meson: convert qapi-specific to meson
[qemu.git] / hw / net / virtio-net.c
1 /*
2 * Virtio Network Device
3 *
4 * Copyright IBM, Corp. 2007
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 */
13
14 #include "qemu/osdep.h"
15 #include "qemu/atomic.h"
16 #include "qemu/iov.h"
17 #include "qemu/main-loop.h"
18 #include "qemu/module.h"
19 #include "hw/virtio/virtio.h"
20 #include "net/net.h"
21 #include "net/checksum.h"
22 #include "net/tap.h"
23 #include "qemu/error-report.h"
24 #include "qemu/timer.h"
25 #include "qemu/option.h"
26 #include "qemu/option_int.h"
27 #include "qemu/config-file.h"
28 #include "qapi/qmp/qdict.h"
29 #include "hw/virtio/virtio-net.h"
30 #include "net/vhost_net.h"
31 #include "net/announce.h"
32 #include "hw/virtio/virtio-bus.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-events-net.h"
35 #include "hw/qdev-properties.h"
36 #include "qapi/qapi-types-migration.h"
37 #include "qapi/qapi-events-migration.h"
38 #include "hw/virtio/virtio-access.h"
39 #include "migration/misc.h"
40 #include "standard-headers/linux/ethtool.h"
41 #include "sysemu/sysemu.h"
42 #include "trace.h"
43 #include "monitor/qdev.h"
44 #include "hw/pci/pci.h"
45 #include "net_rx_pkt.h"
46 #include "hw/virtio/vhost.h"
47
48 #define VIRTIO_NET_VM_VERSION 11
49
50 #define MAC_TABLE_ENTRIES 64
51 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */
52
53 /* previously fixed value */
54 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
55 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
56
57 /* for now, only allow larger queues; with virtio-1, guest can downsize */
58 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
59 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
60
61 #define VIRTIO_NET_IP4_ADDR_SIZE 8 /* ipv4 saddr + daddr */
62
63 #define VIRTIO_NET_TCP_FLAG 0x3F
64 #define VIRTIO_NET_TCP_HDR_LENGTH 0xF000
65
66 /* IPv4 max payload, 16 bits in the header */
67 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
68 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
69
70 /* header length value in ip header without option */
71 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
72
73 #define VIRTIO_NET_IP6_ADDR_SIZE 32 /* ipv6 saddr + daddr */
74 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
75
76 /* Purge coalesced packets timer interval, This value affects the performance
77 a lot, and should be tuned carefully, '300000'(300us) is the recommended
78 value to pass the WHQL test, '50000' can gain 2x netperf throughput with
79 tso/gso/gro 'off'. */
80 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
81
82 #define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \
83 VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \
84 VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \
85 VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \
86 VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \
87 VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \
88 VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \
89 VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \
90 VIRTIO_NET_RSS_HASH_TYPE_UDP_EX)
91
92 static VirtIOFeature feature_sizes[] = {
93 {.flags = 1ULL << VIRTIO_NET_F_MAC,
94 .end = endof(struct virtio_net_config, mac)},
95 {.flags = 1ULL << VIRTIO_NET_F_STATUS,
96 .end = endof(struct virtio_net_config, status)},
97 {.flags = 1ULL << VIRTIO_NET_F_MQ,
98 .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
99 {.flags = 1ULL << VIRTIO_NET_F_MTU,
100 .end = endof(struct virtio_net_config, mtu)},
101 {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX,
102 .end = endof(struct virtio_net_config, duplex)},
103 {.flags = (1ULL << VIRTIO_NET_F_RSS) | (1ULL << VIRTIO_NET_F_HASH_REPORT),
104 .end = endof(struct virtio_net_config, supported_hash_types)},
105 {}
106 };
107
108 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
109 {
110 VirtIONet *n = qemu_get_nic_opaque(nc);
111
112 return &n->vqs[nc->queue_index];
113 }
114
115 static int vq2q(int queue_index)
116 {
117 return queue_index / 2;
118 }
119
120 /* TODO
121 * - we could suppress RX interrupt if we were so inclined.
122 */
123
124 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
125 {
126 VirtIONet *n = VIRTIO_NET(vdev);
127 struct virtio_net_config netcfg;
128 NetClientState *nc = qemu_get_queue(n->nic);
129
130 int ret = 0;
131 memset(&netcfg, 0 , sizeof(struct virtio_net_config));
132 virtio_stw_p(vdev, &netcfg.status, n->status);
133 virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues);
134 virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
135 memcpy(netcfg.mac, n->mac, ETH_ALEN);
136 virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed);
137 netcfg.duplex = n->net_conf.duplex;
138 netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE;
139 virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length,
140 virtio_host_has_feature(vdev, VIRTIO_NET_F_RSS) ?
141 VIRTIO_NET_RSS_MAX_TABLE_LEN : 1);
142 virtio_stl_p(vdev, &netcfg.supported_hash_types,
143 VIRTIO_NET_RSS_SUPPORTED_HASHES);
144 memcpy(config, &netcfg, n->config_size);
145
146 /*
147 * Is this VDPA? No peer means not VDPA: there's no way to
148 * disconnect/reconnect a VDPA peer.
149 */
150 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
151 ret = vhost_net_get_config(get_vhost_net(nc->peer), (uint8_t *)&netcfg,
152 n->config_size);
153 if (ret != -1) {
154 memcpy(config, &netcfg, n->config_size);
155 }
156 }
157 }
158
159 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
160 {
161 VirtIONet *n = VIRTIO_NET(vdev);
162 struct virtio_net_config netcfg = {};
163 NetClientState *nc = qemu_get_queue(n->nic);
164
165 memcpy(&netcfg, config, n->config_size);
166
167 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
168 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
169 memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
170 memcpy(n->mac, netcfg.mac, ETH_ALEN);
171 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
172 }
173
174 /*
175 * Is this VDPA? No peer means not VDPA: there's no way to
176 * disconnect/reconnect a VDPA peer.
177 */
178 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
179 vhost_net_set_config(get_vhost_net(nc->peer),
180 (uint8_t *)&netcfg, 0, n->config_size,
181 VHOST_SET_CONFIG_TYPE_MASTER);
182 }
183 }
184
185 static bool virtio_net_started(VirtIONet *n, uint8_t status)
186 {
187 VirtIODevice *vdev = VIRTIO_DEVICE(n);
188 return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
189 (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
190 }
191
192 static void virtio_net_announce_notify(VirtIONet *net)
193 {
194 VirtIODevice *vdev = VIRTIO_DEVICE(net);
195 trace_virtio_net_announce_notify();
196
197 net->status |= VIRTIO_NET_S_ANNOUNCE;
198 virtio_notify_config(vdev);
199 }
200
201 static void virtio_net_announce_timer(void *opaque)
202 {
203 VirtIONet *n = opaque;
204 trace_virtio_net_announce_timer(n->announce_timer.round);
205
206 n->announce_timer.round--;
207 virtio_net_announce_notify(n);
208 }
209
210 static void virtio_net_announce(NetClientState *nc)
211 {
212 VirtIONet *n = qemu_get_nic_opaque(nc);
213 VirtIODevice *vdev = VIRTIO_DEVICE(n);
214
215 /*
216 * Make sure the virtio migration announcement timer isn't running
217 * If it is, let it trigger announcement so that we do not cause
218 * confusion.
219 */
220 if (n->announce_timer.round) {
221 return;
222 }
223
224 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
225 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
226 virtio_net_announce_notify(n);
227 }
228 }
229
230 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
231 {
232 VirtIODevice *vdev = VIRTIO_DEVICE(n);
233 NetClientState *nc = qemu_get_queue(n->nic);
234 int queues = n->multiqueue ? n->max_queues : 1;
235
236 if (!get_vhost_net(nc->peer)) {
237 return;
238 }
239
240 if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
241 !!n->vhost_started) {
242 return;
243 }
244 if (!n->vhost_started) {
245 int r, i;
246
247 if (n->needs_vnet_hdr_swap) {
248 error_report("backend does not support %s vnet headers; "
249 "falling back on userspace virtio",
250 virtio_is_big_endian(vdev) ? "BE" : "LE");
251 return;
252 }
253
254 /* Any packets outstanding? Purge them to avoid touching rings
255 * when vhost is running.
256 */
257 for (i = 0; i < queues; i++) {
258 NetClientState *qnc = qemu_get_subqueue(n->nic, i);
259
260 /* Purge both directions: TX and RX. */
261 qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
262 qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
263 }
264
265 if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
266 r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
267 if (r < 0) {
268 error_report("%uBytes MTU not supported by the backend",
269 n->net_conf.mtu);
270
271 return;
272 }
273 }
274
275 n->vhost_started = 1;
276 r = vhost_net_start(vdev, n->nic->ncs, queues);
277 if (r < 0) {
278 error_report("unable to start vhost net: %d: "
279 "falling back on userspace virtio", -r);
280 n->vhost_started = 0;
281 }
282 } else {
283 vhost_net_stop(vdev, n->nic->ncs, queues);
284 n->vhost_started = 0;
285 }
286 }
287
288 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
289 NetClientState *peer,
290 bool enable)
291 {
292 if (virtio_is_big_endian(vdev)) {
293 return qemu_set_vnet_be(peer, enable);
294 } else {
295 return qemu_set_vnet_le(peer, enable);
296 }
297 }
298
299 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
300 int queues, bool enable)
301 {
302 int i;
303
304 for (i = 0; i < queues; i++) {
305 if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
306 enable) {
307 while (--i >= 0) {
308 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
309 }
310
311 return true;
312 }
313 }
314
315 return false;
316 }
317
318 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
319 {
320 VirtIODevice *vdev = VIRTIO_DEVICE(n);
321 int queues = n->multiqueue ? n->max_queues : 1;
322
323 if (virtio_net_started(n, status)) {
324 /* Before using the device, we tell the network backend about the
325 * endianness to use when parsing vnet headers. If the backend
326 * can't do it, we fallback onto fixing the headers in the core
327 * virtio-net code.
328 */
329 n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
330 queues, true);
331 } else if (virtio_net_started(n, vdev->status)) {
332 /* After using the device, we need to reset the network backend to
333 * the default (guest native endianness), otherwise the guest may
334 * lose network connectivity if it is rebooted into a different
335 * endianness.
