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"
18 #include "qemu/main-loop.h"
19 #include "qemu/module.h"
20 #include "hw/virtio/virtio.h"
22 #include "net/checksum.h"
24 #include "qemu/error-report.h"
25 #include "qemu/timer.h"
26 #include "qemu/option.h"
27 #include "qemu/option_int.h"
28 #include "qemu/config-file.h"
29 #include "qapi/qmp/qdict.h"
30 #include "hw/virtio/virtio-net.h"
31 #include "net/vhost_net.h"
32 #include "net/announce.h"
33 #include "hw/virtio/virtio-bus.h"
34 #include "qapi/error.h"
35 #include "qapi/qapi-events-net.h"
36 #include "hw/qdev-properties.h"
37 #include "qapi/qapi-types-migration.h"
38 #include "qapi/qapi-events-migration.h"
39 #include "hw/virtio/virtio-access.h"
40 #include "migration/misc.h"
41 #include "standard-headers/linux/ethtool.h"
42 #include "sysemu/sysemu.h"
44 #include "monitor/qdev.h"
45 #include "hw/pci/pci.h"
46 #include "net_rx_pkt.h"
47 #include "hw/virtio/vhost.h"
48 #include "sysemu/qtest.h"
50 #define VIRTIO_NET_VM_VERSION 11
52 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */
54 /* previously fixed value */
55 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
56 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
58 /* for now, only allow larger queue_pairs; with virtio-1, guest can downsize */
59 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
60 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
62 #define VIRTIO_NET_IP4_ADDR_SIZE 8 /* ipv4 saddr + daddr */
64 #define VIRTIO_NET_TCP_FLAG 0x3F
65 #define VIRTIO_NET_TCP_HDR_LENGTH 0xF000
67 /* IPv4 max payload, 16 bits in the header */
68 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
69 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
71 /* header length value in ip header without option */
72 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
74 #define VIRTIO_NET_IP6_ADDR_SIZE 32 /* ipv6 saddr + daddr */
75 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
77 /* Purge coalesced packets timer interval, This value affects the performance
78 a lot, and should be tuned carefully, '300000'(300us) is the recommended
79 value to pass the WHQL test, '50000' can gain 2x netperf throughput with
81 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
83 #define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \
84 VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \
85 VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \
86 VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \
87 VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \
88 VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \
89 VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \
90 VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \
91 VIRTIO_NET_RSS_HASH_TYPE_UDP_EX)
93 static const VirtIOFeature feature_sizes
[] = {
94 {.flags
= 1ULL << VIRTIO_NET_F_MAC
,
95 .end
= endof(struct virtio_net_config
, mac
)},
96 {.flags
= 1ULL << VIRTIO_NET_F_STATUS
,
97 .end
= endof(struct virtio_net_config
, status
)},
98 {.flags
= 1ULL << VIRTIO_NET_F_MQ
,
99 .end
= endof(struct virtio_net_config
, max_virtqueue_pairs
)},
100 {.flags
= 1ULL << VIRTIO_NET_F_MTU
,
101 .end
= endof(struct virtio_net_config
, mtu
)},
102 {.flags
= 1ULL << VIRTIO_NET_F_SPEED_DUPLEX
,
103 .end
= endof(struct virtio_net_config
, duplex
)},
104 {.flags
= (1ULL << VIRTIO_NET_F_RSS
) | (1ULL << VIRTIO_NET_F_HASH_REPORT
),
105 .end
= endof(struct virtio_net_config
, supported_hash_types
)},
109 static const VirtIOConfigSizeParams cfg_size_params
= {
110 .min_size
= endof(struct virtio_net_config
, mac
),
111 .max_size
= sizeof(struct virtio_net_config
),
112 .feature_sizes
= feature_sizes
115 static VirtIONetQueue
*virtio_net_get_subqueue(NetClientState
*nc
)
117 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
119 return &n
->vqs
[nc
->queue_index
];
122 static int vq2q(int queue_index
)
124 return queue_index
/ 2;
127 static void flush_or_purge_queued_packets(NetClientState
*nc
)
133 qemu_flush_or_purge_queued_packets(nc
->peer
, true);
134 assert(!virtio_net_get_subqueue(nc
)->async_tx
.elem
);
138 * - we could suppress RX interrupt if we were so inclined.
141 static void virtio_net_get_config(VirtIODevice
*vdev
, uint8_t *config
)
143 VirtIONet
*n
= VIRTIO_NET(vdev
);
144 struct virtio_net_config netcfg
;
145 NetClientState
*nc
= qemu_get_queue(n
->nic
);
146 static const MACAddr zero
= { .a
= { 0, 0, 0, 0, 0, 0 } };
149 memset(&netcfg
, 0 , sizeof(struct virtio_net_config
));
150 virtio_stw_p(vdev
, &netcfg
.status
, n
->status
);
151 virtio_stw_p(vdev
, &netcfg
.max_virtqueue_pairs
, n
->max_queue_pairs
);
152 virtio_stw_p(vdev
, &netcfg
.mtu
, n
->net_conf
.mtu
);
153 memcpy(netcfg
.mac
, n
->mac
, ETH_ALEN
);
154 virtio_stl_p(vdev
, &netcfg
.speed
, n
->net_conf
.speed
);
155 netcfg
.duplex
= n
->net_conf
.duplex
;
156 netcfg
.rss_max_key_size
= VIRTIO_NET_RSS_MAX_KEY_SIZE
;
157 virtio_stw_p(vdev
, &netcfg
.rss_max_indirection_table_length
,
158 virtio_host_has_feature(vdev
, VIRTIO_NET_F_RSS
) ?
159 VIRTIO_NET_RSS_MAX_TABLE_LEN
: 1);
160 virtio_stl_p(vdev
, &netcfg
.supported_hash_types
,
161 VIRTIO_NET_RSS_SUPPORTED_HASHES
);
162 memcpy(config
, &netcfg
, n
->config_size
);
165 * Is this VDPA? No peer means not VDPA: there's no way to
166 * disconnect/reconnect a VDPA peer.
168 if (nc
->peer
&& nc
->peer
->info
->type
== NET_CLIENT_DRIVER_VHOST_VDPA
) {
169 ret
= vhost_net_get_config(get_vhost_net(nc
->peer
), (uint8_t *)&netcfg
,
176 * Some NIC/kernel combinations present 0 as the mac address. As that
177 * is not a legal address, try to proceed with the address from the
178 * QEMU command line in the hope that the address has been configured
179 * correctly elsewhere - just not reported by the device.
181 if (memcmp(&netcfg
.mac
, &zero
, sizeof(zero
)) == 0) {
182 info_report("Zero hardware mac address detected. Ignoring.");
183 memcpy(netcfg
.mac
, n
->mac
, ETH_ALEN
);
186 netcfg
.status
|= virtio_tswap16(vdev
,
187 n
->status
& VIRTIO_NET_S_ANNOUNCE
);
188 memcpy(config
, &netcfg
, n
->config_size
);
192 static void virtio_net_set_config(VirtIODevice
*vdev
, const uint8_t *config
)
194 VirtIONet
*n
= VIRTIO_NET(vdev
);
195 struct virtio_net_config netcfg
= {};
196 NetClientState
*nc
= qemu_get_queue(n
->nic
);
198 memcpy(&netcfg
, config
, n
->config_size
);
200 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
) &&
201 !virtio_vdev_has_feature(vdev
, VIRTIO_F_VERSION_1
) &&
202 memcmp(netcfg
.mac
, n
->mac
, ETH_ALEN
)) {
203 memcpy(n
->mac
, netcfg
.mac
, ETH_ALEN
);
204 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->mac
);
208 * Is this VDPA? No peer means not VDPA: there's no way to
209 * disconnect/reconnect a VDPA peer.
211 if (nc
->peer
&& nc
->peer
->info
->type
== NET_CLIENT_DRIVER_VHOST_VDPA
) {
212 vhost_net_set_config(get_vhost_net(nc
->peer
),
213 (uint8_t *)&netcfg
, 0, n
->config_size
,
214 VHOST_SET_CONFIG_TYPE_MASTER
);
218 static bool virtio_net_started(VirtIONet
*n
, uint8_t status
)
220 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
221 return (status
& VIRTIO_CONFIG_S_DRIVER_OK
) &&
222 (n
->status
& VIRTIO_NET_S_LINK_UP
) && vdev
->vm_running
;
225 static void virtio_net_announce_notify(VirtIONet
*net
)
227 VirtIODevice
*vdev
= VIRTIO_DEVICE(net
);
228 trace_virtio_net_announce_notify();
230 net
->status
|= VIRTIO_NET_S_ANNOUNCE
;
231 virtio_notify_config(vdev
);
234 static void virtio_net_announce_timer(void *opaque
)
236 VirtIONet
*n
= opaque
;
237 trace_virtio_net_announce_timer(n
->announce_timer
.round
);
239 n
->announce_timer
.round
--;
240 virtio_net_announce_notify(n
);
243 static void virtio_net_announce(NetClientState
*nc
)
245 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
246 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
249 * Make sure the virtio migration announcement timer isn't running
250 * If it is, let it trigger announcement so that we do not cause
253 if (n
->announce_timer
.round
) {
257 if (virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
) &&
258 virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
)) {
259 virtio_net_announce_notify(n
);
263 static void virtio_net_vhost_status(VirtIONet
*n
, uint8_t status
)
265 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
266 NetClientState
*nc
= qemu_get_queue(n
->nic
);
267 int queue_pairs
= n
->multiqueue
? n
->max_queue_pairs
: 1;
268 int cvq
= virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
) ?
269 n
->max_ncs
- n
->max_queue_pairs
: 0;
271 if (!get_vhost_net(nc
->peer
)) {
275 if ((virtio_net_started(n
, status
) && !nc
->peer
->link_down
) ==
276 !!n
->vhost_started
) {
279 if (!n
->vhost_started
) {
282 if (n
->needs_vnet_hdr_swap
) {
283 error_report("backend does not support %s vnet headers; "
284 "falling back on userspace virtio",
285 virtio_is_big_endian(vdev
) ? "BE" : "LE");
289 /* Any packets outstanding? Purge them to avoid touching rings
290 * when vhost is running.
292 for (i
= 0; i
< queue_pairs
; i
++) {
293 NetClientState
*qnc
= qemu_get_subqueue(n
->nic
, i
);
295 /* Purge both directions: TX and RX. */
296 qemu_net_queue_purge(qnc
->peer
->incoming_queue
, qnc
);
297 qemu_net_queue_purge(qnc
->incoming_queue
, qnc
->peer
);
300 if (virtio_has_feature(vdev
->guest_features
, VIRTIO_NET_F_MTU
)) {
301 r
= vhost_net_set_mtu(get_vhost_net(nc
->peer
), n
->net_conf
.mtu
);
303 error_report("%uBytes MTU not supported by the backend",
310 n
->vhost_started
= 1;
311 r
= vhost_net_start(vdev
, n
->nic
->ncs
, queue_pairs
, cvq
);
313 error_report("unable to start vhost net: %d: "
314 "falling back on userspace virtio", -r
);
315 n
->vhost_started
= 0;
318 vhost_net_stop(vdev
, n
->nic
->ncs
, queue_pairs
, cvq
);
319 n
->vhost_started
= 0;
323 static int virtio_net_set_vnet_endian_one(VirtIODevice
*vdev
,
324 NetClientState
*peer
,
327 if (virtio_is_big_endian(vdev
)) {
328 return qemu_set_vnet_be(peer
, enable
);
330 return qemu_set_vnet_le(peer
, enable
);
334 static bool virtio_net_set_vnet_endian(VirtIODevice
*vdev
, NetClientState
*ncs
,
335 int queue_pairs
, bool enable
)
339 for (i
= 0; i
< queue_pairs
; i
++) {
340 if (virtio_net_set_vnet_endian_one(vdev
, ncs
[i
].peer
, enable
) < 0 &&
343 virtio_net_set_vnet_endian_one(vdev
, ncs
[i
].peer
, false);
353 static void virtio_net_vnet_endian_status(VirtIONet
*n
, uint8_t status
)
355 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
356 int queue_pairs
= n
->multiqueue
? n
->max_queue_pairs
: 1;
358 if (virtio_net_started(n
, status
)) {
359 /* Before using the device, we tell the network backend about the
360 * endianness to use when parsing vnet headers. If the backend
361 * can't do it, we fallback onto fixing the headers in the core
364 n
->needs_vnet_hdr_swap
= virtio_net_set_vnet_endian(vdev
, n
->nic
->ncs
,
366 } else if (virtio_net_started(n
, vdev
->status
)) {
367 /* After using the device, we need to reset the network backend to
368 * the default (guest native endianness), otherwise the guest may
369 * lose network connectivity if it is rebooted into a different
372 virtio_net_set_vnet_endian(vdev
, n
->nic
->ncs
, queue_pairs
, false);
376 static void virtio_net_drop_tx_queue_data(VirtIODevice
*vdev
, VirtQueue
*vq
)
378 unsigned int dropped
= virtqueue_drop_all(vq
);
380 virtio_notify(vdev
, vq
);
384 static void virtio_net_set_status(struct VirtIODevice
*vdev
, uint8_t status
)
386 VirtIONet
*n
= VIRTIO_NET(vdev
);
389 uint8_t queue_status
;
391 virtio_net_vnet_endian_status(n
, status
);
392 virtio_net_vhost_status(n
, status
);
394 for (i
= 0; i
< n
->max_queue_pairs
; i
++) {
395 NetClientState
*ncs
= qemu_get_subqueue(n
->nic
, i
);
399 if ((!n
->multiqueue
&& i
!= 0) || i
>= n
->curr_queue_pairs
) {
402 queue_status
= status
;
405 virtio_net_started(n
, queue_status
) && !n
->vhost_started
;
408 qemu_flush_queued_packets(ncs
);
411 if (!q
->tx_waiting
) {
417 timer_mod(q
->tx_timer
,
418 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) + n
->tx_timeout
);
420 qemu_bh_schedule(q
->tx_bh
);
424 timer_del(q
->tx_timer
);
426 qemu_bh_cancel(q
->tx_bh
);
428 if ((n
->status
& VIRTIO_NET_S_LINK_UP
) == 0 &&
429 (queue_status
& VIRTIO_CONFIG_S_DRIVER_OK
) &&
431 /* if tx is waiting we are likely have some packets in tx queue
432 * and disabled notification */
434 virtio_queue_set_notification(q
->tx_vq
, 1);
435 virtio_net_drop_tx_queue_data(vdev
, q
->tx_vq
);
441 static void virtio_net_set_link_status(NetClientState
*nc
)
443 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
444 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
445 uint16_t old_status
= n
->status
;
448 n
->status
&= ~VIRTIO_NET_S_LINK_UP
;
450 n
->status
|= VIRTIO_NET_S_LINK_UP
;
452 if (n
->status
!= old_status
)
453 virtio_notify_config(vdev
);
455 virtio_net_set_status(vdev
, vdev
->status
);
458 static void rxfilter_notify(NetClientState
*nc
)
460 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
462 if (nc
->rxfilter_notify_enabled
) {
463 char *path
= object_get_canonical_path(OBJECT(n
->qdev
));
464 qapi_event_send_nic_rx_filter_changed(n
->netclient_name
, path
);
467 /* disable event notification to avoid events flooding */
468 nc
->rxfilter_notify_enabled
= 0;
472 static intList
*get_vlan_table(VirtIONet
*n
)
478 for (i
= 0; i
< MAX_VLAN
>> 5; i
++) {
479 for (j
= 0; n
->vlans
[i
] && j
<= 0x1f; j
++) {
480 if (n
->vlans
[i
] & (1U << j
)) {
481 QAPI_LIST_PREPEND(list
, (i
<< 5) + j
);
489 static RxFilterInfo
*virtio_net_query_rxfilter(NetClientState
*nc
)
491 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
492 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
497 info
= g_malloc0(sizeof(*info
));
498 info
->name
= g_strdup(nc
->name
);
499 info
->promiscuous
= n
->promisc
;
502 info
->unicast
= RX_STATE_NONE
;
503 } else if (n
->alluni
) {
504 info
->unicast
= RX_STATE_ALL
;
506 info
->unicast
= RX_STATE_NORMAL
;
510 info
->multicast
= RX_STATE_NONE
;
511 } else if (n
->allmulti
) {
512 info
->multicast
= RX_STATE_ALL
;
514 info
->multicast
= RX_STATE_NORMAL
;
517 info
->broadcast_allowed
= n
->nobcast
;
518 info
->multicast_overflow
= n
->mac_table
.multi_overflow
;
519 info
->unicast_overflow
= n
->mac_table
.uni_overflow
;
521 info
->main_mac
= qemu_mac_strdup_printf(n
->mac
);
524 for (i
= 0; i
< n
->mac_table
.first_multi
; i
++) {
525 QAPI_LIST_PREPEND(str_list
,
526 qemu_mac_strdup_printf(n
->mac_table
.macs
+ i
* ETH_ALEN
));
528 info
->unicast_table
= str_list
;
531 for (i
= n
->mac_table
.first_multi
; i
< n
->mac_table
.in_use
; i
++) {
532 QAPI_LIST_PREPEND(str_list
,
533 qemu_mac_strdup_printf(n
->mac_table
.macs
+ i
* ETH_ALEN
));
535 info
->multicast_table
= str_list
;
536 info
->vlan_table
= get_vlan_table(n
);
538 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_VLAN
)) {
539 info
->vlan
= RX_STATE_ALL
;
540 } else if (!info
->vlan_table
) {
541 info
->vlan
= RX_STATE_NONE
;
543 info
->vlan
= RX_STATE_NORMAL
;
546 /* enable event notification after query */
547 nc
->rxfilter_notify_enabled
= 1;
552 static void virtio_net_queue_reset(VirtIODevice
*vdev
, uint32_t queue_index
)
554 VirtIONet
*n
= VIRTIO_NET(vdev
);
557 /* validate queue_index and skip for cvq */
558 if (queue_index
>= n
->max_queue_pairs
* 2) {
562 nc
= qemu_get_subqueue(n
->nic
, vq2q(queue_index
));
568 if (get_vhost_net(nc
->peer
) &&
569 nc
->peer
->info
->type
== NET_CLIENT_DRIVER_TAP
) {
570 vhost_net_virtqueue_reset(vdev
, nc
, queue_index
);
573 flush_or_purge_queued_packets(nc
);
576 static void virtio_net_queue_enable(VirtIODevice
*vdev
, uint32_t queue_index
)
578 VirtIONet
*n
= VIRTIO_NET(vdev
);
582 /* validate queue_index and skip for cvq */
583 if (queue_index
>= n
->max_queue_pairs
* 2) {
587 nc
= qemu_get_subqueue(n
->nic
, vq2q(queue_index
));
589 if (!nc
->peer
|| !vdev
->vhost_started
) {
593 if (get_vhost_net(nc
->peer
) &&
594 nc
->peer
->info
->type
== NET_CLIENT_DRIVER_TAP
) {
595 r
= vhost_net_virtqueue_restart(vdev
, nc
, queue_index
);
597 error_report("unable to restart vhost net virtqueue: %d, "
598 "when resetting the queue", queue_index
);
603 static void virtio_net_reset(VirtIODevice
*vdev
)
605 VirtIONet
*n
= VIRTIO_NET(vdev
);
608 /* Reset back to compatibility mode */
615 /* multiqueue is disabled by default */
616 n
->curr_queue_pairs
= 1;
617 timer_del(n
->announce_timer
.tm
);
618 n
->announce_timer
.round
= 0;
619 n
->status
&= ~VIRTIO_NET_S_ANNOUNCE
;
621 /* Flush any MAC and VLAN filter table state */
622 n
->mac_table
.in_use
= 0;
623 n
->mac_table
.first_multi
= 0;
624 n
->mac_table
.multi_overflow
= 0;
625 n
->mac_table
.uni_overflow
= 0;
626 memset(n
->mac_table
.macs
, 0, MAC_TABLE_ENTRIES
* ETH_ALEN
);
627 memcpy(&n
->mac
[0], &n
->nic
->conf
->macaddr
, sizeof(n
->mac
));
628 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->mac
);
629 memset(n
->vlans
, 0, MAX_VLAN
>> 3);
631 /* Flush any async TX */
632 for (i
= 0; i
< n
->max_queue_pairs
; i
++) {
633 flush_or_purge_queued_packets(qemu_get_subqueue(n
->nic
, i
));
637 static void peer_test_vnet_hdr(VirtIONet
*n
)
639 NetClientState
*nc
= qemu_get_queue(n
->nic
);
644 n
->has_vnet_hdr
= qemu_has_vnet_hdr(nc
->peer
);
647 static int peer_has_vnet_hdr(VirtIONet
*n
)
649 return n
->has_vnet_hdr
;
652 static int peer_has_ufo(VirtIONet
*n
)
654 if (!peer_has_vnet_hdr(n
))
657 n
->has_ufo
= qemu_has_ufo(qemu_get_queue(n
->nic
)->peer
);
662 static void virtio_net_set_mrg_rx_bufs(VirtIONet
*n
, int mergeable_rx_bufs
,
663 int version_1
, int hash_report
)
668 n
->mergeable_rx_bufs
= mergeable_rx_bufs
;
671 n
->guest_hdr_len
= hash_report
?
