier_pending
= virtio_net_guest_notifier_pending
;
3232 vdc
->legacy_features
|= (0x1 << VIRTIO_NET_F_GSO
);
3233 vdc
->post_load
= virtio_net_post_load_virtio
;
3234 vdc
->vmsd
= &vmstate_virtio_net_device
;
3235 vdc
->primary_unplug_pending
= primary_unplug_pending
;
3238 static const TypeInfo virtio_net_info
= {
3239 .name
= TYPE_VIRTIO_NET
,
3240 .parent
= TYPE_VIRTIO_DEVICE
,
3241 .instance_size
= sizeof(VirtIONet
),
3242 .instance_init
= virtio_net_instance_init
,
3243 .class_init
= virtio_net_class_init
,
3246 static void virtio_register_types(void)
3248 type_register_static(&virtio_net_info
);
3251 type_init(virtio_register_types
)