336 */
337 virtio_net_set_vnet_endian(vdev, n->nic->ncs, queues, false);
338 }
339 }
340
341 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq)
342 {
343 unsigned int dropped = virtqueue_drop_all(vq);
344 if (dropped) {
345 virtio_notify(vdev, vq);
346 }
347 }
348
349 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
350 {
351 VirtIONet *n = VIRTIO_NET(vdev);
352 VirtIONetQueue *q;
353 int i;
354 uint8_t queue_status;
355
356 virtio_net_vnet_endian_status(n, status);
357 virtio_net_vhost_status(n, status);
358
359 for (i = 0; i < n->max_queues; i++) {
360 NetClientState *ncs = qemu_get_subqueue(n->nic, i);
361 bool queue_started;
362 q = &n->vqs[i];
363
364 if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
365 queue_status = 0;
366 } else {
367 queue_status = status;
368 }
369 queue_started =
370 virtio_net_started(n, queue_status) && !n->vhost_started;
371
372 if (queue_started) {
373 qemu_flush_queued_packets(ncs);
374 }
375
376 if (!q->tx_waiting) {
377 continue;
378 }
379
380 if (queue_started) {
381 if (q->tx_timer) {
382 timer_mod(q->tx_timer,
383 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
384 } else {
385 qemu_bh_schedule(q->tx_bh);
386 }
387 } else {
388 if (q->tx_timer) {
389 timer_del(q->tx_timer);
390 } else {
391 qemu_bh_cancel(q->tx_bh);
392 }
393 if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 &&
394 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) &&
395 vdev->vm_running) {
396 /* if tx is waiting we are likely have some packets in tx queue
397 * and disabled notification */
398 q->tx_waiting = 0;
399 virtio_queue_set_notification(q->tx_vq, 1);
400 virtio_net_drop_tx_queue_data(vdev, q->tx_vq);
401 }
402 }
403 }
404 }
405
406 static void virtio_net_set_link_status(NetClientState *nc)
407 {
408 VirtIONet *n = qemu_get_nic_opaque(nc);
409 VirtIODevice *vdev = VIRTIO_DEVICE(n);
410 uint16_t old_status = n->status;
411
412 if (nc->link_down)
413 n->status &= ~VIRTIO_NET_S_LINK_UP;
414 else
415 n->status |= VIRTIO_NET_S_LINK_UP;
416
417 if (n->status != old_status)
418 virtio_notify_config(vdev);
419
420 virtio_net_set_status(vdev, vdev->status);
421 }
422
423 static void rxfilter_notify(NetClientState *nc)
424 {
425 VirtIONet *n = qemu_get_nic_opaque(nc);
426
427 if (nc->rxfilter_notify_enabled) {
428 char *path = object_get_canonical_path(OBJECT(n->qdev));
429 qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
430 n->netclient_name, path);
431 g_free(path);
432
433 /* disable event notification to avoid events flooding */
434 nc->rxfilter_notify_enabled = 0;
435 }
436 }
437
438 static intList *get_vlan_table(VirtIONet *n)
439 {
440 intList *list, *entry;
441 int i, j;
442
443 list = NULL;
444 for (i = 0; i < MAX_VLAN >> 5; i++) {
445 for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
446 if (n->vlans[i] & (1U << j)) {
447 entry = g_malloc0(sizeof(*entry));
448 entry->value = (i << 5) + j;
449 entry->next = list;
450 list = entry;
451 }
452 }
453 }
454
455 return list;
456 }
457
458 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
459 {
460 VirtIONet *n = qemu_get_nic_opaque(nc);
461 VirtIODevice *vdev = VIRTIO_DEVICE(n);
462 RxFilterInfo *info;
463 strList *str_list, *entry;
464 int i;
465
466 info = g_malloc0(sizeof(*info));
467 info->name = g_strdup(nc->name);
468 info->promiscuous = n->promisc;
469
470 if (n->nouni) {
471 info->unicast = RX_STATE_NONE;
472 } else if (n->alluni) {
473 info->unicast = RX_STATE_ALL;
474 } else {
475 info->unicast = RX_STATE_NORMAL;
476 }
477
478 if (n->nomulti) {
479 info->multicast = RX_STATE_NONE;
480 } else if (n->allmulti) {
481 info->multicast = RX_STATE_ALL;
482 } else {
483 info->multicast = RX_STATE_NORMAL;
484 }
485
486 info->broadcast_allowed = n->nobcast;
487 info->multicast_overflow = n->mac_table.multi_overflow;
488 info->unicast_overflow = n->mac_table.uni_overflow;
489
490 info->main_mac = qemu_mac_strdup_printf(n->mac);
491
492 str_list = NULL;
493 for (i = 0; i < n->mac_table.first_multi; i++) {
494 entry = g_malloc0(sizeof(*entry));
495 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
496 entry->next = str_list;
497 str_list = entry;
498 }
499 info->unicast_table = str_list;
500
501 str_list = NULL;
502 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
503 entry = g_malloc0(sizeof(*entry));
504 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
505 entry->next = str_list;
506 str_list = entry;
507 }
508 info->multicast_table = str_list;
509 info->vlan_table = get_vlan_table(n);
510
511 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
512 info->vlan = RX_STATE_ALL;
513 } else if (!info->vlan_table) {
514 info->vlan = RX_STATE_NONE;
515 } else {
516 info->vlan = RX_STATE_NORMAL;
517 }
518
519 /* enable event notification after query */
520 nc->rxfilter_notify_enabled = 1;
521
522 return info;
523 }
524
525 static void virtio_net_reset(VirtIODevice *vdev)
526 {
527 VirtIONet *n = VIRTIO_NET(vdev);
528 int i;
529
530 /* Reset back to compatibility mode */
531 n->promisc = 1;
532 n->allmulti = 0;
533 n->alluni = 0;
534 n->nomulti = 0;
535 n->nouni = 0;
536 n->nobcast = 0;
537 /* multiqueue is disabled by default */
538 n->curr_queues = 1;
539 timer_del(n->announce_timer.tm);
540 n->announce_timer.round = 0;
541 n->status &= ~VIRTIO_NET_S_ANNOUNCE;
542
543 /* Flush any MAC and VLAN filter table state */
544 n->mac_table.in_use = 0;
545 n->mac_table.first_multi = 0;
546 n->mac_table.multi_overflow = 0;
547 n->mac_table.uni_overflow = 0;
548 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
549 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
550 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
551 memset(n->vlans, 0, MAX_VLAN >> 3);
552
553 /* Flush any async TX */
554 for (i = 0; i < n->max_queues; i++) {
555 NetClientState *nc = qemu_get_subqueue(n->nic, i);
556
557 if (nc->peer) {
558 qemu_flush_or_purge_queued_packets(nc->peer, true);
559 assert(!virtio_net_get_subqueue(nc)->async_tx.elem);
560 }
561 }
562 }
563
564 static void peer_test_vnet_hdr(VirtIONet *n)
565 {
566 NetClientState *nc = qemu_get_queue(n->nic);
567 if (!nc->peer) {
568 return;
569 }
570
571 n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
572 }
573
574 static int peer_has_vnet_hdr(VirtIONet *n)
575 {
576 return n->has_vnet_hdr;
577 }
578
579 static int peer_has_ufo(VirtIONet *n)
580 {
581 if (!peer_has_vnet_hdr(n))
582 return 0;
583
584 n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
585
586 return n->has_ufo;
587 }
588
589 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
590 int version_1, int hash_report)
591 {
592 int i;
593 NetClientState *nc;
594
595 n->mergeable_rx_bufs = mergeable_rx_bufs;
596
597 if (version_1) {
598 n->guest_hdr_len = hash_report ?
599 sizeof(struct virtio_net_hdr_v1_hash) :
600 sizeof(struct virtio_net_hdr_mrg_rxbuf);
601 n->rss_data.populate_hash = !!hash_report;
602 } else {
603 n->guest_hdr_len = n->mergeable_rx_bufs ?
604 sizeof(struct virtio_net_hdr_mrg_rxbuf) :
605 sizeof(struct virtio_net_hdr);
606 }
607
608 for (i = 0; i < n->max_queues; i++) {
609 nc = qemu_get_subqueue(n->nic, i);
610
611 if (peer_has_vnet_hdr(n) &&
612 qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
613 qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
614 n->host_hdr_len = n->guest_hdr_len;
615 }
616 }
617 }
618
619 static int virtio_net_max_tx_queue_size(VirtIONet *n)
620 {
621 NetClientState *peer = n->nic_conf.peers.ncs[0];
622
623 /*
624 * Backends other than vhost-user don't support max queue size.
625 */
626 if (!peer) {
627 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
628 }
629
630 if (peer->info->type != NET_CLIENT_DRIVER_VHOST_USER) {
631 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
632 }
633
634 return VIRTQUEUE_MAX_SIZE;
635 }
636
637 static int peer_attach(VirtIONet *n, int index)
638 {
639 NetClientState *nc = qemu_get_subqueue(n->nic, index);
640
641 if (!nc->peer) {
642 return 0;
643 }
644
645 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
646 vhost_set_vring_enable(nc->peer, 1);
647 }
648
649 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
650 return 0;
651 }
652
653 if (n->max_queues == 1) {
654 return 0;
655 }
656
657 return tap_enable(nc->peer);
658 }
659
660 static int peer_detach(VirtIONet *n, int index)
661 {
662 NetClientState *nc = qemu_get_subqueue(n->nic, index);
663
664 if (!nc->peer) {
665 return 0;
666 }
667
668 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
669 vhost_set_vring_enable(nc->peer, 0);
670 }
671
672 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
673 return 0;
674 }
675
676 return tap_disable(nc->peer);
677 }
678
679 static void virtio_net_set_queues(VirtIONet *n)
680 {
681 int i;
682 int r;
683
684 if (n->nic->peer_deleted) {
685 return;
686 }
687
688 for (i = 0; i < n->max_queues; i++) {
689 if (i < n->curr_queues) {
690 r = peer_attach(n, i);
691 assert(!r);
692 } else {
693 r = peer_detach(n, i);
694 assert(!r);
695 }
696 }
697 }
698
699 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
700
701 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
702 Error **errp)
703 {
704 VirtIONet *n = VIRTIO_NET(vdev);
705 NetClientState *nc = qemu_get_queue(n->nic);
706
707 /* Firstly sync all virtio-net possible supported features */
708 features |= n->host_features;
709
710 virtio_add_feature(&features, VIRTIO_NET_F_MAC);
711
712 if (!peer_has_vnet_hdr(n)) {
713 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
714 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
715 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
716 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
717
718 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
719 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
720 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
721 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
722
723 virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT);
724 }
725
726 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
727 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
728 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
729 }
730
731 if (!get_vhost_net(nc->peer)) {
732 return features;
733 }
734
735 virtio_clear_feature(&features, VIRTIO_NET_F_RSS);
736 virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT);
737 features = vhost_net_get_features(get_vhost_net(nc->peer), features);
738 vdev->backend_features = features;
739
740 if (n->mtu_bypass_backend &&
741 (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) {
742 features |= (1ULL << VIRTIO_NET_F_MTU);
743 }
744
745 return features;
746 }
747
748 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
749 {
750 uint64_t features = 0;
751
752 /* Linux kernel 2.6.25. It understood MAC (as everyone must),
753 * but also these: */
754 virtio_add_feature(&features, VIRTIO_NET_F_MAC);
755 virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
756 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
757 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
758 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
759
760 return features;
761 }
762
763 static void virtio_net_apply_guest_offloads(VirtIONet *n)
764 {
765 qemu_set_offload(qemu_get_queue(n->nic)->peer,
766 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
767 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
768 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
769 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
770 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
771 }
772
773 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
774 {
775 static const uint64_t guest_offloads_mask =
776 (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
777 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
778 (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
779 (1ULL << VIRTIO_NET_F_GUEST_ECN) |
780 (1ULL << VIRTIO_NET_F_GUEST_UFO);
781
782 return guest_offloads_mask & features;
783 }
784
785 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
786 {
787 VirtIODevice *vdev = VIRTIO_DEVICE(n);
788 return virtio_net_guest_offloads_by_features(vdev->guest_features);
789 }
790
791 static void failover_add_primary(VirtIONet *n, Error **errp)
792 {
793 Error *err = NULL;