672 sizeof(struct virtio_net_hdr_v1_hash
) :
673 sizeof(struct virtio_net_hdr_mrg_rxbuf
);
674 n
->rss_data
.populate_hash
= !!hash_report
;
676 n
->guest_hdr_len
= n
->mergeable_rx_bufs
?
677 sizeof(struct virtio_net_hdr_mrg_rxbuf
) :
678 sizeof(struct virtio_net_hdr
);
681 for (i
= 0; i
< n
->max_queue_pairs
; i
++) {
682 nc
= qemu_get_subqueue(n
->nic
, i
);
684 if (peer_has_vnet_hdr(n
) &&
685 qemu_has_vnet_hdr_len(nc
->peer
, n
->guest_hdr_len
)) {
686 qemu_set_vnet_hdr_len(nc
->peer
, n
->guest_hdr_len
);
687 n
->host_hdr_len
= n
->guest_hdr_len
;
692 static int virtio_net_max_tx_queue_size(VirtIONet
*n
)
694 NetClientState
*peer
= n
->nic_conf
.peers
.ncs
[0];
697 * Backends other than vhost-user or vhost-vdpa don't support max queue
701 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
;
704 switch(peer
->info
->type
) {
705 case NET_CLIENT_DRIVER_VHOST_USER
:
706 case NET_CLIENT_DRIVER_VHOST_VDPA
:
707 return VIRTQUEUE_MAX_SIZE
;
709 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
;
713 static int peer_attach(VirtIONet
*n
, int index
)
715 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, index
);
721 if (nc
->peer
->info
->type
== NET_CLIENT_DRIVER_VHOST_USER
) {
722 vhost_set_vring_enable(nc
->peer
, 1);
725 if (nc
->peer
->info
->type
!= NET_CLIENT_DRIVER_TAP
) {
729 if (n
->max_queue_pairs
== 1) {
733 return tap_enable(nc
->peer
);
736 static int peer_detach(VirtIONet
*n
, int index
)
738 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, index
);
744 if (nc
->peer
->info
->type
== NET_CLIENT_DRIVER_VHOST_USER
) {
745 vhost_set_vring_enable(nc
->peer
, 0);
748 if (nc
->peer
->info
->type
!= NET_CLIENT_DRIVER_TAP
) {
752 return tap_disable(nc
->peer
);
755 static void virtio_net_set_queue_pairs(VirtIONet
*n
)
760 if (n
->nic
->peer_deleted
) {
764 for (i
= 0; i
< n
->max_queue_pairs
; i
++) {
765 if (i
< n
->curr_queue_pairs
) {
766 r
= peer_attach(n
, i
);
769 r
= peer_detach(n
, i
);
775 static void virtio_net_set_multiqueue(VirtIONet
*n
, int multiqueue
);
777 static uint64_t virtio_net_get_features(VirtIODevice
*vdev
, uint64_t features
,
780 VirtIONet
*n
= VIRTIO_NET(vdev
);
781 NetClientState
*nc
= qemu_get_queue(n
->nic
);
783 /* Firstly sync all virtio-net possible supported features */
784 features
|= n
->host_features
;
786 virtio_add_feature(&features
, VIRTIO_NET_F_MAC
);
788 if (!peer_has_vnet_hdr(n
)) {
789 virtio_clear_feature(&features
, VIRTIO_NET_F_CSUM
);
790 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_TSO4
);
791 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_TSO6
);
792 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_ECN
);
794 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_CSUM
);
795 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_TSO4
);
796 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_TSO6
);
797 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_ECN
);
799 virtio_clear_feature(&features
, VIRTIO_NET_F_HASH_REPORT
);
802 if (!peer_has_vnet_hdr(n
) || !peer_has_ufo(n
)) {
803 virtio_clear_feature(&features
, VIRTIO_NET_F_GUEST_UFO
);
804 virtio_clear_feature(&features
, VIRTIO_NET_F_HOST_UFO
);
807 if (!get_vhost_net(nc
->peer
)) {
808 virtio_add_feature(&features
, VIRTIO_F_RING_RESET
);
812 if (!ebpf_rss_is_loaded(&n
->ebpf_rss
)) {
813 virtio_clear_feature(&features
, VIRTIO_NET_F_RSS
);
815 features
= vhost_net_get_features(get_vhost_net(nc
->peer
), features
);
816 vdev
->backend_features
= features
;
818 if (n
->mtu_bypass_backend
&&
819 (n
->host_features
& 1ULL << VIRTIO_NET_F_MTU
)) {
820 features
|= (1ULL << VIRTIO_NET_F_MTU
);
826 static uint64_t virtio_net_bad_features(VirtIODevice
*vdev
)
828 uint64_t features
= 0;
830 /* Linux kernel 2.6.25. It understood MAC (as everyone must),
832 virtio_add_feature(&features
, VIRTIO_NET_F_MAC
);
833 virtio_add_feature(&features
, VIRTIO_NET_F_CSUM
);
834 virtio_add_feature(&features
, VIRTIO_NET_F_HOST_TSO4
);
835 virtio_add_feature(&features
, VIRTIO_NET_F_HOST_TSO6
);
836 virtio_add_feature(&features
, VIRTIO_NET_F_HOST_ECN
);
841 static void virtio_net_apply_guest_offloads(VirtIONet
*n
)
843 qemu_set_offload(qemu_get_queue(n
->nic
)->peer
,
844 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_CSUM
)),
845 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_TSO4
)),
846 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_TSO6
)),
847 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_ECN
)),
848 !!(n
->curr_guest_offloads
& (1ULL << VIRTIO_NET_F_GUEST_UFO
)));
851 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features
)
853 static const uint64_t guest_offloads_mask
=
854 (1ULL << VIRTIO_NET_F_GUEST_CSUM
) |
855 (1ULL << VIRTIO_NET_F_GUEST_TSO4
) |
856 (1ULL << VIRTIO_NET_F_GUEST_TSO6
) |
857 (1ULL << VIRTIO_NET_F_GUEST_ECN
) |
858 (1ULL << VIRTIO_NET_F_GUEST_UFO
);
860 return guest_offloads_mask
& features
;
863 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet
*n
)
865 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
866 return virtio_net_guest_offloads_by_features(vdev
->guest_features
);
875 * Set the failover primary device
877 * @opaque: FailoverId to setup
878 * @opts: opts for device we are handling
879 * @errp: returns an error if this function fails
881 static int failover_set_primary(DeviceState
*dev
, void *opaque
)
883 FailoverDevice
*fdev
= opaque
;
884 PCIDevice
*pci_dev
= (PCIDevice
*)
885 object_dynamic_cast(OBJECT(dev
), TYPE_PCI_DEVICE
);
891 if (!g_strcmp0(pci_dev
->failover_pair_id
, fdev
->n
->netclient_name
)) {
900 * Find the primary device for this failover virtio-net
902 * @n: VirtIONet device
903 * @errp: returns an error if this function fails
905 static DeviceState
*failover_find_primary_device(VirtIONet
*n
)
907 FailoverDevice fdev
= {
911 qbus_walk_children(sysbus_get_default(), failover_set_primary
, NULL
,
916 static void failover_add_primary(VirtIONet
*n
, Error
**errp
)
919 DeviceState
*dev
= failover_find_primary_device(n
);
925 if (!n
->primary_opts
) {
926 error_setg(errp
, "Primary device not found");
927 error_append_hint(errp
, "Virtio-net failover will not work. Make "
928 "sure primary device has parameter"
929 " failover_pair_id=%s\n", n
->netclient_name
);
933 dev
= qdev_device_add_from_qdict(n
->primary_opts
,
934 n
->primary_opts_from_json
,
937 qobject_unref(n
->primary_opts
);
938 n
->primary_opts
= NULL
;
940 object_unref(OBJECT(dev
));
942 error_propagate(errp
, err
);
945 static void virtio_net_set_features(VirtIODevice
*vdev
, uint64_t features
)
947 VirtIONet
*n
= VIRTIO_NET(vdev
);
951 if (n
->mtu_bypass_backend
&&
952 !virtio_has_feature(vdev
->backend_features
, VIRTIO_NET_F_MTU
)) {
953 features
&= ~(1ULL << VIRTIO_NET_F_MTU
);
956 virtio_net_set_multiqueue(n
,
957 virtio_has_feature(features
, VIRTIO_NET_F_RSS
) ||
958 virtio_has_feature(features
, VIRTIO_NET_F_MQ
));
960 virtio_net_set_mrg_rx_bufs(n
,
961 virtio_has_feature(features
,
962 VIRTIO_NET_F_MRG_RXBUF
),
963 virtio_has_feature(features
,
965 virtio_has_feature(features
,
966 VIRTIO_NET_F_HASH_REPORT
));
968 n
->rsc4_enabled
= virtio_has_feature(features
, VIRTIO_NET_F_RSC_EXT
) &&
969 virtio_has_feature(features
, VIRTIO_NET_F_GUEST_TSO4
);
970 n
->rsc6_enabled
= virtio_has_feature(features
, VIRTIO_NET_F_RSC_EXT
) &&
971 virtio_has_feature(features
, VIRTIO_NET_F_GUEST_TSO6
);
972 n
->rss_data
.redirect
= virtio_has_feature(features
, VIRTIO_NET_F_RSS
);
974 if (n
->has_vnet_hdr
) {
975 n
->curr_guest_offloads
=
976 virtio_net_guest_offloads_by_features(features
);
977 virtio_net_apply_guest_offloads(n
);
980 for (i
= 0; i
< n
->max_queue_pairs
; i
++) {
981 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, i
);
983 if (!get_vhost_net(nc
->peer
)) {
986 vhost_net_ack_features(get_vhost_net(nc
->peer
), features
);
989 * keep acked_features in NetVhostUserState up-to-date so it
990 * can't miss any features configured by guest virtio driver.