794
795 if (n->primary_dev) {
796 return;
797 }
798
799 n->primary_device_opts = qemu_opts_find(qemu_find_opts("device"),
800 n->primary_device_id);
801 if (n->primary_device_opts) {
802 n->primary_dev = qdev_device_add(n->primary_device_opts, &err);
803 if (err) {
804 qemu_opts_del(n->primary_device_opts);
805 }
806 if (n->primary_dev) {
807 n->primary_bus = n->primary_dev->parent_bus;
808 if (err) {
809 qdev_unplug(n->primary_dev, &err);
810 qdev_set_id(n->primary_dev, "");
811
812 }
813 }
814 } else {
815 error_setg(errp, "Primary device not found");
816 error_append_hint(errp, "Virtio-net failover will not work. Make "
817 "sure primary device has parameter"
818 " failover_pair_id=<virtio-net-id>\n");
819 }
820 if (err) {
821 error_propagate(errp, err);
822 }
823 }
824
825 static int is_my_primary(void *opaque, QemuOpts *opts, Error **errp)
826 {
827 VirtIONet *n = opaque;
828 int ret = 0;
829
830 const char *standby_id = qemu_opt_get(opts, "failover_pair_id");
831
832 if (standby_id != NULL && (g_strcmp0(standby_id, n->netclient_name) == 0)) {
833 n->primary_device_id = g_strdup(opts->id);
834 ret = 1;
835 }
836
837 return ret;
838 }
839
840 static DeviceState *virtio_net_find_primary(VirtIONet *n, Error **errp)
841 {
842 DeviceState *dev = NULL;
843 Error *err = NULL;
844
845 if (qemu_opts_foreach(qemu_find_opts("device"),
846 is_my_primary, n, &err)) {
847 if (err) {
848 error_propagate(errp, err);
849 return NULL;
850 }
851 if (n->primary_device_id) {
852 dev = qdev_find_recursive(sysbus_get_default(),
853 n->primary_device_id);
854 } else {
855 error_setg(errp, "Primary device id not found");
856 return NULL;
857 }
858 }
859 return dev;
860 }
861
862
863
864 static DeviceState *virtio_connect_failover_devices(VirtIONet *n,
865 DeviceState *dev,
866 Error **errp)
867 {
868 DeviceState *prim_dev = NULL;
869 Error *err = NULL;
870
871 prim_dev = virtio_net_find_primary(n, &err);
872 if (prim_dev) {
873 n->primary_device_id = g_strdup(prim_dev->id);
874 n->primary_device_opts = prim_dev->opts;
875 } else {
876 if (err) {
877 error_propagate(errp, err);
878 }
879 }
880
881 return prim_dev;
882 }
883
884 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
885 {
886 VirtIONet *n = VIRTIO_NET(vdev);
887 Error *err = NULL;
888 int i;
889
890 if (n->mtu_bypass_backend &&
891 !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) {
892 features &= ~(1ULL << VIRTIO_NET_F_MTU);
893 }
894
895 virtio_net_set_multiqueue(n,
896 virtio_has_feature(features, VIRTIO_NET_F_RSS) ||
897 virtio_has_feature(features, VIRTIO_NET_F_MQ));
898
899 virtio_net_set_mrg_rx_bufs(n,
900 virtio_has_feature(features,
901 VIRTIO_NET_F_MRG_RXBUF),
902 virtio_has_feature(features,
903 VIRTIO_F_VERSION_1),
904 virtio_has_feature(features,
905 VIRTIO_NET_F_HASH_REPORT));
906
907 n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
908 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4);
909 n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
910 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6);
911 n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS);
912
913 if (n->has_vnet_hdr) {
914 n->curr_guest_offloads =
915 virtio_net_guest_offloads_by_features(features);
916 virtio_net_apply_guest_offloads(n);
917 }
918
919 for (i = 0; i < n->max_queues; i++) {
920 NetClientState *nc = qemu_get_subqueue(n->nic, i);
921
922 if (!get_vhost_net(nc->peer)) {
923 continue;
924 }
925 vhost_net_ack_features(get_vhost_net(nc->peer), features);
926 }
927
928 if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
929 memset(n->vlans, 0, MAX_VLAN >> 3);
930 } else {
931 memset(n->vlans, 0xff, MAX_VLAN >> 3);
932 }
933
934 if (virtio_has_feature(features, VIRTIO_NET_F_STANDBY)) {
935 qapi_event_send_failover_negotiated(n->netclient_name);
936 atomic_set(&n->primary_should_be_hidden, false);
937 failover_add_primary(n, &err);
938 if (err) {
939 n->primary_dev = virtio_connect_failover_devices(n, n->qdev, &err);
940 if (err) {
941 goto out_err;
942 }
943 failover_add_primary(n, &err);
944 if (err) {
945 goto out_err;
946 }
947 }
948 }
949 return;
950
951 out_err:
952 if (err) {
953 warn_report_err(err);
954 }
955 }
956
957 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
958 struct iovec *iov, unsigned int iov_cnt)
959 {
960 uint8_t on;
961 size_t s;
962 NetClientState *nc = qemu_get_queue(n->nic);
963
964 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
965 if (s != sizeof(on)) {
966 return VIRTIO_NET_ERR;
967 }
968
969 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
970 n->promisc = on;
971 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
972 n->allmulti = on;
973 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
974 n->alluni = on;
975 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
976 n->nomulti = on;
977 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
978 n->nouni = on;
979 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
980 n->nobcast = on;
981 } else {
982 return VIRTIO_NET_ERR;
983 }
984
985 rxfilter_notify(nc);
986
987 return VIRTIO_NET_OK;
988 }
989
990 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
991 struct iovec *iov, unsigned int iov_cnt)
992 {
993 VirtIODevice *vdev = VIRTIO_DEVICE(n);
994 uint64_t offloads;
995 size_t s;
996
997 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
998 return VIRTIO_NET_ERR;
999 }
1000
1001 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
1002 if (s != sizeof(offloads)) {
1003 return VIRTIO_NET_ERR;
1004 }
1005
1006 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
1007 uint64_t supported_offloads;
1008
1009 offloads = virtio_ldq_p(vdev, &offloads);
1010
1011 if (!n->has_vnet_hdr) {
1012 return VIRTIO_NET_ERR;
1013 }
1014
1015 n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
1016 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4);
1017 n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
1018 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6);
1019 virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT);
1020
1021 supported_offloads = virtio_net_supported_guest_offloads(n);
1022 if (offloads & ~supported_offloads) {
1023 return VIRTIO_NET_ERR;
1024 }
1025
1026 n->curr_guest_offloads = offloads;
1027 virtio_net_apply_guest_offloads(n);
1028
1029 return VIRTIO_NET_OK;
1030 } else {
1031 return VIRTIO_NET_ERR;
1032 }
1033 }
1034
1035 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
1036 struct iovec *iov, unsigned int iov_cnt)
1037 {
1038 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1039 struct virtio_net_ctrl_mac mac_data;
1040 size_t s;
1041 NetClientState *nc = qemu_get_queue(n->nic);
1042
1043 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
1044 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
1045 return VIRTIO_NET_ERR;
1046 }
1047 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
1048 assert(s == sizeof(n->mac));
1049 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
1050 rxfilter_notify(nc);
1051
1052 return VIRTIO_NET_OK;
1053 }
1054
1055 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
1056 return VIRTIO_NET_ERR;
1057 }
1058
1059 int in_use = 0;
1060 int first_multi = 0;
1061 uint8_t uni_overflow = 0;
1062 uint8_t multi_overflow = 0;
1063 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1064
1065 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1066 sizeof(mac_data.entries));
1067 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1068 if (s != sizeof(mac_data.entries)) {
1069 goto error;
1070 }
1071 iov_discard_front(&iov, &iov_cnt, s);
1072
1073 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
1074 goto error;
1075 }
1076
1077 if (mac_data.entries <= MAC_TABLE_ENTRIES) {
1078 s = iov_to_buf(iov, iov_cnt, 0, macs,
1079 mac_data.entries * ETH_ALEN);
1080 if (s != mac_data.entries * ETH_ALEN) {
1081 goto error;
1082 }
1083 in_use += mac_data.entries;
1084 } else {
1085 uni_overflow = 1;
1086 }
1087
1088 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
1089
1090 first_multi = in_use;
1091
1092 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1093 sizeof(mac_data.entries));
1094 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1095 if (s != sizeof(mac_data.entries)) {
1096 goto error;
1097 }
1098
1099 iov_discard_front(&iov, &iov_cnt, s);
1100
1101 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
1102 goto error;
1103 }
1104
1105 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
1106 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
1107 mac_data.entries * ETH_ALEN);
1108 if (s != mac_data.entries * ETH_ALEN) {
1109 goto error;
1110 }
1111 in_use += mac_data.entries;
1112 } else {
1113 multi_overflow = 1;
1114 }
1115
1116 n->mac_table.in_use = in_use;
1117 n->mac_table.first_multi = first_multi;
1118 n->mac_table.uni_overflow = uni_overflow;
1119 n->mac_table.multi_overflow = multi_overflow;
1120 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
1121 g_free(macs);
1122 rxfilter_notify(nc);
1123
1124 return VIRTIO_NET_OK;
1125
1126 error:
1127 g_free(macs);
1128 return VIRTIO_NET_ERR;
1129 }
1130
1131 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
1132 struct iovec *iov, unsigned int iov_cnt)
1133 {
1134 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1135 uint16_t vid;
1136 size_t s;
1137 NetClientState *nc = qemu_get_queue(n->nic);
1138
1139 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
1140 vid = virtio_lduw_p(vdev, &vid);
1141 if (s != sizeof(vid)) {
1142 return VIRTIO_NET_ERR;
1143 }
1144
1145 if (vid >= MAX_VLAN)
1146 return VIRTIO_NET_ERR;
1147
1148 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
1149 n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
1150 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
1151 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
1152 else
1153 return VIRTIO_NET_ERR;
1154
1155 rxfilter_notify(nc);
1156
1157 return VIRTIO_NET_OK;
1158 }
1159
1160 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
1161 struct iovec *iov, unsigned int iov_cnt)
1162 {
1163 trace_virtio_net_handle_announce(n->announce_timer.round);
1164 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
1165 n->status & VIRTIO_NET_S_ANNOUNCE) {
1166 n->status &= ~VIRTIO_NET_S_ANNOUNCE;
1167 if (n->announce_timer.round) {
1168 qemu_announce_timer_step(&n->announce_timer);
1169 }
1170 return VIRTIO_NET_OK;
1171 } else {
1172 return VIRTIO_NET_ERR;
1173 }
1174 }
1175
1176 static void virtio_net_disable_rss(VirtIONet *n)
1177 {
1178 if (n->rss_data.enabled) {
1179 trace_virtio_net_rss_disable();
1180 }
1181 n->rss_data.enabled = false;
1182 }
1183
1184 static uint16_t virtio_net_handle_rss(VirtIONet *n,
1185 struct iovec *iov,
1186 unsigned int iov_cnt,
1187 bool do_rss)
1188 {
1189 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1190 struct virtio_net_rss_config cfg;
1191 size_t s, offset = 0, size_get;
1192 uint16_t queues, i;
1193 struct {
1194 uint16_t us;
1195 uint8_t b;
1196 } QEMU_PACKED temp;
1197 const char *err_msg = "";
1198 uint32_t err_value = 0;
1199
1200 if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) {
1201 err_msg = "RSS is not negotiated";
1202 goto error;
1203 }
1204 if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) {
1205 err_msg = "Hash report is not negotiated";
1206 goto error;
1207 }
1208 size_get = offsetof(struct virtio_net_rss_config, indirection_table);
1209 s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get);
1210 if (s != size_get) {
1211 err_msg = "Short command buffer";
1212 err_value = (uint32_t)s;
1213 goto error;
1214 }
1215 n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types);
1216 n->rss_data.indirections_len =
1217 virtio_lduw_p(vdev, &cfg.indirection_table_mask);
1218 n->rss_data.indirections_len++;
1219 if (!do_rss) {
1220 n->rss_data.indirections_len = 1;
1221 }
1222 if (!is_power_of_2(n->rss_data.indirections_len)) {
1223 err_msg = "Invalid size of indirection table";
1224 err_value = n->rss_data.indirections_len;
1225 goto error;
1226 }
1227 if (n->rss_data.indirections_len > VIRTIO_NET_RSS_MAX_TABLE_LEN) {
1228 err_msg = "Too large indirection table";
1229 err_value = n->rss_data.indirections_len;
1230 goto error;
1231 }
1232 n->rss_data.default_queue = do_rss ?
1233 virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0;
1234 if (n->rss_data.default_queue >= n->max_queues) {
1235 err_msg = "Invalid default queue";
1236 err_value = n->rss_data.default_queue;
1237 goto error;
1238 }
1239 offset += size_get;
1240 size_get = sizeof(uint16_t) * n->rss_data.indirections_len;
1241 g_free(n->rss_data.indirections_table);
1242 n->rss_data.indirections_table = g_malloc(size_get);
1243 if (!n->rss_data.indirections_table) {
1244 err_msg = "Can't allocate indirections table";
1245 err_value = n->rss_data.indirections_len;
1246 goto error;
1247 }
1248 s = iov_to_buf(iov, iov_cnt, offset,
1249 n->rss_data.indirections_table, size_get);
1250 if (s != size_get) {
1251 err_msg = "Short indirection table buffer";
1252 err_value = (uint32_t)s;
1253 goto error;
1254 }
1255 for (i = 0; i < n->rss_data.indirections_len; ++i) {
1256 uint16_t val = n->rss_data.indirections_table[i];
1257 n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val);
1258 }
1259 offset += size_get;
1260 size_get = sizeof(temp);
1261 s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get);
1262 if (s != size_get) {
1263 err_msg = "Can't get queues";
1264 err_value = (uint32_t)s;
1265 goto error;
1266 }
1267 queues = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queues;
1268 if (queues == 0 || queues > n->max_queues) {
1269 err_msg = "Invalid number of queues";
1270 err_value = queues;
1271 goto error;
1272 }
1273 if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) {
1274 err_msg = "Invalid key size";
1275 err_value = temp.b;
1276 goto error;
1277 }
1278 if (!temp.b && n->rss_data.hash_types) {
1279 err_msg = "No key provided";
1280 err_value = 0;
1281 goto error;
1282 }
1283 if (!temp.b && !n->rss_data.hash_types) {
1284 virtio_net_disable_rss(n);
1285 return queues;
1286 }
1287 offset += size_get;
1288 size_get = temp.b;
1289 s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get);
1290 if (s != size_get) {
1291 err_msg = "Can get key buffer";
1292 err_value = (uint32_t)s;
1293 goto error;
1294 }
1295 n->rss_data.enabled = true;
1296 trace_virtio_net_rss_enable(n->rss_data.hash_types,
1297 n->rss_data.indirections_len,
1298 temp.b);
1299 return queues;
1300 error:
1301 trace_virtio_net_rss_error(err_msg, err_value);
1302 virtio_net_disable_rss(n);
1303 return 0;
1304 }
1305
1306 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
1307 struct iovec *iov, unsigned int iov_cnt)
1308 {
1309 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1310 uint16_t queues;
1311
1312 virtio_net_disable_rss(n);
1313 if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) {
1314 queues = virtio_net_handle_rss(n, iov, iov_cnt, false);
1315 return queues ? VIRTIO_NET_OK : VIRTIO_NET_ERR;
1316 }
1317 if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
1318 queues = virtio_net_handle_rss(n, iov, iov_cnt, true);
1319 } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
1320 struct virtio_net_ctrl_mq mq;
1321 size_t s;
1322 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) {
1323 return VIRTIO_NET_ERR;
1324 }
1325 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
1326 if (s != sizeof(mq)) {
1327 return VIRTIO_NET_ERR;
1328 }
1329 queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
1330
1331 } else {
1332 return VIRTIO_NET_ERR;
1333 }
1334
1335 if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1336 queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1337 queues > n->max_queues ||
1338 !n->multiqueue) {
1339 return VIRTIO_NET_ERR;
1340 }
1341
1342 n->curr_queues = queues;
1343 /* stop the backend before changing the number of queues to avoid handling a
1344 * disabled queue */
1345 virtio_net_set_status(vdev, vdev->status);
1346 virtio_net_set_queues(n);
1347
1348 return VIRTIO_NET_OK;
1349 }
1350
1351 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
1352 {
1353 VirtIONet *n = VIRTIO_NET(vdev);
1354 struct virtio_net_ctrl_hdr ctrl;
1355 virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
1356 VirtQueueElement *elem;
1357 size_t s;
1358 struct iovec *iov, *iov2;
1359 unsigned int iov_cnt;
1360
1361 for (;;) {
1362 elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
1363 if (!elem) {
1364 break;
1365 }
1366 if (iov_size(elem->in_sg, elem->in_num) < sizeof(status) ||
1367 iov_size(elem->out_sg, elem->out_num) < sizeof(ctrl)) {
1368 virtio_error(vdev, "virtio-net ctrl missing headers");
1369 virtqueue_detach_element(vq, elem, 0);
1370 g_free(elem);
1371 break;
1372 }
1373
1374 iov_cnt = elem->out_num;
1375 iov2 = iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num);
1376 s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
1377 iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
1378 if (s != sizeof(ctrl)) {
1379 status = VIRTIO_NET_ERR;
1380 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
1381 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
1382 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
1383 status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
1384 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
1385 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
1386 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
1387 status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
1388 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
1389 status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
1390 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
1391 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
1392 }
1393
1394 s = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, sizeof(status));
1395 assert(s == sizeof(status));
1396
1397 virtqueue_push(vq, elem, sizeof(status));
1398 virtio_notify(vdev, vq);
1399 g_free(iov2);
1400 g_free(elem);
1401 }
1402 }
1403
1404 /* RX */
1405
1406 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
1407 {
1408 VirtIONet *n = VIRTIO_NET(vdev);
1409 int queue_index = vq2q(virtio_get_queue_index(vq));
1410
1411 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
1412 }
1413
1414 static bool virtio_net_can_receive(NetClientState *nc)
1415 {
1416 VirtIONet *n = qemu_get_nic_opaque(nc);
1417 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1418 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1419
1420 if (!vdev->vm_running) {
1421 return false;
1422 }
1423
1424 if (nc->queue_index >= n->curr_queues) {
1425 return false;
1426 }
1427
1428 if (!virtio_queue_ready(q->rx_vq) ||
1429 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1430 return false;
1431 }
1432
1433 return true;
1434 }
1435
1436 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1437 {
1438 VirtIONet *n = q->n;
1439 if (virtio_queue_empty(q->rx_vq) ||
1440 (n->mergeable_rx_bufs &&
1441 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1442 virtio_queue_set_notification(q->rx_vq, 1);
1443
1444 /* To avoid a race condition where the guest has made some buffers
1445 * available after the above check but before notification was
1446 * enabled, check for available buffers again.
1447 */
1448 if (virtio_queue_empty(q->rx_vq) ||
1449 (n->mergeable_rx_bufs &&
1450 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1451 return 0;
1452 }
1453 }
1454
1455 virtio_queue_set_notification(q->rx_vq, 0);
1456 return 1;
1457 }
1458
1459 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1460 {
1461 virtio_tswap16s(vdev, &hdr->hdr_len);
1462 virtio_tswap16s(vdev, &hdr->gso_size);
1463 virtio_tswap16s(vdev, &hdr->csum_start);
1464 virtio_tswap16s(vdev, &hdr->csum_offset);
1465 }
1466
1467 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1468 * it never finds out that the packets don't have valid checksums. This
1469 * causes dhclient to get upset. Fedora's carried a patch for ages to
1470 * fix this with Xen but it hasn't appeared in an upstream release of
1471 * dhclient yet.
1472 *
1473 * To avoid breaking existing guests, we catch udp packets and add
1474 * checksums. This is terrible but it's better than hacking the guest
1475 * kernels.
1476 *
1477 * N.B. if we introduce a zero-copy API, this operation is no longer free so
1478 * we should provide a mechanism to disable it to avoid polluting the host
1479 * cache.
1480 */
1481 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1482 uint8_t *buf, size_t size)
1483 {
1484 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1485 (size > 27 && size < 1500) && /* normal sized MTU */
1486 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1487 (buf[23] == 17) && /* ip.protocol == UDP */
1488 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1489 net_checksum_calculate(buf, size);
1490 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1491 }
1492 }
1493
1494 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1495 const void *buf, size_t size)
1496 {
1497 if (n->has_vnet_hdr) {
1498 /* FIXME this cast is evil */
1499 void *wbuf = (void *)buf;
1500 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1501 size - n->host_hdr_len);
1502
1503 if (n->needs_vnet_hdr_swap) {
1504 virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1505 }
1506 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1507 } else {
1508 struct virtio_net_hdr hdr = {
1509 .flags = 0,
1510 .gso_type = VIRTIO_NET_HDR_GSO_NONE
1511 };
1512 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1513 }
1514 }
1515
1516 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1517 {
1518 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1519 static const uint8_t vlan[] = {0x81, 0x00};
1520 uint8_t *ptr = (uint8_t *)buf;
1521 int i;
1522
1523 if (n->promisc)
1524 return 1;
1525
1526 ptr += n->host_hdr_len;
1527
1528 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1529 int vid = lduw_be_p(ptr + 14) & 0xfff;
1530 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1531 return 0;
1532 }
1533
1534 if (ptr[0] & 1) { // multicast
1535 if (!memcmp(ptr, bcast, sizeof(bcast))) {
1536 return !n->nobcast;
1537 } else if (n->nomulti) {
1538 return 0;
1539 } else if (n->allmulti || n->mac_table.multi_overflow) {
1540 return 1;
1541 }
1542
1543 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1544 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1545 return 1;
1546 }
1547 }
1548 } else { // unicast
1549 if (n->nouni) {
1550 return 0;
1551 } else if (n->alluni || n->mac_table.uni_overflow) {
1552 return 1;
1553 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1554 return 1;
1555 }
1556
1557 for (i = 0; i < n->mac_table.first_multi; i++) {
1558 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1559 return 1;
1560 }
1561 }
1562 }
1563
1564 return 0;
1565 }
1566
1567 static uint8_t virtio_net_get_hash_type(bool isip4,
1568 bool isip6,
1569 bool isudp,
1570 bool istcp,
1571 uint32_t types)
1572 {
1573 if (isip4) {
1574 if (istcp && (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4)) {
1575 return NetPktRssIpV4Tcp;
1576 }
1577 if (isudp && (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4)) {
1578 return NetPktRssIpV4Udp;
1579 }
1580 if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
1581 return NetPktRssIpV4;
1582 }
1583 } else if (isip6) {
1584 uint32_t mask = VIRTIO_NET_RSS_HASH_TYPE_TCP_EX |
1585 VIRTIO_NET_RSS_HASH_TYPE_TCPv6;
1586
1587 if (istcp && (types & mask)) {
1588 return (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) ?
1589 NetPktRssIpV6TcpEx : NetPktRssIpV6Tcp;
1590 }
1591 mask = VIRTIO_NET_RSS_HASH_TYPE_UDP_EX | VIRTIO_NET_RSS_HASH_TYPE_UDPv6;
1592 if (isudp && (types & mask)) {
1593 return (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) ?
1594 NetPktRssIpV6UdpEx : NetPktRssIpV6Udp;
1595 }
1596 mask = VIRTIO_NET_RSS_HASH_TYPE_IP_EX | VIRTIO_NET_RSS_HASH_TYPE_IPv6;
1597 if (types & mask) {
1598 return (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) ?