992 vhost_net_save_acked_features(nc
->peer
);
995 if (virtio_has_feature(features
, VIRTIO_NET_F_CTRL_VLAN
)) {
996 memset(n
->vlans
, 0, MAX_VLAN
>> 3);
998 memset(n
->vlans
, 0xff, MAX_VLAN
>> 3);
1001 if (virtio_has_feature(features
, VIRTIO_NET_F_STANDBY
)) {
1002 qapi_event_send_failover_negotiated(n
->netclient_name
);
1003 qatomic_set(&n
->failover_primary_hidden
, false);
1004 failover_add_primary(n
, &err
);
1006 if (!qtest_enabled()) {
1007 warn_report_err(err
);
1015 static int virtio_net_handle_rx_mode(VirtIONet
*n
, uint8_t cmd
,
1016 struct iovec
*iov
, unsigned int iov_cnt
)
1020 NetClientState
*nc
= qemu_get_queue(n
->nic
);
1022 s
= iov_to_buf(iov
, iov_cnt
, 0, &on
, sizeof(on
));
1023 if (s
!= sizeof(on
)) {
1024 return VIRTIO_NET_ERR
;
1027 if (cmd
== VIRTIO_NET_CTRL_RX_PROMISC
) {
1029 } else if (cmd
== VIRTIO_NET_CTRL_RX_ALLMULTI
) {
1031 } else if (cmd
== VIRTIO_NET_CTRL_RX_ALLUNI
) {
1033 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOMULTI
) {
1035 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOUNI
) {
1037 } else if (cmd
== VIRTIO_NET_CTRL_RX_NOBCAST
) {
1040 return VIRTIO_NET_ERR
;
1043 rxfilter_notify(nc
);
1045 return VIRTIO_NET_OK
;
1048 static int virtio_net_handle_offloads(VirtIONet
*n
, uint8_t cmd
,
1049 struct iovec
*iov
, unsigned int iov_cnt
)
1051 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1055 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
)) {
1056 return VIRTIO_NET_ERR
;
1059 s
= iov_to_buf(iov
, iov_cnt
, 0, &offloads
, sizeof(offloads
));
1060 if (s
!= sizeof(offloads
)) {
1061 return VIRTIO_NET_ERR
;
1064 if (cmd
== VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET
) {
1065 uint64_t supported_offloads
;
1067 offloads
= virtio_ldq_p(vdev
, &offloads
);
1069 if (!n
->has_vnet_hdr
) {
1070 return VIRTIO_NET_ERR
;
1073 n
->rsc4_enabled
= virtio_has_feature(offloads
, VIRTIO_NET_F_RSC_EXT
) &&
1074 virtio_has_feature(offloads
, VIRTIO_NET_F_GUEST_TSO4
);
1075 n
->rsc6_enabled
= virtio_has_feature(offloads
, VIRTIO_NET_F_RSC_EXT
) &&
1076 virtio_has_feature(offloads
, VIRTIO_NET_F_GUEST_TSO6
);
1077 virtio_clear_feature(&offloads
, VIRTIO_NET_F_RSC_EXT
);
1079 supported_offloads
= virtio_net_supported_guest_offloads(n
);
1080 if (offloads
& ~supported_offloads
) {
1081 return VIRTIO_NET_ERR
;
1084 n
->curr_guest_offloads
= offloads
;
1085 virtio_net_apply_guest_offloads(n
);
1087 return VIRTIO_NET_OK
;
1089 return VIRTIO_NET_ERR
;
1093 static int virtio_net_handle_mac(VirtIONet
*n
, uint8_t cmd
,
1094 struct iovec
*iov
, unsigned int iov_cnt
)
1096 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1097 struct virtio_net_ctrl_mac mac_data
;
1099 NetClientState
*nc
= qemu_get_queue(n
->nic
);
1101 if (cmd
== VIRTIO_NET_CTRL_MAC_ADDR_SET
) {
1102 if (iov_size(iov
, iov_cnt
) != sizeof(n
->mac
)) {
1103 return VIRTIO_NET_ERR
;
1105 s
= iov_to_buf(iov
, iov_cnt
, 0, &n
->mac
, sizeof(n
->mac
));
1106 assert(s
== sizeof(n
->mac
));
1107 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->mac
);
1108 rxfilter_notify(nc
);
1110 return VIRTIO_NET_OK
;
1113 if (cmd
!= VIRTIO_NET_CTRL_MAC_TABLE_SET
) {
1114 return VIRTIO_NET_ERR
;
1118 int first_multi
= 0;
1119 uint8_t uni_overflow
= 0;
1120 uint8_t multi_overflow
= 0;
1121 uint8_t *macs
= g_malloc0(MAC_TABLE_ENTRIES
* ETH_ALEN
);
1123 s
= iov_to_buf(iov
, iov_cnt
, 0, &mac_data
.entries
,
1124 sizeof(mac_data
.entries
));
1125 mac_data
.entries
= virtio_ldl_p(vdev
, &mac_data
.entries
);
1126 if (s
!= sizeof(mac_data
.entries
)) {
1129 iov_discard_front(&iov
, &iov_cnt
, s
);
1131 if (mac_data
.entries
* ETH_ALEN
> iov_size(iov
, iov_cnt
)) {
1135 if (mac_data
.entries
<= MAC_TABLE_ENTRIES
) {
1136 s
= iov_to_buf(iov
, iov_cnt
, 0, macs
,
1137 mac_data
.entries
* ETH_ALEN
);
1138 if (s
!= mac_data
.entries
* ETH_ALEN
) {
1141 in_use
+= mac_data
.entries
;
1146 iov_discard_front(&iov
, &iov_cnt
, mac_data
.entries
* ETH_ALEN
);
1148 first_multi
= in_use
;
1150 s
= iov_to_buf(iov
, iov_cnt
, 0, &mac_data
.entries
,
1151 sizeof(mac_data
.entries
));
1152 mac_data
.entries
= virtio_ldl_p(vdev
, &mac_data
.entries
);
1153 if (s
!= sizeof(mac_data
.entries
)) {
1157 iov_discard_front(&iov
, &iov_cnt
, s
);
1159 if (mac_data
.entries
* ETH_ALEN
!= iov_size(iov
, iov_cnt
)) {
1163 if (mac_data
.entries
<= MAC_TABLE_ENTRIES
- in_use
) {
1164 s
= iov_to_buf(iov
, iov_cnt
, 0, &macs
[in_use
* ETH_ALEN
],
1165 mac_data
.entries
* ETH_ALEN
);
1166 if (s
!= mac_data
.entries
* ETH_ALEN
) {
1169 in_use
+= mac_data
.entries
;
1174 n
->mac_table
.in_use
= in_use
;
1175 n
->mac_table
.first_multi
= first_multi
;
1176 n
->mac_table
.uni_overflow
= uni_overflow
;
1177 n
->mac_table
.multi_overflow
= multi_overflow
;
1178 memcpy(n
->mac_table
.macs
, macs
, MAC_TABLE_ENTRIES
* ETH_ALEN
);
1180 rxfilter_notify(nc
);
1182 return VIRTIO_NET_OK
;
1186 return VIRTIO_NET_ERR
;
1189 static int virtio_net_handle_vlan_table(VirtIONet
*n
, uint8_t cmd
,
1190 struct iovec
*iov
, unsigned int iov_cnt
)
1192 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1195 NetClientState
*nc
= qemu_get_queue(n
->nic
);
1197 s
= iov_to_buf(iov
, iov_cnt
, 0, &vid
, sizeof(vid
));
1198 vid
= virtio_lduw_p(vdev
, &vid
);
1199 if (s
!= sizeof(vid
)) {
1200 return VIRTIO_NET_ERR
;
1203 if (vid
>= MAX_VLAN
)
1204 return VIRTIO_NET_ERR
;
1206 if (cmd
== VIRTIO_NET_CTRL_VLAN_ADD
)
1207 n
->vlans
[vid
>> 5] |= (1U << (vid
& 0x1f));
1208 else if (cmd
== VIRTIO_NET_CTRL_VLAN_DEL
)
1209 n
->vlans
[vid
>> 5] &= ~(1U << (vid
& 0x1f));
1211 return VIRTIO_NET_ERR
;
1213 rxfilter_notify(nc
);
1215 return VIRTIO_NET_OK
;
1218 static int virtio_net_handle_announce(VirtIONet
*n
, uint8_t cmd
,
1219 struct iovec
*iov
, unsigned int iov_cnt
)
1221 trace_virtio_net_handle_announce(n
->announce_timer
.round
);
1222 if (cmd
== VIRTIO_NET_CTRL_ANNOUNCE_ACK
&&
1223 n
->status
& VIRTIO_NET_S_ANNOUNCE
) {
1224 n
->status
&= ~VIRTIO_NET_S_ANNOUNCE
;
1225 if (n
->announce_timer
.round
) {
1226 qemu_announce_timer_step(&n
->announce_timer
);
1228 return VIRTIO_NET_OK
;
1230 return VIRTIO_NET_ERR
;
1234 static void virtio_net_detach_epbf_rss(VirtIONet
*n
);
1236 static void virtio_net_disable_rss(VirtIONet
*n
)
1238 if (n
->rss_data
.enabled
) {
1239 trace_virtio_net_rss_disable();
1241 n
->rss_data
.enabled
= false;
1243 virtio_net_detach_epbf_rss(n
);
1246 static bool virtio_net_attach_ebpf_to_backend(NICState
*nic
, int prog_fd
)
1248 NetClientState
*nc
= qemu_get_peer(qemu_get_queue(nic
), 0);
1249 if (nc
== NULL
|| nc
->info
->set_steering_ebpf
== NULL
) {
1253 return nc
->info
->set_steering_ebpf(nc
, prog_fd
);
1256 static void rss_data_to_rss_config(struct VirtioNetRssData
*data
,
1257 struct EBPFRSSConfig
*config
)
1259 config
->redirect
= data
->redirect
;
1260 config
->populate_hash
= data
->populate_hash
;
1261 config
->hash_types
= data
->hash_types
;
1262 config
->indirections_len
= data
->indirections_len
;
1263 config
->default_queue
= data
->default_queue
;
1266 static bool virtio_net_attach_epbf_rss(VirtIONet
*n
)
1268 struct EBPFRSSConfig config
= {};
1270 if (!ebpf_rss_is_loaded(&n
->ebpf_rss
)) {
1274 rss_data_to_rss_config(&n
->rss_data
, &config
);
1276 if (!ebpf_rss_set_all(&n
->ebpf_rss
, &config
,
1277 n
->rss_data
.indirections_table
, n
->rss_data
.key
)) {
1281 if (!virtio_net_attach_ebpf_to_backend(n
->nic
, n
->ebpf_rss
.program_fd
)) {
1288 static void virtio_net_detach_epbf_rss(VirtIONet
*n
)
1290 virtio_net_attach_ebpf_to_backend(n
->nic
, -1);
1293 static bool virtio_net_load_ebpf(VirtIONet
*n
)
1295 if (!virtio_net_attach_ebpf_to_backend(n
->nic
, -1)) {
1296 /* backend does't support steering ebpf */
1300 return ebpf_rss_load(&n
->ebpf_rss
);
1303 static void virtio_net_unload_ebpf(VirtIONet
*n
)
1305 virtio_net_attach_ebpf_to_backend(n
->nic
, -1);
1306 ebpf_rss_unload(&n
->ebpf_rss
);
1309 static uint16_t virtio_net_handle_rss(VirtIONet
*n
,
1311 unsigned int iov_cnt
,
1314 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1315 struct virtio_net_rss_config cfg
;
1316 size_t s
, offset
= 0, size_get
;
1317 uint16_t queue_pairs
, i
;
1322 const char *err_msg
= "";
1323 uint32_t err_value
= 0;
1325 if (do_rss
&& !virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_RSS
)) {
1326 err_msg
= "RSS is not negotiated";
1329 if (!do_rss
&& !virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_HASH_REPORT
)) {
1330 err_msg
= "Hash report is not negotiated";
1333 size_get
= offsetof(struct virtio_net_rss_config
, indirection_table
);
1334 s
= iov_to_buf(iov
, iov_cnt
, offset
, &cfg
, size_get
);
1335 if (s
!= size_get
) {
1336 err_msg
= "Short command buffer";
1337 err_value
= (uint32_t)s
;
1340 n
->rss_data
.hash_types
= virtio_ldl_p(vdev
, &cfg
.hash_types
);
1341 n
->rss_data
.indirections_len
=
1342 virtio_lduw_p(vdev
, &cfg
.indirection_table_mask
);
1343 n
->rss_data
.indirections_len
++;
1345 n
->rss_data
.indirections_len
= 1;
1347 if (!is_power_of_2(n
->rss_data
.indirections_len
)) {
1348 err_msg
= "Invalid size of indirection table";
1349 err_value
= n
->rss_data
.indirections_len
;
1352 if (n
->rss_data
.indirections_len
> VIRTIO_NET_RSS_MAX_TABLE_LEN
) {
1353 err_msg
= "Too large indirection table";
1354 err_value
= n
->rss_data
.indirections_len
;
1357 n
->rss_data
.default_queue
= do_rss
?
1358 virtio_lduw_p(vdev
, &cfg
.unclassified_queue
) : 0;
1359 if (n
->rss_data
.default_queue
>= n
->max_queue_pairs
) {
1360 err_msg
= "Invalid default queue";
1361 err_value
= n
->rss_data
.default_queue
;
1365 size_get
= sizeof(uint16_t) * n
->rss_data
.indirections_len
;
1366 g_free(n
->rss_data
.indirections_table
);
1367 n
->rss_data
.indirections_table
= g_malloc(size_get
);
1368 if (!n
->rss_data
.indirections_table
) {
1369 err_msg
= "Can't allocate indirections table";
1370 err_value
= n
->rss_data
.indirections_len
;
1373 s
= iov_to_buf(iov
, iov_cnt
, offset
,
1374 n
->rss_data
.indirections_table
, size_get
);
1375 if (s
!= size_get
) {
1376 err_msg
= "Short indirection table buffer";
1377 err_value
= (uint32_t)s
;
1380 for (i
= 0; i
< n
->rss_data
.indirections_len
; ++i
) {
1381 uint16_t val
= n
->rss_data
.indirections_table
[i
];
1382 n
->rss_data
.indirections_table
[i
] = virtio_lduw_p(vdev
, &val
);
1385 size_get
= sizeof(temp
);
1386 s
= iov_to_buf(iov
, iov_cnt
, offset
, &temp
, size_get
);
1387 if (s
!= size_get
) {
1388 err_msg
= "Can't get queue_pairs";
1389 err_value
= (uint32_t)s
;
1392 queue_pairs
= do_rss
? virtio_lduw_p(vdev
, &temp
.us
) : n
->curr_queue_pairs
;
1393 if (queue_pairs
== 0 || queue_pairs
> n
->max_queue_pairs
) {
1394 err_msg
= "Invalid number of queue_pairs";
1395 err_value
= queue_pairs
;
1398 if (temp
.b
> VIRTIO_NET_RSS_MAX_KEY_SIZE
) {
1399 err_msg
= "Invalid key size";
1403 if (!temp
.b
&& n
->rss_data
.hash_types
) {
1404 err_msg
= "No key provided";
1408 if (!temp
.b
&& !n
->rss_data
.hash_types
) {
1409 virtio_net_disable_rss(n
);
1414 s
= iov_to_buf(iov
, iov_cnt
, offset
, n
->rss_data
.key
, size_get
);
1415 if (s
!= size_get
) {
1416 err_msg
= "Can get key buffer";
1417 err_value
= (uint32_t)s
;
1420 n
->rss_data
.enabled
= true;
1422 if (!n
->rss_data
.populate_hash
) {
1423 if (!virtio_net_attach_epbf_rss(n
)) {
1424 /* EBPF must be loaded for vhost */
1425 if (get_vhost_net(qemu_get_queue(n
->nic
)->peer
)) {
1426 warn_report("Can't load eBPF RSS for vhost");
1429 /* fallback to software RSS */
1430 warn_report("Can't load eBPF RSS - fallback to software RSS");
1431 n
->rss_data
.enabled_software_rss
= true;
1434 /* use software RSS for hash populating */
1435 /* and detach eBPF if was loaded before */
1436 virtio_net_detach_epbf_rss(n
);
1437 n
->rss_data
.enabled_software_rss
= true;
1440 trace_virtio_net_rss_enable(n
->rss_data
.hash_types
,
1441 n
->rss_data
.indirections_len
,
1445 trace_virtio_net_rss_error(err_msg
, err_value
);
1446 virtio_net_disable_rss(n
);
1450 static int virtio_net_handle_mq(VirtIONet
*n
, uint8_t cmd
,
1451 struct iovec
*iov
, unsigned int iov_cnt
)
1453 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1454 uint16_t queue_pairs
;
1455 NetClientState
*nc
= qemu_get_queue(n
->nic
);
1457 virtio_net_disable_rss(n
);
1458 if (cmd
== VIRTIO_NET_CTRL_MQ_HASH_CONFIG
) {
1459 queue_pairs
= virtio_net_handle_rss(n
, iov
, iov_cnt
, false);
1460 return queue_pairs
? VIRTIO_NET_OK
: VIRTIO_NET_ERR
;
1462 if (cmd
== VIRTIO_NET_CTRL_MQ_RSS_CONFIG
) {
1463 queue_pairs
= virtio_net_handle_rss(n
, iov
, iov_cnt
, true);
1464 } else if (cmd
== VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
) {
1465 struct virtio_net_ctrl_mq mq
;
1467 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_MQ
)) {
1468 return VIRTIO_NET_ERR
;
1470 s
= iov_to_buf(iov
, iov_cnt
, 0, &mq
, sizeof(mq
));
1471 if (s
!= sizeof(mq
)) {
1472 return VIRTIO_NET_ERR
;
1474 queue_pairs
= virtio_lduw_p(vdev
, &mq
.virtqueue_pairs
);
1477 return VIRTIO_NET_ERR
;
1480 if (queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
1481 queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
1482 queue_pairs
> n
->max_queue_pairs
||
1484 return VIRTIO_NET_ERR
;
1487 n
->curr_queue_pairs
= queue_pairs
;
1488 if (nc
->peer
&& nc
->peer
->info
->type
== NET_CLIENT_DRIVER_VHOST_VDPA
) {
1490 * Avoid updating the backend for a vdpa device: We're only interested
1491 * in updating the device model queues.
1493 return VIRTIO_NET_OK
;
1495 /* stop the backend before changing the number of queue_pairs to avoid handling a
1497 virtio_net_set_status(vdev
, vdev
->status
);
1498 virtio_net_set_queue_pairs(n
);
1500 return VIRTIO_NET_OK
;
1503 size_t virtio_net_handle_ctrl_iov(VirtIODevice
*vdev
,
1504 const struct iovec
*in_sg
, unsigned in_num
,
1505 const struct iovec
*out_sg
,
1508 VirtIONet
*n
= VIRTIO_NET(vdev
);
1509 struct virtio_net_ctrl_hdr ctrl
;
1510 virtio_net_ctrl_ack status
= VIRTIO_NET_ERR
;
1512 struct iovec
*iov
, *iov2
;
1514 if (iov_size(in_sg
, in_num
) < sizeof(status
) ||
1515 iov_size(out_sg
, out_num
) < sizeof(ctrl
)) {
1516 virtio_error(vdev
, "virtio-net ctrl missing headers");
1520 iov2
= iov
= g_memdup2(out_sg
, sizeof(struct iovec
) * out_num
);
1521 s
= iov_to_buf(iov
, out_num
, 0, &ctrl
, sizeof(ctrl
));
1522 iov_discard_front(&iov
, &out_num
, sizeof(ctrl
));
1523 if (s
!= sizeof(ctrl
)) {
1524 status
= VIRTIO_NET_ERR
;
1525 } else if (ctrl
.class == VIRTIO_NET_CTRL_RX
) {
1526 status
= virtio_net_handle_rx_mode(n
, ctrl
.cmd
, iov
, out_num
);
1527 } else if (ctrl
.class == VIRTIO_NET_CTRL_MAC
) {
1528 status
= virtio_net_handle_mac(n
, ctrl
.cmd
, iov
, out_num
);
1529 } else if (ctrl
.class == VIRTIO_NET_CTRL_VLAN
) {
1530 status
= virtio_net_handle_vlan_table(n
, ctrl
.cmd
, iov
, out_num
);
1531 } else if (ctrl
.class == VIRTIO_NET_CTRL_ANNOUNCE
) {
1532 status
= virtio_net_handle_announce(n
, ctrl
.cmd
, iov
, out_num
);
1533 } else if (ctrl
.class == VIRTIO_NET_CTRL_MQ
) {
1534 status
= virtio_net_handle_mq(n
, ctrl
.cmd
, iov
, out_num
);
1535 } else if (ctrl
.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS
) {
1536 status
= virtio_net_handle_offloads(n
, ctrl
.cmd
, iov
, out_num
);
1539 s
= iov_from_buf(in_sg
, in_num
, 0, &status
, sizeof(status
));
1540 assert(s
== sizeof(status
));
1543 return sizeof(status
);
1546 static void virtio_net_handle_ctrl(VirtIODevice
*vdev
, VirtQueue
*vq
)
1548 VirtQueueElement
*elem
;
1552 elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
));
1557 written
= virtio_net_handle_ctrl_iov(vdev
, elem
->in_sg
, elem
->in_num
,
1558 elem
->out_sg
, elem
->out_num
);
1560 virtqueue_push(vq
, elem
, written
);
1561 virtio_notify(vdev
, vq
);
1564 virtqueue_detach_element(vq
, elem
, 0);
1573 static void virtio_net_handle_rx(VirtIODevice
*vdev
, VirtQueue
*vq
)
1575 VirtIONet
*n
= VIRTIO_NET(vdev
);
1576 int queue_index
= vq2q(virtio_get_queue_index(vq
));
1578 qemu_flush_queued_packets(qemu_get_subqueue(n
->nic
, queue_index
));
1581 static bool virtio_net_can_receive(NetClientState
*nc
)
1583 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
1584 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1585 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
1587 if (!vdev
->vm_running
) {
1591 if (nc
->queue_index
>= n
->curr_queue_pairs
) {
1595 if (!virtio_queue_ready(q
->rx_vq
) ||
1596 !(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
1603 static int virtio_net_has_buffers(VirtIONetQueue
*q
, int bufsize
)
1605 VirtIONet
*n
= q
->n
;
1606 if (virtio_queue_empty(q
->rx_vq
) ||
1607 (n
->mergeable_rx_bufs
&&
1608 !virtqueue_avail_bytes(q
->rx_vq
, bufsize
, 0))) {
1609 virtio_queue_set_notification(q
->rx_vq
, 1);
1611 /* To avoid a race condition where the guest has made some buffers
1612 * available after the above check but before notification was
1613 * enabled, check for available buffers again.