1599 NetPktRssIpV6Ex : NetPktRssIpV6;
1600 }
1601 }
1602 return 0xff;
1603 }
1604
1605 static void virtio_set_packet_hash(const uint8_t *buf, uint8_t report,
1606 uint32_t hash)
1607 {
1608 struct virtio_net_hdr_v1_hash *hdr = (void *)buf;
1609 hdr->hash_value = hash;
1610 hdr->hash_report = report;
1611 }
1612
1613 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf,
1614 size_t size)
1615 {
1616 VirtIONet *n = qemu_get_nic_opaque(nc);
1617 unsigned int index = nc->queue_index, new_index = index;
1618 struct NetRxPkt *pkt = n->rx_pkt;
1619 uint8_t net_hash_type;
1620 uint32_t hash;
1621 bool isip4, isip6, isudp, istcp;
1622 static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = {
1623 VIRTIO_NET_HASH_REPORT_IPv4,
1624 VIRTIO_NET_HASH_REPORT_TCPv4,
1625 VIRTIO_NET_HASH_REPORT_TCPv6,
1626 VIRTIO_NET_HASH_REPORT_IPv6,
1627 VIRTIO_NET_HASH_REPORT_IPv6_EX,
1628 VIRTIO_NET_HASH_REPORT_TCPv6_EX,
1629 VIRTIO_NET_HASH_REPORT_UDPv4,
1630 VIRTIO_NET_HASH_REPORT_UDPv6,
1631 VIRTIO_NET_HASH_REPORT_UDPv6_EX
1632 };
1633
1634 net_rx_pkt_set_protocols(pkt, buf + n->host_hdr_len,
1635 size - n->host_hdr_len);
1636 net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
1637 if (isip4 && (net_rx_pkt_get_ip4_info(pkt)->fragment)) {
1638 istcp = isudp = false;
1639 }
1640 if (isip6 && (net_rx_pkt_get_ip6_info(pkt)->fragment)) {
1641 istcp = isudp = false;
1642 }
1643 net_hash_type = virtio_net_get_hash_type(isip4, isip6, isudp, istcp,
1644 n->rss_data.hash_types);
1645 if (net_hash_type > NetPktRssIpV6UdpEx) {
1646 if (n->rss_data.populate_hash) {
1647 virtio_set_packet_hash(buf, VIRTIO_NET_HASH_REPORT_NONE, 0);
1648 }
1649 return n->rss_data.redirect ? n->rss_data.default_queue : -1;
1650 }
1651
1652 hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key);
1653
1654 if (n->rss_data.populate_hash) {
1655 virtio_set_packet_hash(buf, reports[net_hash_type], hash);
1656 }
1657
1658 if (n->rss_data.redirect) {
1659 new_index = hash & (n->rss_data.indirections_len - 1);
1660 new_index = n->rss_data.indirections_table[new_index];
1661 }
1662
1663 return (index == new_index) ? -1 : new_index;
1664 }
1665
1666 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
1667 size_t size, bool no_rss)
1668 {
1669 VirtIONet *n = qemu_get_nic_opaque(nc);
1670 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1671 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1672 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1673 struct virtio_net_hdr_mrg_rxbuf mhdr;
1674 unsigned mhdr_cnt = 0;
1675 size_t offset, i, guest_offset;
1676
1677 if (!virtio_net_can_receive(nc)) {
1678 return -1;
1679 }
1680
1681 if (!no_rss && n->rss_data.enabled) {
1682 int index = virtio_net_process_rss(nc, buf, size);
1683 if (index >= 0) {
1684 NetClientState *nc2 = qemu_get_subqueue(n->nic, index);
1685 return virtio_net_receive_rcu(nc2, buf, size, true);
1686 }
1687 }
1688
1689 /* hdr_len refers to the header we supply to the guest */
1690 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1691 return 0;
1692 }
1693
1694 if (!receive_filter(n, buf, size))
1695 return size;
1696
1697 offset = i = 0;
1698
1699 while (offset < size) {
1700 VirtQueueElement *elem;
1701 int len, total;
1702 const struct iovec *sg;
1703
1704 total = 0;
1705
1706 elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1707 if (!elem) {
1708 if (i) {
1709 virtio_error(vdev, "virtio-net unexpected empty queue: "
1710 "i %zd mergeable %d offset %zd, size %zd, "
1711 "guest hdr len %zd, host hdr len %zd "
1712 "guest features 0x%" PRIx64,
1713 i, n->mergeable_rx_bufs, offset, size,
1714 n->guest_hdr_len, n->host_hdr_len,
1715 vdev->guest_features);
1716 }
1717 return -1;
1718 }
1719
1720 if (elem->in_num < 1) {
1721 virtio_error(vdev,
1722 "virtio-net receive queue contains no in buffers");
1723 virtqueue_detach_element(q->rx_vq, elem, 0);
1724 g_free(elem);
1725 return -1;
1726 }
1727
1728 sg = elem->in_sg;
1729 if (i == 0) {
1730 assert(offset == 0);
1731 if (n->mergeable_rx_bufs) {
1732 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1733 sg, elem->in_num,
1734 offsetof(typeof(mhdr), num_buffers),
1735 sizeof(mhdr.num_buffers));
1736 }
1737
1738 receive_header(n, sg, elem->in_num, buf, size);
1739 if (n->rss_data.populate_hash) {
1740 offset = sizeof(mhdr);
1741 iov_from_buf(sg, elem->in_num, offset,
1742 buf + offset, n->host_hdr_len - sizeof(mhdr));
1743 }
1744 offset = n->host_hdr_len;
1745 total += n->guest_hdr_len;
1746 guest_offset = n->guest_hdr_len;
1747 } else {
1748 guest_offset = 0;
1749 }
1750
1751 /* copy in packet. ugh */
1752 len = iov_from_buf(sg, elem->in_num, guest_offset,
1753 buf + offset, size - offset);
1754 total += len;
1755 offset += len;
1756 /* If buffers can't be merged, at this point we
1757 * must have consumed the complete packet.
1758 * Otherwise, drop it. */
1759 if (!n->mergeable_rx_bufs && offset < size) {
1760 virtqueue_unpop(q->rx_vq, elem, total);
1761 g_free(elem);
1762 return size;
1763 }
1764
1765 /* signal other side */
1766 virtqueue_fill(q->rx_vq, elem, total, i++);
1767 g_free(elem);
1768 }
1769
1770 if (mhdr_cnt) {
1771 virtio_stw_p(vdev, &mhdr.num_buffers, i);
1772 iov_from_buf(mhdr_sg, mhdr_cnt,
1773 0,
1774 &mhdr.num_buffers, sizeof mhdr.num_buffers);
1775 }
1776
1777 virtqueue_flush(q->rx_vq, i);
1778 virtio_notify(vdev, q->rx_vq);
1779
1780 return size;
1781 }
1782
1783 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf,
1784 size_t size)
1785 {
1786 RCU_READ_LOCK_GUARD();
1787
1788 return virtio_net_receive_rcu(nc, buf, size, false);
1789 }
1790
1791 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain,
1792 const uint8_t *buf,
1793 VirtioNetRscUnit *unit)
1794 {
1795 uint16_t ip_hdrlen;
1796 struct ip_header *ip;
1797
1798 ip = (struct ip_header *)(buf + chain->n->guest_hdr_len
1799 + sizeof(struct eth_header));
1800 unit->ip = (void *)ip;
1801 ip_hdrlen = (ip->ip_ver_len & 0xF) << 2;
1802 unit->ip_plen = &ip->ip_len;
1803 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen);
1804 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1805 unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen;
1806 }
1807
1808 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain,
1809 const uint8_t *buf,
1810 VirtioNetRscUnit *unit)
1811 {
1812 struct ip6_header *ip6;
1813
1814 ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len
1815 + sizeof(struct eth_header));
1816 unit->ip = ip6;
1817 unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
1818 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip)
1819 + sizeof(struct ip6_header));
1820 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1821
1822 /* There is a difference between payload lenght in ipv4 and v6,
1823 ip header is excluded in ipv6 */
1824 unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen;
1825 }
1826
1827 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain,
1828 VirtioNetRscSeg *seg)
1829 {
1830 int ret;
1831 struct virtio_net_hdr_v1 *h;
1832
1833 h = (struct virtio_net_hdr_v1 *)seg->buf;
1834 h->flags = 0;
1835 h->gso_type = VIRTIO_NET_HDR_GSO_NONE;
1836
1837 if (seg->is_coalesced) {
1838 h->rsc.segments = seg->packets;
1839 h->rsc.dup_acks = seg->dup_ack;
1840 h->flags = VIRTIO_NET_HDR_F_RSC_INFO;
1841 if (chain->proto == ETH_P_IP) {
1842 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1843 } else {
1844 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1845 }
1846 }
1847
1848 ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size);
1849 QTAILQ_REMOVE(&chain->buffers, seg, next);
1850 g_free(seg->buf);
1851 g_free(seg);
1852
1853 return ret;
1854 }
1855
1856 static void virtio_net_rsc_purge(void *opq)
1857 {
1858 VirtioNetRscSeg *seg, *rn;
1859 VirtioNetRscChain *chain = (VirtioNetRscChain *)opq;
1860
1861 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) {
1862 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
1863 chain->stat.purge_failed++;
1864 continue;
1865 }
1866 }
1867
1868 chain->stat.timer++;
1869 if (!QTAILQ_EMPTY(&chain->buffers)) {
1870 timer_mod(chain->drain_timer,
1871 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
1872 }
1873 }
1874
1875 static void virtio_net_rsc_cleanup(VirtIONet *n)
1876 {
1877 VirtioNetRscChain *chain, *rn_chain;
1878 VirtioNetRscSeg *seg, *rn_seg;
1879
1880 QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) {
1881 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) {
1882 QTAILQ_REMOVE(&chain->buffers, seg, next);
1883 g_free(seg->buf);
1884 g_free(seg);
1885 }
1886
1887 timer_del(chain->drain_timer);
1888 timer_free(chain->drain_timer);
1889 QTAILQ_REMOVE(&n->rsc_chains, chain, next);
1890 g_free(chain);
1891 }
1892 }
1893
1894 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain,
1895 NetClientState *nc,
1896 const uint8_t *buf, size_t size)
1897 {
1898 uint16_t hdr_len;
1899 VirtioNetRscSeg *seg;
1900
1901 hdr_len = chain->n->guest_hdr_len;
1902 seg = g_malloc(sizeof(VirtioNetRscSeg));
1903 seg->buf = g_malloc(hdr_len + sizeof(struct eth_header)
1904 + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD);
1905 memcpy(seg->buf, buf, size);
1906 seg->size = size;
1907 seg->packets = 1;
1908 seg->dup_ack = 0;
1909 seg->is_coalesced = 0;
1910 seg->nc = nc;
1911
1912 QTAILQ_INSERT_TAIL(&chain->buffers, seg, next);
1913 chain->stat.cache++;
1914
1915 switch (chain->proto) {
1916 case ETH_P_IP:
1917 virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
1918 break;
1919 case ETH_P_IPV6:
1920 virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit);
1921 break;
1922 default:
1923 g_assert_not_reached();
1924 }
1925 }
1926
1927 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain,
1928 VirtioNetRscSeg *seg,
1929 const uint8_t *buf,
1930 struct tcp_header *n_tcp,
1931 struct tcp_header *o_tcp)
1932 {
1933 uint32_t nack, oack;
1934 uint16_t nwin, owin;
1935
1936 nack = htonl(n_tcp->th_ack);
1937 nwin = htons(n_tcp->th_win);
1938 oack = htonl(o_tcp->th_ack);
1939 owin = htons(o_tcp->th_win);
1940
1941 if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) {
1942 chain->stat.ack_out_of_win++;
1943 return RSC_FINAL;
1944 } else if (nack == oack) {
1945 /* duplicated ack or window probe */
1946 if (nwin == owin) {
1947 /* duplicated ack, add dup ack count due to whql test up to 1 */
1948 chain->stat.dup_ack++;
1949 return RSC_FINAL;
1950 } else {
1951 /* Coalesce window update */
1952 o_tcp->th_win = n_tcp->th_win;
1953 chain->stat.win_update++;
1954 return RSC_COALESCE;
1955 }
1956 } else {
1957 /* pure ack, go to 'C', finalize*/
1958 chain->stat.pure_ack++;
1959 return RSC_FINAL;
1960 }
1961 }
1962
1963 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain,
1964 VirtioNetRscSeg *seg,
1965 const uint8_t *buf,
1966 VirtioNetRscUnit *n_unit)
1967 {
1968 void *data;
1969 uint16_t o_ip_len;
1970 uint32_t nseq, oseq;
1971 VirtioNetRscUnit *o_unit;
1972
1973 o_unit = &seg->unit;
1974 o_ip_len = htons(*o_unit->ip_plen);
1975 nseq = htonl(n_unit->tcp->th_seq);
1976 oseq = htonl(o_unit->tcp->th_seq);
1977
1978 /* out of order or retransmitted. */
1979 if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) {
1980 chain->stat.data_out_of_win++;
1981 return RSC_FINAL;
1982 }
1983
1984 data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen;
1985 if (nseq == oseq) {
1986 if ((o_unit->payload == 0) && n_unit->payload) {
1987 /* From no payload to payload, normal case, not a dup ack or etc */
1988 chain->stat.data_after_pure_ack++;
1989 goto coalesce;
1990 } else {
1991 return virtio_net_rsc_handle_ack(chain, seg, buf,
1992 n_unit->tcp, o_unit->tcp);
1993 }
1994 } else if ((nseq - oseq) != o_unit->payload) {
1995 /* Not a consistent packet, out of order */
1996 chain->stat.data_out_of_order++;
1997 return RSC_FINAL;
1998 } else {
1999 coalesce:
2000 if ((o_ip_len + n_unit->payload) > chain->max_payload) {
2001 chain->stat.over_size++;
2002 return RSC_FINAL;
2003 }
2004
2005 /* Here comes the right data, the payload length in v4/v6 is different,
2006 so use the field value to update and record the new data len */
2007 o_unit->payload += n_unit->payload; /* update new data len */
2008
2009 /* update field in ip header */
2010 *o_unit->ip_plen = htons(o_ip_len + n_unit->payload);
2011
2012 /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
2013 for windows guest, while this may change the behavior for linux
2014 guest (only if it uses RSC feature). */
2015 o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags;
2016
2017 o_unit->tcp->th_ack = n_unit->tcp->th_ack;
2018 o_unit->tcp->th_win = n_unit->tcp->th_win;
2019
2020 memmove(seg->buf + seg->size, data, n_unit->payload);
2021 seg->size += n_unit->payload;
2022 seg->packets++;
2023 chain->stat.coalesced++;
2024 return RSC_COALESCE;
2025 }
2026 }
2027
2028 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain,
2029 VirtioNetRscSeg *seg,
2030 const uint8_t *buf, size_t size,
2031 VirtioNetRscUnit *unit)
2032 {
2033 struct ip_header *ip1, *ip2;
2034
2035 ip1 = (struct ip_header *)(unit->ip);
2036 ip2 = (struct ip_header *)(seg->unit.ip);
2037 if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst)
2038 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2039 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2040 chain->stat.no_match++;
2041 return RSC_NO_MATCH;
2042 }
2043
2044 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2045 }
2046
2047 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain,
2048 VirtioNetRscSeg *seg,
2049 const uint8_t *buf, size_t size,
2050 VirtioNetRscUnit *unit)
2051 {
2052 struct ip6_header *ip1, *ip2;
2053
2054 ip1 = (struct ip6_header *)(unit->ip);
2055 ip2 = (struct ip6_header *)(seg->unit.ip);
2056 if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address))
2057 || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address))
2058 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2059 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2060 chain->stat.no_match++;
2061 return RSC_NO_MATCH;
2062 }
2063
2064 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2065 }
2066
2067 /* Packets with 'SYN' should bypass, other flag should be sent after drain
2068 * to prevent out of order */
2069 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain,
2070 struct tcp_header *tcp)
2071 {
2072 uint16_t tcp_hdr;
2073 uint16_t tcp_flag;
2074
2075 tcp_flag = htons(tcp->th_offset_flags);
2076 tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10;
2077 tcp_flag &= VIRTIO_NET_TCP_FLAG;
2078 tcp_flag = htons(tcp->th_offset_flags) & 0x3F;
2079 if (tcp_flag & TH_SYN) {
2080 chain->stat.tcp_syn++;
2081 return RSC_BYPASS;
2082 }
2083
2084 if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) {
2085 chain->stat.tcp_ctrl_drain++;
2086 return RSC_FINAL;
2087 }
2088
2089 if (tcp_hdr > sizeof(struct tcp_header)) {
2090 chain->stat.tcp_all_opt++;
2091 return RSC_FINAL;
2092 }
2093
2094 return RSC_CANDIDATE;
2095 }
2096
2097 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain,
2098 NetClientState *nc,
2099 const uint8_t *buf, size_t size,
2100 VirtioNetRscUnit *unit)
2101 {
2102 int ret;
2103 VirtioNetRscSeg *seg, *nseg;
2104
2105 if (QTAILQ_EMPTY(&chain->buffers)) {
2106 chain->stat.empty_cache++;
2107 virtio_net_rsc_cache_buf(chain, nc, buf, size);
2108 timer_mod(chain->drain_timer,
2109 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
2110 return size;
2111 }
2112
2113 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2114 if (chain->proto == ETH_P_IP) {
2115 ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
2116 } else {
2117 ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit);
2118 }
2119
2120 if (ret == RSC_FINAL) {
2121 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2122 /* Send failed */
2123 chain->stat.final_failed++;
2124 return 0;
2125 }
2126
2127 /* Send current packet */
2128 return virtio_net_do_receive(nc, buf, size);
2129 } else if (ret == RSC_NO_MATCH) {
2130 continue;
2131 } else {
2132 /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
2133 seg->is_coalesced = 1;
2134 return size;
2135 }
2136 }
2137
2138 chain->stat.no_match_cache++;
2139 virtio_net_rsc_cache_buf(chain, nc, buf, size);
2140 return size;
2141 }
2142
2143 /* Drain a connection data, this is to avoid out of order segments */
2144 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain,
2145 NetClientState *nc,
2146 const uint8_t *buf, size_t size,
2147 uint16_t ip_start, uint16_t ip_size,
2148 uint16_t tcp_port)
2149 {
2150 VirtioNetRscSeg *seg, *nseg;
2151 uint32_t ppair1, ppair2;
2152
2153 ppair1 = *(uint32_t *)(buf + tcp_port);
2154 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2155 ppair2 = *(uint32_t *)(seg->buf + tcp_port);
2156 if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size)
2157 || (ppair1 != ppair2)) {
2158 continue;
2159 }
2160 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2161 chain->stat.drain_failed++;
2162 }
2163
2164 break;
2165 }
2166
2167 return virtio_net_do_receive(nc, buf, size);
2168 }
2169
2170 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain,
2171 struct ip_header *ip,
2172 const uint8_t *buf, size_t size)
2173 {
2174 uint16_t ip_len;
2175
2176 /* Not an ipv4 packet */
2177 if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) {
2178 chain->stat.ip_option++;
2179 return RSC_BYPASS;
2180 }
2181
2182 /* Don't handle packets with ip option */
2183 if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) {
2184 chain->stat.ip_option++;
2185 return RSC_BYPASS;
2186 }
2187
2188 if (ip->ip_p != IPPROTO_TCP) {
2189 chain->stat.bypass_not_tcp++;
2190 return RSC_BYPASS;
2191 }
2192
2193 /* Don't handle packets with ip fragment */
2194 if (!(htons(ip->ip_off) & IP_DF)) {
2195 chain->stat.ip_frag++;
2196 return RSC_BYPASS;
2197 }
2198
2199 /* Don't handle packets with ecn flag */
2200 if (IPTOS_ECN(ip->ip_tos)) {
2201 chain->stat.ip_ecn++;
2202 return RSC_BYPASS;
2203 }
2204
2205 ip_len = htons(ip->ip_len);
2206 if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header))
2207 || ip_len > (size - chain->n->guest_hdr_len -
2208 sizeof(struct eth_header))) {
2209 chain->stat.ip_hacked++;
2210 return RSC_BYPASS;
2211 }
2212
2213 return RSC_CANDIDATE;
2214 }
2215
2216 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain,
2217 NetClientState *nc,
2218 const uint8_t *buf, size_t size)
2219 {
2220 int32_t ret;
2221 uint16_t hdr_len;
2222 VirtioNetRscUnit unit;
2223
2224 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2225
2226 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)
2227 + sizeof(struct tcp_header))) {
2228 chain->stat.bypass_not_tcp++;
2229 return virtio_net_do_receive(nc, buf, size);
2230 }
2231
2232 virtio_net_rsc_extract_unit4(chain, buf, &unit);
2233 if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size)
2234 != RSC_CANDIDATE) {
2235 return virtio_net_do_receive(nc, buf, size);
2236 }
2237
2238 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2239 if (ret == RSC_BYPASS) {
2240 return virtio_net_do_receive(nc, buf, size);
2241 } else if (ret == RSC_FINAL) {
2242 return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2243 ((hdr_len + sizeof(struct eth_header)) + 12),
2244 VIRTIO_NET_IP4_ADDR_SIZE,
2245 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header));
2246 }
2247
2248 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2249 }
2250
2251 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain,
2252 struct ip6_header *ip6,
2253 const uint8_t *buf, size_t size)
2254 {
2255 uint16_t ip_len;
2256
2257 if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4)
2258 != IP_HEADER_VERSION_6) {
2259 return RSC_BYPASS;
2260 }
2261
2262 /* Both option and protocol is checked in this */
2263 if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) {
2264 chain->stat.bypass_not_tcp++;
2265 return RSC_BYPASS;
2266 }
2267
2268 ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
2269 if (ip_len < sizeof(struct tcp_header) ||
2270 ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header)
2271 - sizeof(struct ip6_header))) {
2272 chain->stat.ip_hacked++;
2273 return RSC_BYPASS;
2274 }
2275
2276 /* Don't handle packets with ecn flag */
2277 if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) {
2278 chain->stat.ip_ecn++;
2279 return RSC_BYPASS;
2280 }
2281
2282 return RSC_CANDIDATE;
2283 }
2284
2285 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc,
2286 const uint8_t *buf, size_t size)
2287 {
2288 int32_t ret;
2289 uint16_t hdr_len;
2290 VirtioNetRscChain *chain;
2291 VirtioNetRscUnit unit;
2292
2293 chain = (VirtioNetRscChain *)opq;
2294 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2295
2296 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header)
2297 + sizeof(tcp_header))) {
2298 return virtio_net_do_receive(nc, buf, size);
2299 }
2300
2301 virtio_net_rsc_extract_unit6(chain, buf, &unit);
2302 if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain,
2303 unit.ip, buf, size)) {
2304 return virtio_net_do_receive(nc, buf, size);
2305 }
2306
2307 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2308 if (ret == RSC_BYPASS) {
2309 return virtio_net_do_receive(nc, buf, size);
2310 } else if (ret == RSC_FINAL) {
2311 return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2312 ((hdr_len + sizeof(struct eth_header)) + 8),
2313 VIRTIO_NET_IP6_ADDR_SIZE,
2314 hdr_len + sizeof(struct eth_header)
2315 + sizeof(struct ip6_header));
2316 }
2317
2318 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2319 }
2320
2321 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n,
2322 NetClientState *nc,
2323 uint16_t proto)
2324 {
2325 VirtioNetRscChain *chain;
2326
2327 if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) {
2328 return NULL;
2329 }
2330
2331 QTAILQ_FOREACH(chain, &n->rsc_chains, next) {
2332 if (chain->proto == proto) {
2333 return chain;
2334 }
2335 }
2336
2337 chain = g_malloc(sizeof(*chain));
2338 chain->n = n;
2339 chain->proto = proto;
2340 if (proto == (uint16_t)ETH_P_IP) {
2341 chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
2342 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2343 } else {
2344 chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD;
2345 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2346 }
2347 chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST,
2348 virtio_net_rsc_purge, chain);
2349 memset(&chain->stat, 0, sizeof(chain->stat));
2350
2351 QTAILQ_INIT(&chain->buffers);
2352 QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next);
2353
2354 return chain;
2355 }
2356
2357 static ssize_t virtio_net_rsc_receive(NetClientState *nc,
2358 const uint8_t *buf,
2359 size_t size)
2360 {
2361 uint16_t proto;
2362 VirtioNetRscChain *chain;
2363 struct eth_header *eth;
2364 VirtIONet *n;
2365
2366 n = qemu_get_nic_opaque(nc);
2367 if (size < (n->host_hdr_len + sizeof(struct eth_header))) {
2368 return virtio_net_do_receive(nc, buf, size);
2369 }
2370
2371 eth = (struct eth_header *)(buf + n->guest_hdr_len);
2372 proto = htons(eth->h_proto);
2373
2374 chain = virtio_net_rsc_lookup_chain(n, nc, proto);
2375 if (chain) {
2376 chain->stat.received++;
2377 if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) {
2378 return virtio_net_rsc_receive4(chain, nc, buf, size);
2379 } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) {
2380 return virtio_net_rsc_receive6(chain, nc, buf, size);
2381 }
2382 }
2383 return virtio_net_do_receive(nc, buf, size);
2384 }
2385
2386 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf,
2387 size_t size)
2388 {
2389 VirtIONet *n = qemu_get_nic_opaque(nc);
2390 if ((n->rsc4_enabled || n->rsc6_enabled)) {
2391 return virtio_net_rsc_receive(nc, buf, size);
2392 } else {
2393 return virtio_net_do_receive(nc, buf, size);
2394 }
2395 }
2396
2397 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
2398
2399 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
2400 {
2401 VirtIONet *n = qemu_get_nic_opaque(nc);
2402 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
2403 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2404
2405 virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
2406 virtio_notify(vdev, q->tx_vq);
2407
2408 g_free(q->async_tx.elem);
2409 q->async_tx.elem = NULL;
2410
2411 virtio_queue_set_notification(q->tx_vq, 1);
2412 virtio_net_flush_tx(q);
2413 }
2414
2415 /* TX */
2416 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
2417 {
2418 VirtIONet *n = q->n;
2419 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2420 VirtQueueElement *elem;
2421 int32_t num_packets = 0;
2422 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
2423 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2424 return num_packets;
2425 }
2426
2427 if (q->async_tx.elem) {
2428 virtio_queue_set_notification(q->tx_vq, 0);
2429 return num_packets;
2430 }
2431
2432 for (;;) {
2433 ssize_t ret;
2434 unsigned int out_num;
2435 struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
2436 struct virtio_net_hdr_mrg_rxbuf mhdr;
2437
2438 elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
2439 if (!elem) {
2440 break;
2441 }
2442
2443 out_num = elem->out_num;
2444 out_sg = elem->out_sg;
2445 if (out_num < 1) {
2446 virtio_error(vdev, "virtio-net header not in first element");
2447 virtqueue_detach_element(q->tx_vq, elem, 0);
2448 g_free(elem);
2449 return -EINVAL;
2450 }
2451
2452 if (n->has_vnet_hdr) {
2453 if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
2454 n->guest_hdr_len) {
2455 virtio_error(vdev, "virtio-net header incorrect");
2456 virtqueue_detach_element(q->tx_vq, elem, 0);
2457 g_free(elem);
2458 return -EINVAL;
2459 }
2460 if (n->needs_vnet_hdr_swap) {
2461 virtio_net_hdr_swap(vdev, (void *) &mhdr);
2462 sg2[0].iov_base = &mhdr;
2463 sg2[0].iov_len = n->guest_hdr_len;
2464 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
2465 out_sg, out_num,
2466 n->guest_hdr_len, -1);
2467 if (out_num == VIRTQUEUE_MAX_SIZE) {
2468 goto drop;
2469 }
2470 out_num += 1;
2471 out_sg = sg2;
2472 }
2473 }
2474 /*
2475 * If host wants to see the guest header as is, we can
2476 * pass it on unchanged. Otherwise, copy just the parts
2477 * that host is interested in.