1615 if (virtio_queue_empty(q
->rx_vq
) ||
1616 (n
->mergeable_rx_bufs
&&
1617 !virtqueue_avail_bytes(q
->rx_vq
, bufsize
, 0))) {
1622 virtio_queue_set_notification(q
->rx_vq
, 0);
1626 static void virtio_net_hdr_swap(VirtIODevice
*vdev
, struct virtio_net_hdr
*hdr
)
1628 virtio_tswap16s(vdev
, &hdr
->hdr_len
);
1629 virtio_tswap16s(vdev
, &hdr
->gso_size
);
1630 virtio_tswap16s(vdev
, &hdr
->csum_start
);
1631 virtio_tswap16s(vdev
, &hdr
->csum_offset
);
1634 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1635 * it never finds out that the packets don't have valid checksums. This
1636 * causes dhclient to get upset. Fedora's carried a patch for ages to
1637 * fix this with Xen but it hasn't appeared in an upstream release of
1640 * To avoid breaking existing guests, we catch udp packets and add
1641 * checksums. This is terrible but it's better than hacking the guest
1644 * N.B. if we introduce a zero-copy API, this operation is no longer free so
1645 * we should provide a mechanism to disable it to avoid polluting the host
1648 static void work_around_broken_dhclient(struct virtio_net_hdr
*hdr
,
1649 uint8_t *buf
, size_t size
)
1651 if ((hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) && /* missing csum */
1652 (size
> 27 && size
< 1500) && /* normal sized MTU */
1653 (buf
[12] == 0x08 && buf
[13] == 0x00) && /* ethertype == IPv4 */
1654 (buf
[23] == 17) && /* ip.protocol == UDP */
1655 (buf
[34] == 0 && buf
[35] == 67)) { /* udp.srcport == bootps */
1656 net_checksum_calculate(buf
, size
, CSUM_UDP
);
1657 hdr
->flags
&= ~VIRTIO_NET_HDR_F_NEEDS_CSUM
;
1661 static void receive_header(VirtIONet
*n
, const struct iovec
*iov
, int iov_cnt
,
1662 const void *buf
, size_t size
)
1664 if (n
->has_vnet_hdr
) {
1665 /* FIXME this cast is evil */
1666 void *wbuf
= (void *)buf
;
1667 work_around_broken_dhclient(wbuf
, wbuf
+ n
->host_hdr_len
,
1668 size
- n
->host_hdr_len
);
1670 if (n
->needs_vnet_hdr_swap
) {
1671 virtio_net_hdr_swap(VIRTIO_DEVICE(n
), wbuf
);
1673 iov_from_buf(iov
, iov_cnt
, 0, buf
, sizeof(struct virtio_net_hdr
));
1675 struct virtio_net_hdr hdr
= {
1677 .gso_type
= VIRTIO_NET_HDR_GSO_NONE
1679 iov_from_buf(iov
, iov_cnt
, 0, &hdr
, sizeof hdr
);
1683 static int receive_filter(VirtIONet
*n
, const uint8_t *buf
, int size
)
1685 static const uint8_t bcast
[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1686 static const uint8_t vlan
[] = {0x81, 0x00};
1687 uint8_t *ptr
= (uint8_t *)buf
;
1693 ptr
+= n
->host_hdr_len
;
1695 if (!memcmp(&ptr
[12], vlan
, sizeof(vlan
))) {
1696 int vid
= lduw_be_p(ptr
+ 14) & 0xfff;
1697 if (!(n
->vlans
[vid
>> 5] & (1U << (vid
& 0x1f))))
1701 if (ptr
[0] & 1) { // multicast
1702 if (!memcmp(ptr
, bcast
, sizeof(bcast
))) {
1704 } else if (n
->nomulti
) {
1706 } else if (n
->allmulti
|| n
->mac_table
.multi_overflow
) {
1710 for (i
= n
->mac_table
.first_multi
; i
< n
->mac_table
.in_use
; i
++) {
1711 if (!memcmp(ptr
, &n
->mac_table
.macs
[i
* ETH_ALEN
], ETH_ALEN
)) {
1718 } else if (n
->alluni
|| n
->mac_table
.uni_overflow
) {
1720 } else if (!memcmp(ptr
, n
->mac
, ETH_ALEN
)) {
1724 for (i
= 0; i
< n
->mac_table
.first_multi
; i
++) {
1725 if (!memcmp(ptr
, &n
->mac_table
.macs
[i
* ETH_ALEN
], ETH_ALEN
)) {
1734 static uint8_t virtio_net_get_hash_type(bool isip4
,
1741 if (istcp
&& (types
& VIRTIO_NET_RSS_HASH_TYPE_TCPv4
)) {
1742 return NetPktRssIpV4Tcp
;
1744 if (isudp
&& (types
& VIRTIO_NET_RSS_HASH_TYPE_UDPv4
)) {
1745 return NetPktRssIpV4Udp
;
1747 if (types
& VIRTIO_NET_RSS_HASH_TYPE_IPv4
) {
1748 return NetPktRssIpV4
;
1751 uint32_t mask
= VIRTIO_NET_RSS_HASH_TYPE_TCP_EX
|
1752 VIRTIO_NET_RSS_HASH_TYPE_TCPv6
;
1754 if (istcp
&& (types
& mask
)) {
1755 return (types
& VIRTIO_NET_RSS_HASH_TYPE_TCP_EX
) ?
1756 NetPktRssIpV6TcpEx
: NetPktRssIpV6Tcp
;
1758 mask
= VIRTIO_NET_RSS_HASH_TYPE_UDP_EX
| VIRTIO_NET_RSS_HASH_TYPE_UDPv6
;
1759 if (isudp
&& (types
& mask
)) {
1760 return (types
& VIRTIO_NET_RSS_HASH_TYPE_UDP_EX
) ?
1761 NetPktRssIpV6UdpEx
: NetPktRssIpV6Udp
;
1763 mask
= VIRTIO_NET_RSS_HASH_TYPE_IP_EX
| VIRTIO_NET_RSS_HASH_TYPE_IPv6
;
1765 return (types
& VIRTIO_NET_RSS_HASH_TYPE_IP_EX
) ?
1766 NetPktRssIpV6Ex
: NetPktRssIpV6
;
1772 static void virtio_set_packet_hash(const uint8_t *buf
, uint8_t report
,
1775 struct virtio_net_hdr_v1_hash
*hdr
= (void *)buf
;
1776 hdr
->hash_value
= hash
;
1777 hdr
->hash_report
= report
;
1780 static int virtio_net_process_rss(NetClientState
*nc
, const uint8_t *buf
,
1783 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
1784 unsigned int index
= nc
->queue_index
, new_index
= index
;
1785 struct NetRxPkt
*pkt
= n
->rx_pkt
;
1786 uint8_t net_hash_type
;
1788 bool isip4
, isip6
, isudp
, istcp
;
1789 static const uint8_t reports
[NetPktRssIpV6UdpEx
+ 1] = {
1790 VIRTIO_NET_HASH_REPORT_IPv4
,
1791 VIRTIO_NET_HASH_REPORT_TCPv4
,
1792 VIRTIO_NET_HASH_REPORT_TCPv6
,
1793 VIRTIO_NET_HASH_REPORT_IPv6
,
1794 VIRTIO_NET_HASH_REPORT_IPv6_EX
,
1795 VIRTIO_NET_HASH_REPORT_TCPv6_EX
,
1796 VIRTIO_NET_HASH_REPORT_UDPv4
,
1797 VIRTIO_NET_HASH_REPORT_UDPv6
,
1798 VIRTIO_NET_HASH_REPORT_UDPv6_EX
1801 net_rx_pkt_set_protocols(pkt
, buf
+ n
->host_hdr_len
,
1802 size
- n
->host_hdr_len
);
1803 net_rx_pkt_get_protocols(pkt
, &isip4
, &isip6
, &isudp
, &istcp
);
1804 if (isip4
&& (net_rx_pkt_get_ip4_info(pkt
)->fragment
)) {
1805 istcp
= isudp
= false;
1807 if (isip6
&& (net_rx_pkt_get_ip6_info(pkt
)->fragment
)) {
1808 istcp
= isudp
= false;
1810 net_hash_type
= virtio_net_get_hash_type(isip4
, isip6
, isudp
, istcp
,
1811 n
->rss_data
.hash_types
);
1812 if (net_hash_type
> NetPktRssIpV6UdpEx
) {
1813 if (n
->rss_data
.populate_hash
) {
1814 virtio_set_packet_hash(buf
, VIRTIO_NET_HASH_REPORT_NONE
, 0);
1816 return n
->rss_data
.redirect
? n
->rss_data
.default_queue
: -1;
1819 hash
= net_rx_pkt_calc_rss_hash(pkt
, net_hash_type
, n
->rss_data
.key
);
1821 if (n
->rss_data
.populate_hash
) {
1822 virtio_set_packet_hash(buf
, reports
[net_hash_type
], hash
);
1825 if (n
->rss_data
.redirect
) {
1826 new_index
= hash
& (n
->rss_data
.indirections_len
- 1);
1827 new_index
= n
->rss_data
.indirections_table
[new_index
];
1830 return (index
== new_index
) ? -1 : new_index
;
1833 static ssize_t
virtio_net_receive_rcu(NetClientState
*nc
, const uint8_t *buf
,
1834 size_t size
, bool no_rss
)
1836 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
1837 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
1838 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
1839 VirtQueueElement
*elems
[VIRTQUEUE_MAX_SIZE
];
1840 size_t lens
[VIRTQUEUE_MAX_SIZE
];
1841 struct iovec mhdr_sg
[VIRTQUEUE_MAX_SIZE
];
1842 struct virtio_net_hdr_mrg_rxbuf mhdr
;
1843 unsigned mhdr_cnt
= 0;
1844 size_t offset
, i
, guest_offset
, j
;
1847 if (!virtio_net_can_receive(nc
)) {
1851 if (!no_rss
&& n
->rss_data
.enabled
&& n
->rss_data
.enabled_software_rss
) {
1852 int index
= virtio_net_process_rss(nc
, buf
, size
);
1854 NetClientState
*nc2
= qemu_get_subqueue(n
->nic
, index
);
1855 return virtio_net_receive_rcu(nc2
, buf
, size
, true);
1859 /* hdr_len refers to the header we supply to the guest */
1860 if (!virtio_net_has_buffers(q
, size
+ n
->guest_hdr_len
- n
->host_hdr_len
)) {
1864 if (!receive_filter(n
, buf
, size
))
1869 while (offset
< size
) {
1870 VirtQueueElement
*elem
;
1872 const struct iovec
*sg
;
1876 if (i
== VIRTQUEUE_MAX_SIZE
) {
1877 virtio_error(vdev
, "virtio-net unexpected long buffer chain");
1882 elem
= virtqueue_pop(q
->rx_vq
, sizeof(VirtQueueElement
));
1885 virtio_error(vdev
, "virtio-net unexpected empty queue: "
1886 "i %zd mergeable %d offset %zd, size %zd, "
1887 "guest hdr len %zd, host hdr len %zd "
1888 "guest features 0x%" PRIx64
,
1889 i
, n
->mergeable_rx_bufs
, offset
, size
,
1890 n
->guest_hdr_len
, n
->host_hdr_len
,
1891 vdev
->guest_features
);
1897 if (elem
->in_num
< 1) {
1899 "virtio-net receive queue contains no in buffers");
1900 virtqueue_detach_element(q
->rx_vq
, elem
, 0);
1908 assert(offset
== 0);
1909 if (n
->mergeable_rx_bufs
) {
1910 mhdr_cnt
= iov_copy(mhdr_sg
, ARRAY_SIZE(mhdr_sg
),
1912 offsetof(typeof(mhdr
), num_buffers
),
1913 sizeof(mhdr
.num_buffers
));
1916 receive_header(n
, sg
, elem
->in_num
, buf
, size
);
1917 if (n
->rss_data
.populate_hash
) {
1918 offset
= sizeof(mhdr
);
1919 iov_from_buf(sg
, elem
->in_num
, offset
,
1920 buf
+ offset
, n
->host_hdr_len
- sizeof(mhdr
));
1922 offset
= n
->host_hdr_len
;
1923 total
+= n
->guest_hdr_len
;
1924 guest_offset
= n
->guest_hdr_len
;
1929 /* copy in packet. ugh */
1930 len
= iov_from_buf(sg
, elem
->in_num
, guest_offset
,
1931 buf
+ offset
, size
- offset
);
1934 /* If buffers can't be merged, at this point we
1935 * must have consumed the complete packet.