2478 */
2479 assert(n->host_hdr_len <= n->guest_hdr_len);
2480 if (n->host_hdr_len != n->guest_hdr_len) {
2481 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
2482 out_sg, out_num,
2483 0, n->host_hdr_len);
2484 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
2485 out_sg, out_num,
2486 n->guest_hdr_len, -1);
2487 out_num = sg_num;
2488 out_sg = sg;
2489 }
2490
2491 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
2492 out_sg, out_num, virtio_net_tx_complete);
2493 if (ret == 0) {
2494 virtio_queue_set_notification(q->tx_vq, 0);
2495 q->async_tx.elem = elem;
2496 return -EBUSY;
2497 }
2498
2499 drop:
2500 virtqueue_push(q->tx_vq, elem, 0);
2501 virtio_notify(vdev, q->tx_vq);
2502 g_free(elem);
2503
2504 if (++num_packets >= n->tx_burst) {
2505 break;
2506 }
2507 }
2508 return num_packets;
2509 }
2510
2511 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
2512 {
2513 VirtIONet *n = VIRTIO_NET(vdev);
2514 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2515
2516 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2517 virtio_net_drop_tx_queue_data(vdev, vq);
2518 return;
2519 }
2520
2521 /* This happens when device was stopped but VCPU wasn't. */
2522 if (!vdev->vm_running) {
2523 q->tx_waiting = 1;
2524 return;
2525 }
2526
2527 if (q->tx_waiting) {
2528 virtio_queue_set_notification(vq, 1);
2529 timer_del(q->tx_timer);
2530 q->tx_waiting = 0;
2531 if (virtio_net_flush_tx(q) == -EINVAL) {
2532 return;
2533 }
2534 } else {
2535 timer_mod(q->tx_timer,
2536 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2537 q->tx_waiting = 1;
2538 virtio_queue_set_notification(vq, 0);
2539 }
2540 }
2541
2542 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
2543 {
2544 VirtIONet *n = VIRTIO_NET(vdev);
2545 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2546
2547 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2548 virtio_net_drop_tx_queue_data(vdev, vq);
2549 return;
2550 }
2551
2552 if (unlikely(q->tx_waiting)) {
2553 return;
2554 }
2555 q->tx_waiting = 1;
2556 /* This happens when device was stopped but VCPU wasn't. */
2557 if (!vdev->vm_running) {
2558 return;
2559 }
2560 virtio_queue_set_notification(vq, 0);
2561 qemu_bh_schedule(q->tx_bh);
2562 }
2563
2564 static void virtio_net_tx_timer(void *opaque)
2565 {
2566 VirtIONetQueue *q = opaque;
2567 VirtIONet *n = q->n;
2568 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2569 /* This happens when device was stopped but BH wasn't. */
2570 if (!vdev->vm_running) {
2571 /* Make sure tx waiting is set, so we'll run when restarted. */
2572 assert(q->tx_waiting);
2573 return;
2574 }
2575
2576 q->tx_waiting = 0;
2577
2578 /* Just in case the driver is not ready on more */
2579 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2580 return;
2581 }
2582
2583 virtio_queue_set_notification(q->tx_vq, 1);
2584 virtio_net_flush_tx(q);
2585 }
2586
2587 static void virtio_net_tx_bh(void *opaque)
2588 {
2589 VirtIONetQueue *q = opaque;
2590 VirtIONet *n = q->n;
2591 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2592 int32_t ret;
2593
2594 /* This happens when device was stopped but BH wasn't. */
2595 if (!vdev->vm_running) {
2596 /* Make sure tx waiting is set, so we'll run when restarted. */
2597 assert(q->tx_waiting);
2598 return;
2599 }
2600
2601 q->tx_waiting = 0;
2602
2603 /* Just in case the driver is not ready on more */
2604 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
2605 return;
2606 }
2607
2608 ret = virtio_net_flush_tx(q);
2609 if (ret == -EBUSY || ret == -EINVAL) {
2610 return; /* Notification re-enable handled by tx_complete or device
2611 * broken */
2612 }
2613
2614 /* If we flush a full burst of packets, assume there are
2615 * more coming and immediately reschedule */
2616 if (ret >= n->tx_burst) {
2617 qemu_bh_schedule(q->tx_bh);
2618 q->tx_waiting = 1;
2619 return;
2620 }
2621
2622 /* If less than a full burst, re-enable notification and flush
2623 * anything that may have come in while we weren't looking. If
2624 * we find something, assume the guest is still active and reschedule */
2625 virtio_queue_set_notification(q->tx_vq, 1);
2626 ret = virtio_net_flush_tx(q);
2627 if (ret == -EINVAL) {
2628 return;
2629 } else if (ret > 0) {
2630 virtio_queue_set_notification(q->tx_vq, 0);
2631 qemu_bh_schedule(q->tx_bh);
2632 q->tx_waiting = 1;
2633 }
2634 }
2635
2636 static void virtio_net_add_queue(VirtIONet *n, int index)
2637 {
2638 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2639
2640 n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
2641 virtio_net_handle_rx);
2642
2643 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
2644 n->vqs[index].tx_vq =
2645 virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2646 virtio_net_handle_tx_timer);
2647 n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
2648 virtio_net_tx_timer,
2649 &n->vqs[index]);
2650 } else {
2651 n->vqs[index].tx_vq =
2652 virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2653 virtio_net_handle_tx_bh);
2654 n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
2655 }
2656
2657 n->vqs[index].tx_waiting = 0;
2658 n->vqs[index].n = n;
2659 }
2660
2661 static void virtio_net_del_queue(VirtIONet *n, int index)
2662 {
2663 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2664 VirtIONetQueue *q = &n->vqs[index];
2665 NetClientState *nc = qemu_get_subqueue(n->nic, index);
2666
2667 qemu_purge_queued_packets(nc);
2668
2669 virtio_del_queue(vdev, index * 2);
2670 if (q->tx_timer) {
2671 timer_del(q->tx_timer);
2672 timer_free(q->tx_timer);
2673 q->tx_timer = NULL;
2674 } else {
2675 qemu_bh_delete(q->tx_bh);
2676 q->tx_bh = NULL;
2677 }
2678 q->tx_waiting = 0;
2679 virtio_del_queue(vdev, index * 2 + 1);
2680 }
2681
2682 static void virtio_net_change_num_queues(VirtIONet *n, int new_max_queues)
2683 {
2684 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2685 int old_num_queues = virtio_get_num_queues(vdev);
2686 int new_num_queues = new_max_queues * 2 + 1;
2687 int i;
2688
2689 assert(old_num_queues >= 3);
2690 assert(old_num_queues % 2 == 1);
2691
2692 if (old_num_queues == new_num_queues) {
2693 return;
2694 }
2695
2696 /*
2697 * We always need to remove and add ctrl vq if
2698 * old_num_queues != new_num_queues. Remove ctrl_vq first,
2699 * and then we only enter one of the following two loops.
2700 */
2701 virtio_del_queue(vdev, old_num_queues - 1);
2702
2703 for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
2704 /* new_num_queues < old_num_queues */
2705 virtio_net_del_queue(n, i / 2);
2706 }
2707
2708 for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
2709 /* new_num_queues > old_num_queues */
2710 virtio_net_add_queue(n, i / 2);
2711 }
2712
2713 /* add ctrl_vq last */
2714 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
2715 }
2716
2717 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
2718 {
2719 int max = multiqueue ? n->max_queues : 1;
2720
2721 n->multiqueue = multiqueue;
2722 virtio_net_change_num_queues(n, max);
2723
2724 virtio_net_set_queues(n);
2725 }
2726
2727 static int virtio_net_post_load_device(void *opaque, int version_id)
2728 {
2729 VirtIONet *n = opaque;
2730 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2731 int i, link_down;
2732
2733 trace_virtio_net_post_load_device();
2734 virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
2735 virtio_vdev_has_feature(vdev,
2736 VIRTIO_F_VERSION_1),
2737 virtio_vdev_has_feature(vdev,
2738 VIRTIO_NET_F_HASH_REPORT));
2739
2740 /* MAC_TABLE_ENTRIES may be different from the saved image */
2741 if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
2742 n->mac_table.in_use = 0;
2743 }
2744
2745 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
2746 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
2747 }
2748
2749 /*
2750 * curr_guest_offloads will be later overwritten by the
2751 * virtio_set_features_nocheck call done from the virtio_load.
2752 * Here we make sure it is preserved and restored accordingly
2753 * in the virtio_net_post_load_virtio callback.
2754 */
2755 n->saved_guest_offloads = n->curr_guest_offloads;
2756
2757 virtio_net_set_queues(n);
2758
2759 /* Find the first multicast entry in the saved MAC filter */
2760 for (i = 0; i < n->mac_table.in_use; i++) {
2761 if (n->mac_table.macs[i * ETH_ALEN] & 1) {
2762 break;
2763 }
2764 }
2765 n->mac_table.first_multi = i;
2766
2767 /* nc.link_down can't be migrated, so infer link_down according
2768 * to link status bit in n->status */
2769 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
2770 for (i = 0; i < n->max_queues; i++) {
2771 qemu_get_subqueue(n->nic, i)->link_down = link_down;
2772 }
2773
2774 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
2775 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
2776 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
2777 QEMU_CLOCK_VIRTUAL,
2778 virtio_net_announce_timer, n);
2779 if (n->announce_timer.round) {
2780 timer_mod(n->announce_timer.tm,
2781 qemu_clock_get_ms(n->announce_timer.type));
2782 } else {
2783 qemu_announce_timer_del(&n->announce_timer, false);
2784 }
2785 }
2786
2787 if (n->rss_data.enabled) {
2788 trace_virtio_net_rss_enable(n->rss_data.hash_types,
2789 n->rss_data.indirections_len,
2790 sizeof(n->rss_data.key));
2791 } else {
2792 trace_virtio_net_rss_disable();
2793 }
2794 return 0;
2795 }
2796
2797 static int virtio_net_post_load_virtio(VirtIODevice *vdev)
2798 {
2799 VirtIONet *n = VIRTIO_NET(vdev);
2800 /*
2801 * The actual needed state is now in saved_guest_offloads,
2802 * see virtio_net_post_load_device for detail.