1936 * Otherwise, drop it. */
1937 if (!n
->mergeable_rx_bufs
&& offset
< size
) {
1938 virtqueue_unpop(q
->rx_vq
, elem
, total
);
1950 virtio_stw_p(vdev
, &mhdr
.num_buffers
, i
);
1951 iov_from_buf(mhdr_sg
, mhdr_cnt
,
1953 &mhdr
.num_buffers
, sizeof mhdr
.num_buffers
);
1956 for (j
= 0; j
< i
; j
++) {
1957 /* signal other side */
1958 virtqueue_fill(q
->rx_vq
, elems
[j
], lens
[j
], j
);
1962 virtqueue_flush(q
->rx_vq
, i
);
1963 virtio_notify(vdev
, q
->rx_vq
);
1968 for (j
= 0; j
< i
; j
++) {
1969 virtqueue_detach_element(q
->rx_vq
, elems
[j
], lens
[j
]);
1976 static ssize_t
virtio_net_do_receive(NetClientState
*nc
, const uint8_t *buf
,
1979 RCU_READ_LOCK_GUARD();
1981 return virtio_net_receive_rcu(nc
, buf
, size
, false);
1984 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain
*chain
,
1986 VirtioNetRscUnit
*unit
)
1989 struct ip_header
*ip
;
1991 ip
= (struct ip_header
*)(buf
+ chain
->n
->guest_hdr_len
1992 + sizeof(struct eth_header
));
1993 unit
->ip
= (void *)ip
;
1994 ip_hdrlen
= (ip
->ip_ver_len
& 0xF) << 2;
1995 unit
->ip_plen
= &ip
->ip_len
;
1996 unit
->tcp
= (struct tcp_header
*)(((uint8_t *)unit
->ip
) + ip_hdrlen
);
1997 unit
->tcp_hdrlen
= (htons(unit
->tcp
->th_offset_flags
) & 0xF000) >> 10;
1998 unit
->payload
= htons(*unit
->ip_plen
) - ip_hdrlen
- unit
->tcp_hdrlen
;
2001 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain
*chain
,
2003 VirtioNetRscUnit
*unit
)
2005 struct ip6_header
*ip6
;
2007 ip6
= (struct ip6_header
*)(buf
+ chain
->n
->guest_hdr_len
2008 + sizeof(struct eth_header
));
2010 unit
->ip_plen
= &(ip6
->ip6_ctlun
.ip6_un1
.ip6_un1_plen
);
2011 unit
->tcp
= (struct tcp_header
*)(((uint8_t *)unit
->ip
)
2012 + sizeof(struct ip6_header
));
2013 unit
->tcp_hdrlen
= (htons(unit
->tcp
->th_offset_flags
) & 0xF000) >> 10;
2015 /* There is a difference between payload lenght in ipv4 and v6,
2016 ip header is excluded in ipv6 */
2017 unit
->payload
= htons(*unit
->ip_plen
) - unit
->tcp_hdrlen
;
2020 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain
*chain
,
2021 VirtioNetRscSeg
*seg
)
2024 struct virtio_net_hdr_v1
*h
;
2026 h
= (struct virtio_net_hdr_v1
*)seg
->buf
;
2028 h
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
2030 if (seg
->is_coalesced
) {
2031 h
->rsc
.segments
= seg
->packets
;
2032 h
->rsc
.dup_acks
= seg
->dup_ack
;
2033 h
->flags
= VIRTIO_NET_HDR_F_RSC_INFO
;
2034 if (chain
->proto
== ETH_P_IP
) {
2035 h
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
2037 h
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
2041 ret
= virtio_net_do_receive(seg
->nc
, seg
->buf
, seg
->size
);
2042 QTAILQ_REMOVE(&chain
->buffers
, seg
, next
);
2049 static void virtio_net_rsc_purge(void *opq
)
2051 VirtioNetRscSeg
*seg
, *rn
;
2052 VirtioNetRscChain
*chain
= (VirtioNetRscChain
*)opq
;
2054 QTAILQ_FOREACH_SAFE(seg
, &chain
->buffers
, next
, rn
) {
2055 if (virtio_net_rsc_drain_seg(chain
, seg
) == 0) {
2056 chain
->stat
.purge_failed
++;
2061 chain
->stat
.timer
++;
2062 if (!QTAILQ_EMPTY(&chain
->buffers
)) {
2063 timer_mod(chain
->drain_timer
,
2064 qemu_clock_get_ns(QEMU_CLOCK_HOST
) + chain
->n
->rsc_timeout
);
2068 static void virtio_net_rsc_cleanup(VirtIONet
*n
)
2070 VirtioNetRscChain
*chain
, *rn_chain
;
2071 VirtioNetRscSeg
*seg
, *rn_seg
;
2073 QTAILQ_FOREACH_SAFE(chain
, &n
->rsc_chains
, next
, rn_chain
) {
2074 QTAILQ_FOREACH_SAFE(seg
, &chain
->buffers
, next
, rn_seg
) {
2075 QTAILQ_REMOVE(&chain
->buffers
, seg
, next
);
2080 timer_free(chain
->drain_timer
);
2081 QTAILQ_REMOVE(&n
->rsc_chains
, chain
, next
);
2086 static void virtio_net_rsc_cache_buf(VirtioNetRscChain
*chain
,
2088 const uint8_t *buf
, size_t size
)
2091 VirtioNetRscSeg
*seg
;
2093 hdr_len
= chain
->n
->guest_hdr_len
;
2094 seg
= g_new(VirtioNetRscSeg
, 1);
2095 seg
->buf
= g_malloc(hdr_len
+ sizeof(struct eth_header
)
2096 + sizeof(struct ip6_header
) + VIRTIO_NET_MAX_TCP_PAYLOAD
);
2097 memcpy(seg
->buf
, buf
, size
);
2101 seg
->is_coalesced
= 0;
2104 QTAILQ_INSERT_TAIL(&chain
->buffers
, seg
, next
);
2105 chain
->stat
.cache
++;
2107 switch (chain
->proto
) {
2109 virtio_net_rsc_extract_unit4(chain
, seg
->buf
, &seg
->unit
);
2112 virtio_net_rsc_extract_unit6(chain
, seg
->buf
, &seg
->unit
);
2115 g_assert_not_reached();
2119 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain
*chain
,
2120 VirtioNetRscSeg
*seg
,
2122 struct tcp_header
*n_tcp
,
2123 struct tcp_header
*o_tcp
)
2125 uint32_t nack
, oack
;
2126 uint16_t nwin
, owin
;
2128 nack
= htonl(n_tcp
->th_ack
);
2129 nwin
= htons(n_tcp
->th_win
);
2130 oack
= htonl(o_tcp
->th_ack
);
2131 owin
= htons(o_tcp
->th_win
);
2133 if ((nack
- oack
) >= VIRTIO_NET_MAX_TCP_PAYLOAD
) {
2134 chain
->stat
.ack_out_of_win
++;
2136 } else if (nack
== oack
) {
2137 /* duplicated ack or window probe */
2139 /* duplicated ack, add dup ack count due to whql test up to 1 */
2140 chain
->stat
.dup_ack
++;
2143 /* Coalesce window update */
2144 o_tcp
->th_win
= n_tcp
->th_win
;
2145 chain
->stat
.win_update
++;
2146 return RSC_COALESCE
;
2149 /* pure ack, go to 'C', finalize*/
2150 chain
->stat
.pure_ack
++;
2155 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain
*chain
,
2156 VirtioNetRscSeg
*seg
,
2158 VirtioNetRscUnit
*n_unit
)
2162 uint32_t nseq
, oseq
;
2163 VirtioNetRscUnit
*o_unit
;
2165 o_unit
= &seg
->unit
;
2166 o_ip_len
= htons(*o_unit
->ip_plen
);
2167 nseq
= htonl(n_unit
->tcp
->th_seq
);
2168 oseq
= htonl(o_unit
->tcp
->th_seq
);
2170 /* out of order or retransmitted. */
2171 if ((nseq
- oseq
) > VIRTIO_NET_MAX_TCP_PAYLOAD
) {
2172 chain
->stat
.data_out_of_win
++;
2176 data
= ((uint8_t *)n_unit
->tcp
) + n_unit
->tcp_hdrlen
;
2178 if ((o_unit
->payload
== 0) && n_unit
->payload
) {
2179 /* From no payload to payload, normal case, not a dup ack or etc */
2180 chain
->stat
.data_after_pure_ack
++;
2183 return virtio_net_rsc_handle_ack(chain
, seg
, buf
,
2184 n_unit
->tcp
, o_unit
->tcp
);
2186 } else if ((nseq
- oseq
) != o_unit
->payload
) {
2187 /* Not a consistent packet, out of order */
2188 chain
->stat
.data_out_of_order
++;
2192 if ((o_ip_len
+ n_unit
->payload
) > chain
->max_payload
) {
2193 chain
->stat
.over_size
++;
2197 /* Here comes the right data, the payload length in v4/v6 is different,
2198 so use the field value to update and record the new data len */
2199 o_unit
->payload
+= n_unit
->payload
; /* update new data len */
2201 /* update field in ip header */
2202 *o_unit
->ip_plen
= htons(o_ip_len
+ n_unit
->payload
);
2204 /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
2205 for windows guest, while this may change the behavior for linux
2206 guest (only if it uses RSC feature). */
2207 o_unit
->tcp
->th_offset_flags
= n_unit
->tcp
->th_offset_flags
;
2209 o_unit
->tcp
->th_ack
= n_unit
->tcp
->th_ack
;
2210 o_unit
->tcp
->th_win
= n_unit
->tcp
->th_win
;
2212 memmove(seg
->buf
+ seg
->size
, data
, n_unit
->payload
);
2213 seg
->size
+= n_unit
->payload
;
2215 chain
->stat
.coalesced
++;
2216 return RSC_COALESCE
;
2220 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain
*chain
,
2221 VirtioNetRscSeg
*seg
,
2222 const uint8_t *buf
, size_t size
,
2223 VirtioNetRscUnit
*unit
)
2225 struct ip_header
*ip1
, *ip2
;
2227 ip1
= (struct ip_header
*)(unit
->ip
);
2228 ip2
= (struct ip_header
*)(seg
->unit
.ip
);
2229 if ((ip1
->ip_src
^ ip2
->ip_src
) || (ip1
->ip_dst
^ ip2
->ip_dst
)
2230 || (unit
->tcp
->th_sport
^ seg
->unit
.tcp
->th_sport
)
2231 || (unit
->tcp
->th_dport
^ seg
->unit
.tcp
->th_dport
)) {
2232 chain
->stat
.no_match
++;
2233 return RSC_NO_MATCH
;
2236 return virtio_net_rsc_coalesce_data(chain
, seg
, buf
, unit
);
2239 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain
*chain
,
2240 VirtioNetRscSeg
*seg
,
2241 const uint8_t *buf
, size_t size
,
2242 VirtioNetRscUnit
*unit
)
2244 struct ip6_header
*ip1
, *ip2
;
2246 ip1
= (struct ip6_header
*)(unit
->ip
);
2247 ip2
= (struct ip6_header
*)(seg
->unit
.ip
);
2248 if (memcmp(&ip1
->ip6_src
, &ip2
->ip6_src
, sizeof(struct in6_address
))
2249 || memcmp(&ip1
->ip6_dst
, &ip2
->ip6_dst
, sizeof(struct in6_address
))
2250 || (unit
->tcp
->th_sport
^ seg
->unit
.tcp
->th_sport
)
2251 || (unit
->tcp
->th_dport
^ seg
->unit
.tcp
->th_dport
)) {
2252 chain
->stat
.no_match
++;
2253 return RSC_NO_MATCH
;
2256 return virtio_net_rsc_coalesce_data(chain
, seg
, buf
, unit
);
2259 /* Packets with 'SYN' should bypass, other flag should be sent after drain
2260 * to prevent out of order */
2261 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain
*chain
,
2262 struct tcp_header
*tcp
)
2267 tcp_flag
= htons(tcp
->th_offset_flags
);
2268 tcp_hdr
= (tcp_flag
& VIRTIO_NET_TCP_HDR_LENGTH
) >> 10;
2269 tcp_flag
&= VIRTIO_NET_TCP_FLAG
;
2270 if (tcp_flag
& TH_SYN
) {
2271 chain
->stat
.tcp_syn
++;
2275 if (tcp_flag
& (TH_FIN
| TH_URG
| TH_RST
| TH_ECE
| TH_CWR
)) {
2276 chain
->stat
.tcp_ctrl_drain
++;
2280 if (tcp_hdr
> sizeof(struct tcp_header
)) {
2281 chain
->stat
.tcp_all_opt
++;
2285 return RSC_CANDIDATE
;
2288 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain
*chain
,
2290 const uint8_t *buf
, size_t size
,
2291 VirtioNetRscUnit
*unit
)
2294 VirtioNetRscSeg
*seg
, *nseg
;
2296 if (QTAILQ_EMPTY(&chain
->buffers
)) {
2297 chain
->stat
.empty_cache
++;
2298 virtio_net_rsc_cache_buf(chain
, nc
, buf
, size
);
2299 timer_mod(chain
->drain_timer
,
2300 qemu_clock_get_ns(QEMU_CLOCK_HOST
) + chain
->n
->rsc_timeout
);
2304 QTAILQ_FOREACH_SAFE(seg
, &chain
->buffers
, next
, nseg
) {
2305 if (chain
->proto
== ETH_P_IP
) {
2306 ret
= virtio_net_rsc_coalesce4(chain
, seg
, buf
, size
, unit
);
2308 ret
= virtio_net_rsc_coalesce6(chain
, seg
, buf
, size
, unit
);
2311 if (ret
== RSC_FINAL
) {
2312 if (virtio_net_rsc_drain_seg(chain
, seg
) == 0) {
2314 chain
->stat
.final_failed
++;
2318 /* Send current packet */
2319 return virtio_net_do_receive(nc
, buf
, size
);
2320 } else if (ret
== RSC_NO_MATCH
) {
2323 /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
2324 seg
->is_coalesced
= 1;
2329 chain
->stat
.no_match_cache
++;
2330 virtio_net_rsc_cache_buf(chain
, nc
, buf
, size
);
2334 /* Drain a connection data, this is to avoid out of order segments */
2335 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain
*chain
,
2337 const uint8_t *buf
, size_t size
,
2338 uint16_t ip_start
, uint16_t ip_size
,
2341 VirtioNetRscSeg
*seg
, *nseg
;
2342 uint32_t ppair1
, ppair2
;
2344 ppair1
= *(uint32_t *)(buf
+ tcp_port
);
2345 QTAILQ_FOREACH_SAFE(seg
, &chain
->buffers
, next
, nseg
) {
2346 ppair2
= *(uint32_t *)(seg
->buf
+ tcp_port
);
2347 if (memcmp(buf
+ ip_start
, seg
->buf
+ ip_start
, ip_size
)
2348 || (ppair1
!= ppair2
)) {
2351 if (virtio_net_rsc_drain_seg(chain
, seg
) == 0) {
2352 chain
->stat
.drain_failed
++;
2358 return virtio_net_do_receive(nc
, buf
, size
);
2361 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain
*chain
,
2362 struct ip_header
*ip
,
2363 const uint8_t *buf
, size_t size
)
2367 /* Not an ipv4 packet */
2368 if (((ip
->ip_ver_len
& 0xF0) >> 4) != IP_HEADER_VERSION_4
) {
2369 chain
->stat
.ip_option
++;
2373 /* Don't handle packets with ip option */
2374 if ((ip
->ip_ver_len
& 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH
) {
2375 chain
->stat
.ip_option
++;
2379 if (ip
->ip_p
!= IPPROTO_TCP
) {
2380 chain
->stat
.bypass_not_tcp
++;
2384 /* Don't handle packets with ip fragment */
2385 if (!(htons(ip
->ip_off
) & IP_DF
)) {
2386 chain
->stat
.ip_frag
++;
2390 /* Don't handle packets with ecn flag */
2391 if (IPTOS_ECN(ip
->ip_tos
)) {
2392 chain
->stat
.ip_ecn
++;
2396 ip_len
= htons(ip
->ip_len
);
2397 if (ip_len
< (sizeof(struct ip_header
) + sizeof(struct tcp_header
))
2398 || ip_len
> (size
- chain
->n
->guest_hdr_len
-
2399 sizeof(struct eth_header
))) {
2400 chain
->stat
.ip_hacked
++;
2404 return RSC_CANDIDATE
;
2407 static size_t virtio_net_rsc_receive4(VirtioNetRscChain
*chain
,
2409 const uint8_t *buf
, size_t size
)
2413 VirtioNetRscUnit unit
;
2415 hdr_len
= ((VirtIONet
*)(chain
->n
))->guest_hdr_len
;
2417 if (size
< (hdr_len
+ sizeof(struct eth_header
) + sizeof(struct ip_header
)
2418 + sizeof(struct tcp_header
))) {
2419 chain
->stat
.bypass_not_tcp
++;
2420 return virtio_net_do_receive(nc
, buf
, size
);
2423 virtio_net_rsc_extract_unit4(chain
, buf
, &unit
);
2424 if (virtio_net_rsc_sanity_check4(chain
, unit
.ip
, buf
, size
)
2426 return virtio_net_do_receive(nc
, buf
, size
);
2429 ret
= virtio_net_rsc_tcp_ctrl_check(chain
, unit
.tcp
);
2430 if (ret
== RSC_BYPASS
) {
2431 return virtio_net_do_receive(nc
, buf
, size
);
2432 } else if (ret
== RSC_FINAL
) {
2433 return virtio_net_rsc_drain_flow(chain
, nc
, buf
, size
,
2434 ((hdr_len
+ sizeof(struct eth_header
)) + 12),
2435 VIRTIO_NET_IP4_ADDR_SIZE
,
2436 hdr_len
+ sizeof(struct eth_header
) + sizeof(struct ip_header
));
2439 return virtio_net_rsc_do_coalesce(chain
, nc
, buf
, size
, &unit
);
2442 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain
*chain
,
2443 struct ip6_header
*ip6
,
2444 const uint8_t *buf
, size_t size
)
2448 if (((ip6
->ip6_ctlun
.ip6_un1
.ip6_un1_flow
& 0xF0) >> 4)
2449 != IP_HEADER_VERSION_6
) {
2453 /* Both option and protocol is checked in this */
2454 if (ip6
->ip6_ctlun
.ip6_un1
.ip6_un1_nxt
!= IPPROTO_TCP
) {
2455 chain
->stat
.bypass_not_tcp
++;
2459 ip_len
= htons(ip6
->ip6_ctlun
.ip6_un1
.ip6_un1_plen
);
2460 if (ip_len
< sizeof(struct tcp_header
) ||
2461 ip_len
> (size
- chain
->n
->guest_hdr_len
- sizeof(struct eth_header
)
2462 - sizeof(struct ip6_header
))) {
2463 chain
->stat
.ip_hacked
++;
2467 /* Don't handle packets with ecn flag */
2468 if (IP6_ECN(ip6
->ip6_ctlun
.ip6_un3
.ip6_un3_ecn
)) {
2469 chain
->stat
.ip_ecn
++;
2473 return RSC_CANDIDATE
;
2476 static size_t virtio_net_rsc_receive6(void *opq
, NetClientState
*nc
,
2477 const uint8_t *buf
, size_t size
)
2481 VirtioNetRscChain
*chain
;
2482 VirtioNetRscUnit unit
;
2485 hdr_len
= ((VirtIONet
*)(chain
->n
))->guest_hdr_len
;
2487 if (size
< (hdr_len
+ sizeof(struct eth_header
) + sizeof(struct ip6_header
)
2488 + sizeof(tcp_header
))) {
2489 return virtio_net_do_receive(nc
, buf
, size
);
2492 virtio_net_rsc_extract_unit6(chain
, buf
, &unit
);
2493 if (RSC_CANDIDATE
!= virtio_net_rsc_sanity_check6(chain
,
2494 unit
.ip
, buf
, size
)) {
2495 return virtio_net_do_receive(nc
, buf
, size
);
2498 ret
= virtio_net_rsc_tcp_ctrl_check(chain
, unit
.tcp
);
2499 if (ret
== RSC_BYPASS
) {
2500 return virtio_net_do_receive(nc
, buf
, size
);
2501 } else if (ret
== RSC_FINAL
) {
2502 return virtio_net_rsc_drain_flow(chain
, nc
, buf
, size
,
2503 ((hdr_len
+ sizeof(struct eth_header
)) + 8),
2504 VIRTIO_NET_IP6_ADDR_SIZE
,
2505 hdr_len
+ sizeof(struct eth_header
)
2506 + sizeof(struct ip6_header
));
2509 return virtio_net_rsc_do_coalesce(chain
, nc
, buf
, size
, &unit
);
2512 static VirtioNetRscChain
*virtio_net_rsc_lookup_chain(VirtIONet
*n
,
2516 VirtioNetRscChain
*chain
;
2518 if ((proto
!= (uint16_t)ETH_P_IP
) && (proto
!= (uint16_t)ETH_P_IPV6
)) {
2522 QTAILQ_FOREACH(chain
, &n
->rsc_chains
, next
) {
2523 if (chain
->proto
== proto
) {
2528 chain
= g_malloc(sizeof(*chain
));
2530 chain
->proto
= proto
;
2531 if (proto
== (uint16_t)ETH_P_IP
) {
2532 chain
->max_payload
= VIRTIO_NET_MAX_IP4_PAYLOAD
;
2533 chain
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
2535 chain
->max_payload
= VIRTIO_NET_MAX_IP6_PAYLOAD
;
2536 chain
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
2538 chain
->drain_timer
= timer_new_ns(QEMU_CLOCK_HOST
,
2539 virtio_net_rsc_purge
, chain
);
2540 memset(&chain
->stat
, 0, sizeof(chain
->stat
));
2542 QTAILQ_INIT(&chain
->buffers
);
2543 QTAILQ_INSERT_TAIL(&n
->rsc_chains
, chain
, next
);
2548 static ssize_t
virtio_net_rsc_receive(NetClientState
*nc
,
2553 VirtioNetRscChain
*chain
;
2554 struct eth_header
*eth
;
2557 n
= qemu_get_nic_opaque(nc
);
2558 if (size
< (n
->host_hdr_len
+ sizeof(struct eth_header
))) {
2559 return virtio_net_do_receive(nc
, buf
, size
);
2562 eth
= (struct eth_header
*)(buf
+ n
->guest_hdr_len
);
2563 proto
= htons(eth
->h_proto
);
2565 chain
= virtio_net_rsc_lookup_chain(n
, nc
, proto
);
2567 chain
->stat
.received
++;
2568 if (proto
== (uint16_t)ETH_P_IP
&& n
->rsc4_enabled
) {
2569 return virtio_net_rsc_receive4(chain
, nc
, buf
, size
);
2570 } else if (proto
== (uint16_t)ETH_P_IPV6
&& n
->rsc6_enabled
) {
2571 return virtio_net_rsc_receive6(chain
, nc
, buf
, size
);
2574 return virtio_net_do_receive(nc
, buf
, size
);
2577 static ssize_t
virtio_net_receive(NetClientState
*nc
, const uint8_t *buf
,
2580 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
2581 if ((n
->rsc4_enabled
|| n
->rsc6_enabled
)) {
2582 return virtio_net_rsc_receive(nc
, buf
, size
);
2584 return virtio_net_do_receive(nc
, buf
, size
);
2588 static int32_t virtio_net_flush_tx(VirtIONetQueue
*q
);
2590 static void virtio_net_tx_complete(NetClientState
*nc
, ssize_t len
)
2592 VirtIONet
*n
= qemu_get_nic_opaque(nc
);
2593 VirtIONetQueue
*q
= virtio_net_get_subqueue(nc
);
2594 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2597 virtqueue_push(q
->tx_vq
, q
->async_tx
.elem
, 0);
2598 virtio_notify(vdev
, q
->tx_vq
);
2600 g_free(q
->async_tx
.elem
);
2601 q
->async_tx
.elem
= NULL
;
2603 virtio_queue_set_notification(q
->tx_vq
, 1);
2604 ret
= virtio_net_flush_tx(q
);
2605 if (ret
>= n
->tx_burst
) {
2607 * the flush has been stopped by tx_burst
2608 * we will not receive notification for the
2609 * remainining part, so re-schedule
2611 virtio_queue_set_notification(q
->tx_vq
, 0);
2613 qemu_bh_schedule(q
->tx_bh
);
2615 timer_mod(q
->tx_timer
,
2616 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) + n
->tx_timeout
);
2623 static int32_t virtio_net_flush_tx(VirtIONetQueue
*q
)
2625 VirtIONet
*n
= q
->n
;
2626 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2627 VirtQueueElement
*elem
;
2628 int32_t num_packets
= 0;
2629 int queue_index
= vq2q(virtio_get_queue_index(q
->tx_vq
));
2630 if (!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
2634 if (q
->async_tx
.elem
) {
2635 virtio_queue_set_notification(q
->tx_vq
, 0);
2641 unsigned int out_num
;
2642 struct iovec sg
[VIRTQUEUE_MAX_SIZE
], sg2
[VIRTQUEUE_MAX_SIZE
+ 1], *out_sg
;
2643 struct virtio_net_hdr_mrg_rxbuf mhdr
;
2645 elem
= virtqueue_pop(q
->tx_vq
, sizeof(VirtQueueElement
));
2650 out_num
= elem
->out_num
;
2651 out_sg
= elem
->out_sg
;
2653 virtio_error(vdev
, "virtio-net header not in first element");
2654 virtqueue_detach_element(q
->tx_vq
, elem
, 0);
2659 if (n
->has_vnet_hdr
) {
2660 if (iov_to_buf(out_sg
, out_num
, 0, &mhdr
, n
->guest_hdr_len
) <
2662 virtio_error(vdev
, "virtio-net header incorrect");
2663 virtqueue_detach_element(q
->tx_vq
, elem
, 0);
2667 if (n
->needs_vnet_hdr_swap
) {
2668 virtio_net_hdr_swap(vdev
, (void *) &mhdr
);
2669 sg2
[0].iov_base
= &mhdr
;
2670 sg2
[0].iov_len
= n
->guest_hdr_len
;
2671 out_num
= iov_copy(&sg2
[1], ARRAY_SIZE(sg2
) - 1,
2673 n
->guest_hdr_len
, -1);
2674 if (out_num
== VIRTQUEUE_MAX_SIZE
) {
2682 * If host wants to see the guest header as is, we can
2683 * pass it on unchanged. Otherwise, copy just the parts
2684 * that host is interested in.
2686 assert(n
->host_hdr_len
<= n
->guest_hdr_len
);
2687 if (n
->host_hdr_len
!= n
->guest_hdr_len
) {
2688 unsigned sg_num
= iov_copy(sg
, ARRAY_SIZE(sg
),
2690 0, n
->host_hdr_len
);
2691 sg_num
+= iov_copy(sg
+ sg_num
, ARRAY_SIZE(sg
) - sg_num
,
2693 n
->guest_hdr_len
, -1);
2698 ret
= qemu_sendv_packet_async(qemu_get_subqueue(n
->nic
, queue_index
),
2699 out_sg
, out_num
, virtio_net_tx_complete
);
2701 virtio_queue_set_notification(q
->tx_vq
, 0);
2702 q
->async_tx
.elem
= elem
;
2707 virtqueue_push(q
->tx_vq
, elem
, 0);
2708 virtio_notify(vdev
, q
->tx_vq
);
2711 if (++num_packets
>= n
->tx_burst
) {
2718 static void virtio_net_tx_timer(void *opaque
);
2720 static void virtio_net_handle_tx_timer(VirtIODevice
*vdev
, VirtQueue
*vq
)
2722 VirtIONet
*n
= VIRTIO_NET(vdev
);
2723 VirtIONetQueue
*q
= &n
->vqs
[vq2q(virtio_get_queue_index(vq
))];
2725 if (unlikely((n
->status
& VIRTIO_NET_S_LINK_UP
) == 0)) {
2726 virtio_net_drop_tx_queue_data(vdev
, vq
);
2730 /* This happens when device was stopped but VCPU wasn't. */
2731 if (!vdev
->vm_running
) {
2736 if (q
->tx_waiting
) {
2737 /* We already have queued packets, immediately flush */
2738 timer_del(q
->tx_timer
);
2739 virtio_net_tx_timer(q
);
2741 /* re-arm timer to flush it (and more) on next tick */
2742 timer_mod(q
->tx_timer
,
2743 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) + n
->tx_timeout
);
2745 virtio_queue_set_notification(vq
, 0);
2749 static void virtio_net_handle_tx_bh(VirtIODevice
*vdev
, VirtQueue
*vq
)
2751 VirtIONet
*n
= VIRTIO_NET(vdev
);
2752 VirtIONetQueue
*q
= &n
->vqs
[vq2q(virtio_get_queue_index(vq
))];
2754 if (unlikely((n
->status
& VIRTIO_NET_S_LINK_UP
) == 0)) {
2755 virtio_net_drop_tx_queue_data(vdev
, vq
);
2759 if (unlikely(q
->tx_waiting
)) {
2763 /* This happens when device was stopped but VCPU wasn't. */
2764 if (!vdev
->vm_running
) {
2767 virtio_queue_set_notification(vq
, 0);
2768 qemu_bh_schedule(q
->tx_bh
);
2771 static void virtio_net_tx_timer(void *opaque
)
2773 VirtIONetQueue
*q
= opaque
;
2774 VirtIONet
*n
= q
->n
;
2775 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2778 /* This happens when device was stopped but BH wasn't. */
2779 if (!vdev
->vm_running
) {
2780 /* Make sure tx waiting is set, so we'll run when restarted. */
2781 assert(q
->tx_waiting
);
2787 /* Just in case the driver is not ready on more */
2788 if (!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
2792 ret
= virtio_net_flush_tx(q
);
2793 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
2797 * If we flush a full burst of packets, assume there are
2798 * more coming and immediately rearm
2800 if (ret
>= n
->tx_burst
) {
2802 timer_mod(q
->tx_timer
,
2803 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) + n
->tx_timeout
);
2807 * If less than a full burst, re-enable notification and flush
2808 * anything that may have come in while we weren't looking. If
2809 * we find something, assume the guest is still active and rearm
2811 virtio_queue_set_notification(q
->tx_vq
, 1);
2812 ret
= virtio_net_flush_tx(q
);
2814 virtio_queue_set_notification(q
->tx_vq
, 0);
2816 timer_mod(q
->tx_timer
,
2817 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) + n
->tx_timeout
);
2821 static void virtio_net_tx_bh(void *opaque
)
2823 VirtIONetQueue
*q
= opaque
;
2824 VirtIONet
*n
= q
->n
;
2825 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2828 /* This happens when device was stopped but BH wasn't. */
2829 if (!vdev
->vm_running
) {
2830 /* Make sure tx waiting is set, so we'll run when restarted. */
2831 assert(q
->tx_waiting
);
2837 /* Just in case the driver is not ready on more */
2838 if (unlikely(!(vdev
->status
& VIRTIO_CONFIG_S_DRIVER_OK
))) {
2842 ret
= virtio_net_flush_tx(q
);
2843 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
2844 return; /* Notification re-enable handled by tx_complete or device
2848 /* If we flush a full burst of packets, assume there are
2849 * more coming and immediately reschedule */
2850 if (ret
>= n
->tx_burst
) {
2851 qemu_bh_schedule(q
->tx_bh
);
2856 /* If less than a full burst, re-enable notification and flush
2857 * anything that may have come in while we weren't looking. If
2858 * we find something, assume the guest is still active and reschedule */
2859 virtio_queue_set_notification(q
->tx_vq
, 1);
2860 ret
= virtio_net_flush_tx(q
);
2861 if (ret
== -EINVAL
) {
2863 } else if (ret
> 0) {
2864 virtio_queue_set_notification(q
->tx_vq
, 0);
2865 qemu_bh_schedule(q
->tx_bh
);
2870 static void virtio_net_add_queue(VirtIONet
*n
, int index
)
2872 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2874 n
->vqs
[index
].rx_vq
= virtio_add_queue(vdev
, n
->net_conf
.rx_queue_size
,
2875 virtio_net_handle_rx
);
2877 if (n
->net_conf
.tx
&& !strcmp(n
->net_conf
.tx
, "timer")) {
2878 n
->vqs
[index
].tx_vq
=
2879 virtio_add_queue(vdev
, n
->net_conf
.tx_queue_size
,
2880 virtio_net_handle_tx_timer
);
2881 n
->vqs
[index
].tx_timer
= timer_new_ns(QEMU_CLOCK_VIRTUAL
,
2882 virtio_net_tx_timer
,
2885 n
->vqs
[index
].tx_vq
=
2886 virtio_add_queue(vdev
, n
->net_conf
.tx_queue_size
,
2887 virtio_net_handle_tx_bh
);
2888 n
->vqs
[index
].tx_bh
= qemu_bh_new(virtio_net_tx_bh
, &n
->vqs
[index
]);
2891 n
->vqs
[index
].tx_waiting
= 0;
2892 n
->vqs
[index
].n
= n
;
2895 static void virtio_net_del_queue(VirtIONet
*n
, int index
)
2897 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2898 VirtIONetQueue
*q
= &n
->vqs
[index
];
2899 NetClientState
*nc
= qemu_get_subqueue(n
->nic
, index
);
2901 qemu_purge_queued_packets(nc
);
2903 virtio_del_queue(vdev
, index
* 2);
2905 timer_free(q
->tx_timer
);
2908 qemu_bh_delete(q
->tx_bh
);
2912 virtio_del_queue(vdev
, index
* 2 + 1);
2915 static void virtio_net_change_num_queue_pairs(VirtIONet
*n
, int new_max_queue_pairs
)
2917 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2918 int old_num_queues
= virtio_get_num_queues(vdev
);
2919 int new_num_queues
= new_max_queue_pairs
* 2 + 1;
2922 assert(old_num_queues
>= 3);
2923 assert(old_num_queues
% 2 == 1);
2925 if (old_num_queues
== new_num_queues
) {
2930 * We always need to remove and add ctrl vq if
2931 * old_num_queues != new_num_queues. Remove ctrl_vq first,
2932 * and then we only enter one of the following two loops.
2934 virtio_del_queue(vdev
, old_num_queues
- 1);
2936 for (i
= new_num_queues
- 1; i
< old_num_queues
- 1; i
+= 2) {
2937 /* new_num_queues < old_num_queues */
2938 virtio_net_del_queue(n
, i
/ 2);
2941 for (i
= old_num_queues
- 1; i
< new_num_queues
- 1; i
+= 2) {
2942 /* new_num_queues > old_num_queues */
2943 virtio_net_add_queue(n
, i
/ 2);
2946 /* add ctrl_vq last */
2947 n
->ctrl_vq
= virtio_add_queue(vdev
, 64, virtio_net_handle_ctrl
);
2950 static void virtio_net_set_multiqueue(VirtIONet
*n
, int multiqueue
)
2952 int max
= multiqueue
? n
->max_queue_pairs
: 1;
2954 n
->multiqueue
= multiqueue
;
2955 virtio_net_change_num_queue_pairs(n
, max
);
2957 virtio_net_set_queue_pairs(n
);
2960 static int virtio_net_post_load_device(void *opaque
, int version_id
)
2962 VirtIONet
*n
= opaque
;
2963 VirtIODevice
*vdev
= VIRTIO_DEVICE(n
);
2966 trace_virtio_net_post_load_device();
2967 virtio_net_set_mrg_rx_bufs(n
, n
->mergeable_rx_bufs
,
2968 virtio_vdev_has_feature(vdev
,
2969 VIRTIO_F_VERSION_1
),
2970 virtio_vdev_has_feature(vdev
,
2971 VIRTIO_NET_F_HASH_REPORT
));
2973 /* MAC_TABLE_ENTRIES may be different from the saved image */
2974 if (n
->mac_table
.in_use
> MAC_TABLE_ENTRIES
) {
2975 n
->mac_table
.in_use
= 0;
2978 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
)) {
2979 n
->curr_guest_offloads
= virtio_net_supported_guest_offloads(n
);
2983 * curr_guest_offloads will be later overwritten by the
2984 * virtio_set_features_nocheck call done from the virtio_load.
2985 * Here we make sure it is preserved and restored accordingly
2986 * in the virtio_net_post_load_virtio callback.
2988 n
->saved_guest_offloads
= n
->curr_guest_offloads
;
2990 virtio_net_set_queue_pairs(n
);
2992 /* Find the first multicast entry in the saved MAC filter */
2993 for (i
= 0; i
< n
->mac_table
.in_use
; i
++) {
2994 if (n
->mac_table
.macs
[i
* ETH_ALEN
] & 1) {
2998 n
->mac_table
.first_multi
= i
;
3000 /* nc.link_down can't be migrated, so infer link_down according
3001 * to link status bit in n->status */
3002 link_down
= (n
->status
& VIRTIO_NET_S_LINK_UP
) == 0;
3003 for (i
= 0; i
< n
->max_queue_pairs
; i
++) {
3004 qemu_get_subqueue(n
->nic
, i
)->link_down
= link_down
;
3007 if (virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
) &&
3008 virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
)) {
3009 qemu_announce_timer_reset(&n
->announce_timer
, migrate_announce_params(),
3011 virtio_net_announce_timer
, n
);
3012 if (n
->announce_timer
.round
) {
3013 timer_mod(n
->announce_timer
.tm
,
3014 qemu_clock_get_ms(n
->announce_timer
.type
));
3016 qemu_announce_timer_del(&n
->announce_timer
, false);
3020 if (n
->rss_data
.enabled
) {
3021 n
->rss_data
.enabled_software_rss
= n
->rss_data
.populate_hash
;
3022 if (!n
->rss_data
.populate_hash
) {
3023 if (!virtio_net_attach_epbf_rss(n
)) {
3024 if (get_vhost_net(qemu_get_queue(n
->nic
)->peer
)) {
3025 warn_report("Can't post-load eBPF RSS for vhost");
3027 warn_report("Can't post-load eBPF RSS - "
3028 "fallback to software RSS");
3029 n
->rss_data
.enabled_software_rss
= true;
3034 trace_virtio_net_rss_enable(n
->rss_data
.hash_types
,
3035 n
->rss_data
.indirections_len
,
3036 sizeof(n
->rss_data
.key
));
3038 trace_virtio_net_rss_disable();
3043 static int virtio_net_post_load_virtio(VirtIODevice
*vdev
)
3045 VirtIONet
*n
= VIRTIO_NET(vdev
);
3047 * The actual needed state is now in saved_guest_offloads,
3048 * see virtio_net_post_load_device for detail.
3049 * Restore it back and apply the desired offloads.
3051 n
->curr_guest_offloads
= n
->saved_guest_offloads
;
3052 if (peer_has_vnet_hdr(n
)) {
3053 virtio_net_apply_guest_offloads(n
);
3059 /* tx_waiting field of a VirtIONetQueue */
3060 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting
= {
3061 .name
= "virtio-net-queue-tx_waiting",
3062 .fields
= (VMStateField
[]) {
3063 VMSTATE_UINT32(tx_waiting
, VirtIONetQueue
),
3064 VMSTATE_END_OF_LIST()
3068 static bool max_queue_pairs_gt_1(void *opaque
, int version_id
)
3070 return VIRTIO_NET(opaque
)->max_queue_pairs
> 1;
3073 static bool has_ctrl_guest_offloads(void *opaque
, int version_id
)
3075 return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque
),
3076 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
);
3079 static bool mac_table_fits(void *opaque
, int version_id
)
3081 return VIRTIO_NET(opaque
)->mac_table
.in_use
<= MAC_TABLE_ENTRIES
;
3084 static bool mac_table_doesnt_fit(void *opaque
, int version_id
)
3086 return !mac_table_fits(opaque
, version_id
);
3089 /* This temporary type is shared by all the WITH_TMP methods
3090 * although only some fields are used by each.
3092 struct VirtIONetMigTmp
{
3094 VirtIONetQueue
*vqs_1
;
3095 uint16_t curr_queue_pairs_1
;
3097 uint32_t has_vnet_hdr
;
3100 /* The 2nd and subsequent tx_waiting flags are loaded later than
3101 * the 1st entry in the queue_pairs and only if there's more than one
3102 * entry. We use the tmp mechanism to calculate a temporary
3103 * pointer and count and also validate the count.
3106 static int virtio_net_tx_waiting_pre_save(void *opaque
)
3108 struct VirtIONetMigTmp
*tmp
= opaque
;
3110 tmp
->vqs_1
= tmp
->parent
->vqs
+ 1;
3111 tmp
->curr_queue_pairs_1
= tmp
->parent
->curr_queue_pairs
- 1;
3112 if (tmp
->parent
->curr_queue_pairs
== 0) {
3113 tmp
->curr_queue_pairs_1
= 0;
3119 static int virtio_net_tx_waiting_pre_load(void *opaque
)
3121 struct VirtIONetMigTmp
*tmp
= opaque
;
3123 /* Reuse the pointer setup from save */
3124 virtio_net_tx_waiting_pre_save(opaque
);
3126 if (tmp
->parent
->curr_queue_pairs
> tmp
->parent
->max_queue_pairs
) {
3127 error_report("virtio-net: curr_queue_pairs %x > max_queue_pairs %x",
3128 tmp
->parent
->curr_queue_pairs
, tmp
->parent
->max_queue_pairs
);
3133 return 0; /* all good */
3136 static const VMStateDescription vmstate_virtio_net_tx_waiting
= {
3137 .name
= "virtio-net-tx_waiting",
3138 .pre_load
= virtio_net_tx_waiting_pre_load
,
3139 .pre_save
= virtio_net_tx_waiting_pre_save
,
3140 .fields
= (VMStateField
[]) {
3141 VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1
, struct VirtIONetMigTmp
,
3143 vmstate_virtio_net_queue_tx_waiting
,
3144 struct VirtIONetQueue
),
3145 VMSTATE_END_OF_LIST()
3149 /* the 'has_ufo' flag is just tested; if the incoming stream has the
3150 * flag set we need to check that we have it
3152 static int virtio_net_ufo_post_load(void *opaque
, int version_id
)
3154 struct VirtIONetMigTmp
*tmp
= opaque
;
3156 if (tmp
->has_ufo
&& !peer_has_ufo(tmp
->parent
)) {
3157 error_report("virtio-net: saved image requires TUN_F_UFO support");
3164 static int virtio_net_ufo_pre_save(void *opaque
)
3166 struct VirtIONetMigTmp
*tmp
= opaque
;
3168 tmp
->has_ufo
= tmp
->parent
->has_ufo
;
3173 static const VMStateDescription vmstate_virtio_net_has_ufo
= {
3174 .name
= "virtio-net-ufo",
3175 .post_load
= virtio_net_ufo_post_load
,
3176 .pre_save
= virtio_net_ufo_pre_save
,
3177 .fields
= (VMStateField
[]) {
3178 VMSTATE_UINT8(has_ufo
, struct VirtIONetMigTmp
),
3179 VMSTATE_END_OF_LIST()
3183 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
3184 * flag set we need to check that we have it
3186 static int virtio_net_vnet_post_load(void *opaque
, int version_id
)
3188 struct VirtIONetMigTmp
*tmp
= opaque
;
3190 if (tmp
->has_vnet_hdr
&& !peer_has_vnet_hdr(tmp
->parent
)) {
3191 error_report("virtio-net: saved image requires vnet_hdr=on");
3198 static int virtio_net_vnet_pre_save(void *opaque
)
3200 struct VirtIONetMigTmp
*tmp
= opaque
;
3202 tmp
->has_vnet_hdr
= tmp
->parent
->has_vnet_hdr
;
3207 static const VMStateDescription vmstate_virtio_net_has_vnet
= {
3208 .name
= "virtio-net-vnet",
3209 .post_load
= virtio_net_vnet_post_load
,
3210 .pre_save
= virtio_net_vnet_pre_save
,
3211 .fields
= (VMStateField
[]) {
3212 VMSTATE_UINT32(has_vnet_hdr
, struct VirtIONetMigTmp
),
3213 VMSTATE_END_OF_LIST()
3217 static bool virtio_net_rss_needed(void *opaque
)
3219 return VIRTIO_NET(opaque
)->rss_data
.enabled
;
3222 static const VMStateDescription vmstate_virtio_net_rss
= {
3223 .name
= "virtio-net-device/rss",
3225 .minimum_version_id
= 1,
3226 .needed
= virtio_net_rss_needed
,
3227 .fields
= (VMStateField
[]) {
3228 VMSTATE_BOOL(rss_data
.enabled
, VirtIONet
),
3229 VMSTATE_BOOL(rss_data
.redirect
, VirtIONet
),
3230 VMSTATE_BOOL(rss_data
.populate_hash
, VirtIONet
),
3231 VMSTATE_UINT32(rss_data
.hash_types
, VirtIONet
),
3232 VMSTATE_UINT16(rss_data
.indirections_len
, VirtIONet
),
3233 VMSTATE_UINT16(rss_data
.default_queue
, VirtIONet
),
3234 VMSTATE_UINT8_ARRAY(rss_data
.key
, VirtIONet
,
3235 VIRTIO_NET_RSS_MAX_KEY_SIZE
),
3236 VMSTATE_VARRAY_UINT16_ALLOC(rss_data
.indirections_table
, VirtIONet
,
3237 rss_data
.indirections_len
, 0,
3238 vmstate_info_uint16
, uint16_t),
3239 VMSTATE_END_OF_LIST()
3243 static const VMStateDescription vmstate_virtio_net_device
= {
3244 .name
= "virtio-net-device",
3245 .version_id
= VIRTIO_NET_VM_VERSION
,
3246 .minimum_version_id
= VIRTIO_NET_VM_VERSION
,
3247 .post_load
= virtio_net_post_load_device
,
3248 .fields
= (VMStateField
[]) {
3249 VMSTATE_UINT8_ARRAY(mac
, VirtIONet
, ETH_ALEN
),
3250 VMSTATE_STRUCT_POINTER(vqs
, VirtIONet
,
3251 vmstate_virtio_net_queue_tx_waiting
,
3253 VMSTATE_UINT32(mergeable_rx_bufs
, VirtIONet
),
3254 VMSTATE_UINT16(status
, VirtIONet
),
3255 VMSTATE_UINT8(promisc
, VirtIONet
),
3256 VMSTATE_UINT8(allmulti
, VirtIONet
),
3257 VMSTATE_UINT32(mac_table
.in_use
, VirtIONet
),
3259 /* Guarded pair: If it fits we load it, else we throw it away
3260 * - can happen if source has a larger MAC table.; post-load
3261 * sets flags in this case.
3263 VMSTATE_VBUFFER_MULTIPLY(mac_table
.macs
, VirtIONet
,
3264 0, mac_table_fits
, mac_table
.in_use
,
3266 VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet
, mac_table_doesnt_fit
, 0,
3267 mac_table
.in_use
, ETH_ALEN
),
3269 /* Note: This is an array of uint32's that's always been saved as a
3270 * buffer; hold onto your endiannesses; it's actually used as a bitmap
3271 * but based on the uint.
3273 VMSTATE_BUFFER_POINTER_UNSAFE(vlans
, VirtIONet
, 0, MAX_VLAN
>> 3),
3274 VMSTATE_WITH_TMP(VirtIONet
, struct VirtIONetMigTmp
,
3275 vmstate_virtio_net_has_vnet
),
3276 VMSTATE_UINT8(mac_table
.multi_overflow
, VirtIONet
),
3277 VMSTATE_UINT8(mac_table
.uni_overflow
, VirtIONet
),
3278 VMSTATE_UINT8(alluni
, VirtIONet
),
3279 VMSTATE_UINT8(nomulti
, VirtIONet
),
3280 VMSTATE_UINT8(nouni
, VirtIONet
),
3281 VMSTATE_UINT8(nobcast
, VirtIONet
),
3282 VMSTATE_WITH_TMP(VirtIONet
, struct VirtIONetMigTmp
,
3283 vmstate_virtio_net_has_ufo
),
3284 VMSTATE_SINGLE_TEST(max_queue_pairs
, VirtIONet
, max_queue_pairs_gt_1
, 0,
3285 vmstate_info_uint16_equal
, uint16_t),
3286 VMSTATE_UINT16_TEST(curr_queue_pairs
, VirtIONet
, max_queue_pairs_gt_1
),
3287 VMSTATE_WITH_TMP(VirtIONet
, struct VirtIONetMigTmp
,
3288 vmstate_virtio_net_tx_waiting
),
3289 VMSTATE_UINT64_TEST(curr_guest_offloads
, VirtIONet
,
3290 has_ctrl_guest_offloads
),
3291 VMSTATE_END_OF_LIST()
3293 .subsections
= (const VMStateDescription
* []) {
3294 &vmstate_virtio_net_rss
,
3299 static NetClientInfo net_virtio_info
= {
3300 .type
= NET_CLIENT_DRIVER_NIC
,
3301 .size
= sizeof(NICState
),
3302 .can_receive
= virtio_net_can_receive
,
3303 .receive
= virtio_net_receive
,
3304 .link_status_changed
= virtio_net_set_link_status
,
3305 .query_rx_filter
= virtio_net_query_rxfilter
,
3306 .announce
= virtio_net_announce
,
3309 static bool virtio_net_guest_notifier_pending(VirtIODevice
*vdev
, int idx
)
3311 VirtIONet
*n
= VIRTIO_NET(vdev
);
3313 assert(n
->vhost_started
);
3314 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_MQ
) && idx
== 2) {
3315 /* Must guard against invalid features and bogus queue index
3316 * from being set by malicious guest, or penetrated through
3317 * buggy migration stream.
3319 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
)) {
3320 qemu_log_mask(LOG_GUEST_ERROR
,
3321 "%s: bogus vq index ignored\n", __func__
);
3324 nc
= qemu_get_subqueue(n
->nic
, n
->max_queue_pairs
);
3326 nc
= qemu_get_subqueue(n
->nic
, vq2q(idx
));
3329 * Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1
3330 * as the Marco of configure interrupt's IDX, If this driver does not
3331 * support, the function will return false
3334 if (idx
== VIRTIO_CONFIG_IRQ_IDX
) {
3337 return vhost_net_virtqueue_pending(get_vhost_net(nc
->peer
), idx
);
3340 static void virtio_net_guest_notifier_mask(VirtIODevice
*vdev
, int idx
,
3343 VirtIONet
*n
= VIRTIO_NET(vdev
);
3345 assert(n
->vhost_started
);
3346 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_MQ
) && idx
== 2) {
3347 /* Must guard against invalid features and bogus queue index
3348 * from being set by malicious guest, or penetrated through
3349 * buggy migration stream.
3351 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
)) {
3352 qemu_log_mask(LOG_GUEST_ERROR
,
3353 "%s: bogus vq index ignored\n", __func__
);
3356 nc
= qemu_get_subqueue(n
->nic
, n
->max_queue_pairs
);
3358 nc
= qemu_get_subqueue(n
->nic
, vq2q(idx
));
3361 *Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1
3362 * as the Marco of configure interrupt's IDX, If this driver does not
3363 * support, the function will return
3366 if (idx
== VIRTIO_CONFIG_IRQ_IDX
) {
3370 vhost_net_virtqueue_mask(get_vhost_net(nc
->peer
), vdev
, idx
, mask
);
3373 static void virtio_net_set_config_size(VirtIONet
*n
, uint64_t host_features
)
3375 virtio_add_feature(&host_features
, VIRTIO_NET_F_MAC
);
3377 n
->config_size
= virtio_get_config_size(&cfg_size_params
, host_features
);
3380 void virtio_net_set_netclient_name(VirtIONet
*n
, const char *name
,
3384 * The name can be NULL, the netclient name will be type.x.
3386 assert(type
!= NULL
);
3388 g_free(n
->netclient_name
);
3389 g_free(n
->netclient_type
);
3390 n
->netclient_name
= g_strdup(name
);
3391 n
->netclient_type
= g_strdup(type
);
3394 static bool failover_unplug_primary(VirtIONet
*n
, DeviceState
*dev
)
3396 HotplugHandler
*hotplug_ctrl
;
3400 hotplug_ctrl
= qdev_get_hotplug_handler(dev
);
3402 pci_dev
= PCI_DEVICE(dev
);
3403 pci_dev
->partially_hotplugged
= true;
3404 hotplug_handler_unplug_request(hotplug_ctrl
, dev
, &err
);
3406 error_report_err(err
);
3415 static bool failover_replug_primary(VirtIONet
*n
, DeviceState
*dev
,
3419 HotplugHandler
*hotplug_ctrl
;
3420 PCIDevice
*pdev
= PCI_DEVICE(dev
);
3421 BusState
*primary_bus
;
3423 if (!pdev
->partially_hotplugged
) {
3426 primary_bus
= dev
->parent_bus
;
3428 error_setg(errp
, "virtio_net: couldn't find primary bus");
3431 qdev_set_parent_bus(dev
, primary_bus
, &error_abort
);
3432 qatomic_set(&n
->failover_primary_hidden
, false);
3433 hotplug_ctrl
= qdev_get_hotplug_handler(dev
);
3435 hotplug_handler_pre_plug(hotplug_ctrl
, dev
, &err
);
3439 hotplug_handler_plug(hotplug_ctrl
, dev
, &err
);
3441 pdev
->partially_hotplugged
= false;
3444 error_propagate(errp
, err
);
3448 static void virtio_net_handle_migration_primary(VirtIONet
*n
, MigrationState
*s
)
3450 bool should_be_hidden
;
3452 DeviceState
*dev
= failover_find_primary_device(n
);
3458 should_be_hidden
= qatomic_read(&n
->failover_primary_hidden
);
3460 if (migration_in_setup(s
) && !should_be_hidden
) {
3461 if (failover_unplug_primary(n
, dev
)) {
3462 vmstate_unregister(VMSTATE_IF(dev
), qdev_get_vmsd(dev
), dev
);
3463 qapi_event_send_unplug_primary(dev
->id
);
3464 qatomic_set(&n
->failover_primary_hidden
, true);
3466 warn_report("couldn't unplug primary device");
3468 } else if (migration_has_failed(s
)) {
3469 /* We already unplugged the device let's plug it back */
3470 if (!failover_replug_primary(n
, dev
, &err
)) {
3472 error_report_err(err
);
3478 static void virtio_net_migration_state_notifier(Notifier
*notifier
, void *data
)
3480 MigrationState
*s
= data
;
3481 VirtIONet
*n
= container_of(notifier
, VirtIONet
, migration_state
);
3482 virtio_net_handle_migration_primary(n
, s
);
3485 static bool failover_hide_primary_device(DeviceListener
*listener
,
3486 const QDict
*device_opts
,
3490 VirtIONet
*n
= container_of(listener
, VirtIONet
, primary_listener
);
3491 const char *standby_id
;
3497 if (!qdict_haskey(device_opts
, "failover_pair_id")) {
3501 if (!qdict_haskey(device_opts
, "id")) {
3502 error_setg(errp
, "Device with failover_pair_id needs to have id");
3506 standby_id
= qdict_get_str(device_opts
, "failover_pair_id");
3507 if (g_strcmp0(standby_id
, n
->netclient_name
) != 0) {
3512 * The hide helper can be called several times for a given device.
3513 * Check there is only one primary for a virtio-net device but
3514 * don't duplicate the qdict several times if it's called for the same
3517 if (n
->primary_opts
) {
3518 const char *old
, *new;
3519 /* devices with failover_pair_id always have an id */
3520 old
= qdict_get_str(n
->primary_opts
, "id");
3521 new = qdict_get_str(device_opts
, "id");
3522 if (strcmp(old
, new) != 0) {
3523 error_setg(errp
, "Cannot attach more than one primary device to "
3524 "'%s': '%s' and '%s'", n
->netclient_name
, old
, new);
3528 n
->primary_opts
= qdict_clone_shallow(device_opts
);
3529 n
->primary_opts_from_json
= from_json
;
3532 /* failover_primary_hidden is set during feature negotiation */
3533 return qatomic_read(&n
->failover_primary_hidden
);
3536 static void virtio_net_device_realize(DeviceState
*dev
, Error
**errp
)
3538 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
3539 VirtIONet
*n
= VIRTIO_NET(dev
);
3543 if (n
->net_conf
.mtu
) {
3544 n
->host_features
|= (1ULL << VIRTIO_NET_F_MTU
);
3547 if (n
->net_conf
.duplex_str
) {
3548 if (strncmp(n
->net_conf
.duplex_str
, "half", 5) == 0) {
3549 n
->net_conf
.duplex
= DUPLEX_HALF
;
3550 } else if (strncmp(n
->net_conf
.duplex_str
, "full", 5) == 0) {
3551 n
->net_conf
.duplex
= DUPLEX_FULL
;
3553 error_setg(errp
, "'duplex' must be 'half' or 'full'");
3556 n
->host_features
|= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX
);
3558 n
->net_conf
.duplex
= DUPLEX_UNKNOWN
;
3561 if (n
->net_conf
.speed
< SPEED_UNKNOWN
) {
3562 error_setg(errp
, "'speed' must be between 0 and INT_MAX");
3565 if (n
->net_conf
.speed
>= 0) {
3566 n
->host_features
|= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX
);
3570 n
->primary_listener
.hide_device
= failover_hide_primary_device
;
3571 qatomic_set(&n
->failover_primary_hidden
, true);
3572 device_listener_register(&n
->primary_listener
);
3573 n
->migration_state
.notify
= virtio_net_migration_state_notifier
;
3574 add_migration_state_change_notifier(&n
->migration_state
);
3575 n
->host_features
|= (1ULL << VIRTIO_NET_F_STANDBY
);
3578 virtio_net_set_config_size(n
, n
->host_features
);
3579 virtio_init(vdev
, VIRTIO_ID_NET
, n
->config_size
);
3582 * We set a lower limit on RX queue size to what it always was.
3583 * Guests that want a smaller ring can always resize it without
3584 * help from us (using virtio 1 and up).
3586 if (n
->net_conf
.rx_queue_size
< VIRTIO_NET_RX_QUEUE_MIN_SIZE
||
3587 n
->net_conf
.rx_queue_size
> VIRTQUEUE_MAX_SIZE
||
3588 !is_power_of_2(n
->net_conf
.rx_queue_size
)) {
3589 error_setg(errp
, "Invalid rx_queue_size (= %" PRIu16
"), "
3590 "must be a power of 2 between %d and %d.",
3591 n
->net_conf
.rx_queue_size
, VIRTIO_NET_RX_QUEUE_MIN_SIZE
,
3592 VIRTQUEUE_MAX_SIZE
);
3593 virtio_cleanup(vdev
);
3597 if (n
->net_conf
.tx_queue_size
< VIRTIO_NET_TX_QUEUE_MIN_SIZE
||
3598 n
->net_conf
.tx_queue_size
> VIRTQUEUE_MAX_SIZE
||
3599 !is_power_of_2(n
->net_conf
.tx_queue_size
)) {
3600 error_setg(errp
, "Invalid tx_queue_size (= %" PRIu16
"), "
3601 "must be a power of 2 between %d and %d",
3602 n
->net_conf
.tx_queue_size
, VIRTIO_NET_TX_QUEUE_MIN_SIZE
,
3603 VIRTQUEUE_MAX_SIZE
);
3604 virtio_cleanup(vdev
);
3608 n
->max_ncs
= MAX(n
->nic_conf
.peers
.queues
, 1);
3611 * Figure out the datapath queue pairs since the backend could
3612 * provide control queue via peers as well.
3614 if (n
->nic_conf
.peers
.queues
) {
3615 for (i
= 0; i
< n
->max_ncs
; i
++) {
3616 if (n
->nic_conf
.peers
.ncs
[i
]->is_datapath
) {
3617 ++n
->max_queue_pairs
;
3621 n
->max_queue_pairs
= MAX(n
->max_queue_pairs
, 1);
3623 if (n
->max_queue_pairs
* 2 + 1 > VIRTIO_QUEUE_MAX
) {
3624 error_setg(errp
, "Invalid number of queue pairs (= %" PRIu32
"), "
3625 "must be a positive integer less than %d.",
3626 n
->max_queue_pairs
, (VIRTIO_QUEUE_MAX
- 1) / 2);
3627 virtio_cleanup(vdev
);
3630 n
->vqs
= g_new0(VirtIONetQueue
, n
->max_queue_pairs
);
3631 n
->curr_queue_pairs
= 1;
3632 n
->tx_timeout
= n
->net_conf
.txtimer
;
3634 if (n
->net_conf
.tx
&& strcmp(n
->net_conf
.tx
, "timer")
3635 && strcmp(n
->net_conf
.tx
, "bh")) {
3636 warn_report("virtio-net: "
3637 "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
3639 error_printf("Defaulting to \"bh\"");
3642 n
->net_conf
.tx_queue_size
= MIN(virtio_net_max_tx_queue_size(n
),
3643 n
->net_conf
.tx_queue_size
);
3645 for (i
= 0; i
< n
->max_queue_pairs
; i
++) {
3646 virtio_net_add_queue(n
, i
);
3649 n
->ctrl_vq
= virtio_add_queue(vdev
, 64, virtio_net_handle_ctrl
);
3650 qemu_macaddr_default_if_unset(&n
->nic_conf
.macaddr
);
3651 memcpy(&n
->mac
[0], &n
->nic_conf
.macaddr
, sizeof(n
->mac
));
3652 n
->status
= VIRTIO_NET_S_LINK_UP
;
3653 qemu_announce_timer_reset(&n
->announce_timer
, migrate_announce_params(),
3655 virtio_net_announce_timer
, n
);
3656 n
->announce_timer
.round
= 0;
3658 if (n
->netclient_type
) {
3660 * Happen when virtio_net_set_netclient_name has been called.
3662 n
->nic
= qemu_new_nic(&net_virtio_info
, &n
->nic_conf
,
3663 n
->netclient_type
, n
->netclient_name
, n
);
3665 n
->nic
= qemu_new_nic(&net_virtio_info
, &n
->nic_conf
,
3666 object_get_typename(OBJECT(dev
)), dev
->id
, n
);
3669 for (i
= 0; i
< n
->max_queue_pairs
; i
++) {
3670 n
->nic
->ncs
[i
].do_not_pad
= true;
3673 peer_test_vnet_hdr(n
);
3674 if (peer_has_vnet_hdr(n
)) {
3675 for (i
= 0; i
< n
->max_queue_pairs
; i
++) {
3676 qemu_using_vnet_hdr(qemu_get_subqueue(n
->nic
, i
)->peer
, true);
3678 n
->host_hdr_len
= sizeof(struct virtio_net_hdr
);
3680 n
->host_hdr_len
= 0;
3683 qemu_format_nic_info_str(qemu_get_queue(n
->nic
), n
->nic_conf
.macaddr
.a
);
3685 n
->vqs
[0].tx_waiting
= 0;
3686 n
->tx_burst
= n
->net_conf
.txburst
;
3687 virtio_net_set_mrg_rx_bufs(n
, 0, 0, 0);
3688 n
->promisc
= 1; /* for compatibility */
3690 n
->mac_table
.macs
= g_malloc0(MAC_TABLE_ENTRIES
* ETH_ALEN
);
3692 n
->vlans
= g_malloc0(MAX_VLAN
>> 3);
3694 nc
= qemu_get_queue(n
->nic
);
3695 nc
->rxfilter_notify_enabled
= 1;
3697 if (nc
->peer
&& nc
->peer
->info
->type
== NET_CLIENT_DRIVER_VHOST_VDPA
) {
3698 struct virtio_net_config netcfg
= {};
3699 memcpy(&netcfg
.mac
, &n
->nic_conf
.macaddr
, ETH_ALEN
);
3700 vhost_net_set_config(get_vhost_net(nc
->peer
),
3701 (uint8_t *)&netcfg
, 0, ETH_ALEN
, VHOST_SET_CONFIG_TYPE_MASTER
);
3703 QTAILQ_INIT(&n
->rsc_chains
);
3706 net_rx_pkt_init(&n
->rx_pkt
, false);
3708 if (virtio_has_feature(n
->host_features
, VIRTIO_NET_F_RSS
)) {
3709 virtio_net_load_ebpf(n
);
3713 static void virtio_net_device_unrealize(DeviceState
*dev
)
3715 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
3716 VirtIONet
*n
= VIRTIO_NET(dev
);
3717 int i
, max_queue_pairs
;
3719 if (virtio_has_feature(n
->host_features
, VIRTIO_NET_F_RSS
)) {
3720 virtio_net_unload_ebpf(n
);
3723 /* This will stop vhost backend if appropriate. */
3724 virtio_net_set_status(vdev
, 0);
3726 g_free(n
->netclient_name
);
3727 n
->netclient_name
= NULL
;
3728 g_free(n
->netclient_type
);
3729 n
->netclient_type
= NULL
;
3731 g_free(n
->mac_table
.macs
);
3735 qobject_unref(n
->primary_opts
);
3736 device_listener_unregister(&n
->primary_listener
);
3737 remove_migration_state_change_notifier(&n
->migration_state
);
3739 assert(n
->primary_opts
== NULL
);
3742 max_queue_pairs
= n
->multiqueue
? n
->max_queue_pairs
: 1;
3743 for (i
= 0; i
< max_queue_pairs
; i
++) {
3744 virtio_net_del_queue(n
, i
);
3746 /* delete also control vq */
3747 virtio_del_queue(vdev
, max_queue_pairs
* 2);
3748 qemu_announce_timer_del(&n
->announce_timer
, false);
3750 qemu_del_nic(n
->nic
);
3751 virtio_net_rsc_cleanup(n
);
3752 g_free(n
->rss_data
.indirections_table
);
3753 net_rx_pkt_uninit(n
->rx_pkt
);
3754 virtio_cleanup(vdev
);
3757 static void virtio_net_instance_init(Object
*obj
)
3759 VirtIONet
*n
= VIRTIO_NET(obj
);
3762 * The default config_size is sizeof(struct virtio_net_config).
3763 * Can be overriden with virtio_net_set_config_size.
3765 n
->config_size
= sizeof(struct virtio_net_config
);
3766 device_add_bootindex_property(obj
, &n
->nic_conf
.bootindex
,
3767 "bootindex", "/ethernet-phy@0",
3770 ebpf_rss_init(&n
->ebpf_rss
);
3773 static int virtio_net_pre_save(void *opaque
)
3775 VirtIONet
*n
= opaque
;
3777 /* At this point, backend must be stopped, otherwise
3778 * it might keep writing to memory. */
3779 assert(!n
->vhost_started
);
3784 static bool primary_unplug_pending(void *opaque
)
3786 DeviceState
*dev
= opaque
;
3787 DeviceState
*primary
;
3788 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
3789 VirtIONet
*n
= VIRTIO_NET(vdev
);
3791 if (!virtio_vdev_has_feature(vdev
, VIRTIO_NET_F_STANDBY
)) {
3794 primary
= failover_find_primary_device(n
);
3795 return primary
? primary
->pending_deleted_event
: false;
3798 static bool dev_unplug_pending(void *opaque
)
3800 DeviceState
*dev
= opaque
;
3801 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_GET_CLASS(dev
);
3803 return vdc
->primary_unplug_pending(dev
);
3806 static struct vhost_dev
*virtio_net_get_vhost(VirtIODevice
*vdev
)
3808 VirtIONet
*n
= VIRTIO_NET(vdev
);
3809 NetClientState
*nc
= qemu_get_queue(n
->nic
);
3810 struct vhost_net
*net
= get_vhost_net(nc
->peer
);
3814 static const VMStateDescription vmstate_virtio_net
= {
3815 .name
= "virtio-net",
3816 .minimum_version_id
= VIRTIO_NET_VM_VERSION
,
3817 .version_id
= VIRTIO_NET_VM_VERSION
,
3818 .fields
= (VMStateField
[]) {
3819 VMSTATE_VIRTIO_DEVICE
,
3820 VMSTATE_END_OF_LIST()
3822 .pre_save
= virtio_net_pre_save
,
3823 .dev_unplug_pending
= dev_unplug_pending
,
3826 static Property virtio_net_properties
[] = {
3827 DEFINE_PROP_BIT64("csum", VirtIONet
, host_features
,
3828 VIRTIO_NET_F_CSUM
, true),
3829 DEFINE_PROP_BIT64("guest_csum", VirtIONet
, host_features
,
3830 VIRTIO_NET_F_GUEST_CSUM
, true),
3831 DEFINE_PROP_BIT64("gso", VirtIONet
, host_features
, VIRTIO_NET_F_GSO
, true),
3832 DEFINE_PROP_BIT64("guest_tso4", VirtIONet
, host_features
,
3833 VIRTIO_NET_F_GUEST_TSO4
, true),
3834 DEFINE_PROP_BIT64("guest_tso6", VirtIONet
, host_features
,
3835 VIRTIO_NET_F_GUEST_TSO6
, true),
3836 DEFINE_PROP_BIT64("guest_ecn", VirtIONet
, host_features
,
3837 VIRTIO_NET_F_GUEST_ECN
, true),
3838 DEFINE_PROP_BIT64("guest_ufo", VirtIONet
, host_features
,
3839 VIRTIO_NET_F_GUEST_UFO
, true),
3840 DEFINE_PROP_BIT64("guest_announce", VirtIONet
, host_features
,
3841 VIRTIO_NET_F_GUEST_ANNOUNCE
, true),
3842 DEFINE_PROP_BIT64("host_tso4", VirtIONet
, host_features
,
3843 VIRTIO_NET_F_HOST_TSO4
, true),
3844 DEFINE_PROP_BIT64("host_tso6", VirtIONet
, host_features
,
3845 VIRTIO_NET_F_HOST_TSO6
, true),
3846 DEFINE_PROP_BIT64("host_ecn", VirtIONet
, host_features
,
3847 VIRTIO_NET_F_HOST_ECN
, true),
3848 DEFINE_PROP_BIT64("host_ufo", VirtIONet
, host_features
,
3849 VIRTIO_NET_F_HOST_UFO
, true),
3850 DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet
, host_features
,
3851 VIRTIO_NET_F_MRG_RXBUF
, true),
3852 DEFINE_PROP_BIT64("status", VirtIONet
, host_features
,
3853 VIRTIO_NET_F_STATUS
, true),
3854 DEFINE_PROP_BIT64("ctrl_vq", VirtIONet
, host_features
,
3855 VIRTIO_NET_F_CTRL_VQ
, true),
3856 DEFINE_PROP_BIT64("ctrl_rx", VirtIONet
, host_features
,
3857 VIRTIO_NET_F_CTRL_RX
, true),
3858 DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet
, host_features
,
3859 VIRTIO_NET_F_CTRL_VLAN
, true),
3860 DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet
, host_features
,
3861 VIRTIO_NET_F_CTRL_RX_EXTRA
, true),
3862 DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet
, host_features
,
3863 VIRTIO_NET_F_CTRL_MAC_ADDR
, true),
3864 DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet
, host_features
,
3865 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
, true),
3866 DEFINE_PROP_BIT64("mq", VirtIONet
, host_features
, VIRTIO_NET_F_MQ
, false),
3867 DEFINE_PROP_BIT64("rss", VirtIONet
, host_features
,
3868 VIRTIO_NET_F_RSS
, false),
3869 DEFINE_PROP_BIT64("hash", VirtIONet
, host_features
,
3870 VIRTIO_NET_F_HASH_REPORT
, false),
3871 DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet
, host_features
,
3872 VIRTIO_NET_F_RSC_EXT
, false),
3873 DEFINE_PROP_UINT32("rsc_interval", VirtIONet
, rsc_timeout
,
3874 VIRTIO_NET_RSC_DEFAULT_INTERVAL
),
3875 DEFINE_NIC_PROPERTIES(VirtIONet
, nic_conf
),
3876 DEFINE_PROP_UINT32("x-txtimer", VirtIONet
, net_conf
.txtimer
,
3878 DEFINE_PROP_INT32("x-txburst", VirtIONet
, net_conf
.txburst
, TX_BURST
),
3879 DEFINE_PROP_STRING("tx", VirtIONet
, net_conf
.tx
),
3880 DEFINE_PROP_UINT16("rx_queue_size", VirtIONet
, net_conf
.rx_queue_size
,
3881 VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
),
3882 DEFINE_PROP_UINT16("tx_queue_size", VirtIONet
, net_conf
.tx_queue_size
,
3883 VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
),
3884 DEFINE_PROP_UINT16("host_mtu", VirtIONet
, net_conf
.mtu
, 0),
3885 DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet
, mtu_bypass_backend
,
3887 DEFINE_PROP_INT32("speed", VirtIONet
, net_conf
.speed
, SPEED_UNKNOWN
),
3888 DEFINE_PROP_STRING("duplex", VirtIONet
, net_conf
.duplex_str
),
3889 DEFINE_PROP_BOOL("failover", VirtIONet
, failover
, false),
3890 DEFINE_PROP_END_OF_LIST(),
3893 static void virtio_net_class_init(ObjectClass
*klass
, void *data
)
3895 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3896 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
3898 device_class_set_props(dc
, virtio_net_properties
);
3899 dc
->vmsd
= &vmstate_virtio_net
;
3900 set_bit(DEVICE_CATEGORY_NETWORK
, dc
->categories
);
3901 vdc
->realize
= virtio_net_device_realize
;
3902 vdc
->unrealize
= virtio_net_device_unrealize
;
3903 vdc
->get_config
= virtio_net_get_config
;
3904 vdc
->set_config
= virtio_net_set_config
;
3905 vdc
->get_features
= virtio_net_get_features
;
3906 vdc
->set_features
= virtio_net_set_features
;
3907 vdc
->bad_features
= virtio_net_bad_features
;
3908 vdc
->reset
= virtio_net_reset
;
3909 vdc
->queue_reset
= virtio_net_queue_reset
;
3910 vdc
->queue_enable
= virtio_net_queue_enable
;
3911 vdc
->set_status
= virtio_net_set_status
;
3912 vdc
->guest_notifier_mask
= virtio_net_guest_notifier_mask
;
3913 vdc
->guest_notifier_pending
= virtio_net_guest_notifier_pending
;
3914 vdc
->legacy_features
|= (0x1 << VIRTIO_NET_F_GSO
);
3915 vdc
->post_load
= virtio_net_post_load_virtio
;
3916 vdc
->vmsd
= &vmstate_virtio_net_device
;
3917 vdc
->primary_unplug_pending
= primary_unplug_pending
;
3918 vdc
->get_vhost
= virtio_net_get_vhost
;
3921 static const TypeInfo virtio_net_info
= {
3922 .name
= TYPE_VIRTIO_NET
,
3923 .parent
= TYPE_VIRTIO_DEVICE
,
3924 .instance_size
= sizeof(VirtIONet
),
3925 .instance_init
= virtio_net_instance_init
,
3926 .class_init
= virtio_net_class_init
,
3929 static void virtio_register_types(void)
3931 type_register_static(&virtio_net_info
);
3934 type_init(virtio_register_types
)