2803 * Restore it back and apply the desired offloads.
2804 */
2805 n->curr_guest_offloads = n->saved_guest_offloads;
2806 if (peer_has_vnet_hdr(n)) {
2807 virtio_net_apply_guest_offloads(n);
2808 }
2809
2810 return 0;
2811 }
2812
2813 /* tx_waiting field of a VirtIONetQueue */
2814 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = {
2815 .name = "virtio-net-queue-tx_waiting",
2816 .fields = (VMStateField[]) {
2817 VMSTATE_UINT32(tx_waiting, VirtIONetQueue),
2818 VMSTATE_END_OF_LIST()
2819 },
2820 };
2821
2822 static bool max_queues_gt_1(void *opaque, int version_id)
2823 {
2824 return VIRTIO_NET(opaque)->max_queues > 1;
2825 }
2826
2827 static bool has_ctrl_guest_offloads(void *opaque, int version_id)
2828 {
2829 return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque),
2830 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
2831 }
2832
2833 static bool mac_table_fits(void *opaque, int version_id)
2834 {
2835 return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES;
2836 }
2837
2838 static bool mac_table_doesnt_fit(void *opaque, int version_id)
2839 {
2840 return !mac_table_fits(opaque, version_id);
2841 }
2842
2843 /* This temporary type is shared by all the WITH_TMP methods
2844 * although only some fields are used by each.
2845 */
2846 struct VirtIONetMigTmp {
2847 VirtIONet *parent;
2848 VirtIONetQueue *vqs_1;
2849 uint16_t curr_queues_1;
2850 uint8_t has_ufo;
2851 uint32_t has_vnet_hdr;
2852 };
2853
2854 /* The 2nd and subsequent tx_waiting flags are loaded later than
2855 * the 1st entry in the queues and only if there's more than one
2856 * entry. We use the tmp mechanism to calculate a temporary
2857 * pointer and count and also validate the count.
2858 */
2859
2860 static int virtio_net_tx_waiting_pre_save(void *opaque)
2861 {
2862 struct VirtIONetMigTmp *tmp = opaque;
2863
2864 tmp->vqs_1 = tmp->parent->vqs + 1;
2865 tmp->curr_queues_1 = tmp->parent->curr_queues - 1;
2866 if (tmp->parent->curr_queues == 0) {
2867 tmp->curr_queues_1 = 0;
2868 }
2869
2870 return 0;
2871 }
2872
2873 static int virtio_net_tx_waiting_pre_load(void *opaque)
2874 {
2875 struct VirtIONetMigTmp *tmp = opaque;
2876
2877 /* Reuse the pointer setup from save */
2878 virtio_net_tx_waiting_pre_save(opaque);
2879
2880 if (tmp->parent->curr_queues > tmp->parent->max_queues) {
2881 error_report("virtio-net: curr_queues %x > max_queues %x",
2882 tmp->parent->curr_queues, tmp->parent->max_queues);
2883
2884 return -EINVAL;
2885 }
2886
2887 return 0; /* all good */
2888 }
2889
2890 static const VMStateDescription vmstate_virtio_net_tx_waiting = {
2891 .name = "virtio-net-tx_waiting",
2892 .pre_load = virtio_net_tx_waiting_pre_load,
2893 .pre_save = virtio_net_tx_waiting_pre_save,
2894 .fields = (VMStateField[]) {
2895 VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp,
2896 curr_queues_1,
2897 vmstate_virtio_net_queue_tx_waiting,
2898 struct VirtIONetQueue),
2899 VMSTATE_END_OF_LIST()
2900 },
2901 };
2902
2903 /* the 'has_ufo' flag is just tested; if the incoming stream has the
2904 * flag set we need to check that we have it
2905 */
2906 static int virtio_net_ufo_post_load(void *opaque, int version_id)
2907 {
2908 struct VirtIONetMigTmp *tmp = opaque;
2909
2910 if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) {
2911 error_report("virtio-net: saved image requires TUN_F_UFO support");
2912 return -EINVAL;
2913 }
2914
2915 return 0;
2916 }
2917
2918 static int virtio_net_ufo_pre_save(void *opaque)
2919 {
2920 struct VirtIONetMigTmp *tmp = opaque;
2921
2922 tmp->has_ufo = tmp->parent->has_ufo;
2923
2924 return 0;
2925 }
2926
2927 static const VMStateDescription vmstate_virtio_net_has_ufo = {
2928 .name = "virtio-net-ufo",
2929 .post_load = virtio_net_ufo_post_load,
2930 .pre_save = virtio_net_ufo_pre_save,
2931 .fields = (VMStateField[]) {
2932 VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp),
2933 VMSTATE_END_OF_LIST()
2934 },
2935 };
2936
2937 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
2938 * flag set we need to check that we have it
2939 */
2940 static int virtio_net_vnet_post_load(void *opaque, int version_id)
2941 {
2942 struct VirtIONetMigTmp *tmp = opaque;
2943
2944 if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) {
2945 error_report("virtio-net: saved image requires vnet_hdr=on");
2946 return -EINVAL;
2947 }
2948
2949 return 0;
2950 }
2951
2952 static int virtio_net_vnet_pre_save(void *opaque)
2953 {
2954 struct VirtIONetMigTmp *tmp = opaque;
2955
2956 tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr;
2957
2958 return 0;
2959 }
2960
2961 static const VMStateDescription vmstate_virtio_net_has_vnet = {
2962 .name = "virtio-net-vnet",
2963 .post_load = virtio_net_vnet_post_load,
2964 .pre_save = virtio_net_vnet_pre_save,
2965 .fields = (VMStateField[]) {
2966 VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp),
2967 VMSTATE_END_OF_LIST()
2968 },
2969 };
2970
2971 static bool virtio_net_rss_needed(void *opaque)
2972 {
2973 return VIRTIO_NET(opaque)->rss_data.enabled;
2974 }
2975
2976 static const VMStateDescription vmstate_virtio_net_rss = {
2977 .name = "virtio-net-device/rss",
2978 .version_id = 1,
2979 .minimum_version_id = 1,
2980 .needed = virtio_net_rss_needed,
2981 .fields = (VMStateField[]) {
2982 VMSTATE_BOOL(rss_data.enabled, VirtIONet),
2983 VMSTATE_BOOL(rss_data.redirect, VirtIONet),
2984 VMSTATE_BOOL(rss_data.populate_hash, VirtIONet),
2985 VMSTATE_UINT32(rss_data.hash_types, VirtIONet),
2986 VMSTATE_UINT16(rss_data.indirections_len, VirtIONet),
2987 VMSTATE_UINT16(rss_data.default_queue, VirtIONet),
2988 VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet,
2989 VIRTIO_NET_RSS_MAX_KEY_SIZE),
2990 VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet,
2991 rss_data.indirections_len, 0,
2992 vmstate_info_uint16, uint16_t),
2993 VMSTATE_END_OF_LIST()
2994 },
2995 };
2996
2997 static const VMStateDescription vmstate_virtio_net_device = {
2998 .name = "virtio-net-device",
2999 .version_id = VIRTIO_NET_VM_VERSION,
3000 .minimum_version_id = VIRTIO_NET_VM_VERSION,
3001 .post_load = virtio_net_post_load_device,
3002 .fields = (VMStateField[]) {
3003 VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN),
3004 VMSTATE_STRUCT_POINTER(vqs, VirtIONet,
3005 vmstate_virtio_net_queue_tx_waiting,
3006 VirtIONetQueue),
3007 VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet),
3008 VMSTATE_UINT16(status, VirtIONet),
3009 VMSTATE_UINT8(promisc, VirtIONet),
3010 VMSTATE_UINT8(allmulti, VirtIONet),
3011 VMSTATE_UINT32(mac_table.in_use, VirtIONet),
3012
3013 /* Guarded pair: If it fits we load it, else we throw it away
3014 * - can happen if source has a larger MAC table.; post-load
3015 * sets flags in this case.
3016 */
3017 VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet,
3018 0, mac_table_fits, mac_table.in_use,
3019 ETH_ALEN),
3020 VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0,
3021 mac_table.in_use, ETH_ALEN),
3022
3023 /* Note: This is an array of uint32's that's always been saved as a
3024 * buffer; hold onto your endiannesses; it's actually used as a bitmap
3025 * but based on the uint.
3026 */
3027 VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3),
3028 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3029 vmstate_virtio_net_has_vnet),
3030 VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet),
3031 VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet),
3032 VMSTATE_UINT8(alluni, VirtIONet),
3033 VMSTATE_UINT8(nomulti, VirtIONet),
3034 VMSTATE_UINT8(nouni, VirtIONet),
3035 VMSTATE_UINT8(nobcast, VirtIONet),
3036 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3037 vmstate_virtio_net_has_ufo),
3038 VMSTATE_SINGLE_TEST(max_queues, VirtIONet, max_queues_gt_1, 0,
3039 vmstate_info_uint16_equal, uint16_t),
3040 VMSTATE_UINT16_TEST(curr_queues, VirtIONet, max_queues_gt_1),
3041 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3042 vmstate_virtio_net_tx_waiting),
3043 VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet,
3044 has_ctrl_guest_offloads),
3045 VMSTATE_END_OF_LIST()
3046 },
3047 .subsections = (const VMStateDescription * []) {
3048 &vmstate_virtio_net_rss,
3049 NULL
3050 }
3051 };
3052
3053 static NetClientInfo net_virtio_info = {
3054 .type = NET_CLIENT_DRIVER_NIC,
3055 .size = sizeof(NICState),
3056 .can_receive = virtio_net_can_receive,
3057 .receive = virtio_net_receive,
3058 .link_status_changed = virtio_net_set_link_status,
3059 .query_rx_filter = virtio_net_query_rxfilter,
3060 .announce = virtio_net_announce,
3061 };
3062
3063 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
3064 {
3065 VirtIONet *n = VIRTIO_NET(vdev);
3066 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
3067 assert(n->vhost_started);
3068 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
3069 }
3070
3071 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
3072 bool mask)
3073 {
3074 VirtIONet *n = VIRTIO_NET(vdev);
3075 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
3076 assert(n->vhost_started);
3077 vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
3078 vdev, idx, mask);
3079 }
3080
3081 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
3082 {
3083 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
3084
3085 n->config_size = virtio_feature_get_config_size(feature_sizes,
3086 host_features);
3087 }
3088
3089 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
3090 const char *type)
3091 {
3092 /*
3093 * The name can be NULL, the netclient name will be type.x.
3094 */
3095 assert(type != NULL);
3096
3097 g_free(n->netclient_name);
3098 g_free(n->netclient_type);
3099 n->netclient_name = g_strdup(name);
3100 n->netclient_type = g_strdup(type);
3101 }
3102
3103 static bool failover_unplug_primary(VirtIONet *n)
3104 {
3105 HotplugHandler *hotplug_ctrl;
3106 PCIDevice *pci_dev;
3107 Error *err = NULL;
3108
3109 hotplug_ctrl = qdev_get_hotplug_handler(n->primary_dev);
3110 if (hotplug_ctrl) {
3111 pci_dev = PCI_DEVICE(n->primary_dev);
3112 pci_dev->partially_hotplugged = true;
3113 hotplug_handler_unplug_request(hotplug_ctrl, n->primary_dev, &err);
3114 if (err) {
3115 error_report_err(err);
3116 return false;
3117 }
3118 } else {
3119 return false;
3120 }
3121 return true;
3122 }
3123
3124 static bool failover_replug_primary(VirtIONet *n, Error **errp)
3125 {
3126 Error *err = NULL;
3127 HotplugHandler *hotplug_ctrl;
3128 PCIDevice *pdev = PCI_DEVICE(n->primary_dev);
3129
3130 if (!pdev->partially_hotplugged) {
3131 return true;
3132 }
3133 if (!n->primary_device_opts) {
3134 n->primary_device_opts = qemu_opts_from_qdict(
3135 qemu_find_opts("device"),
3136 n->primary_device_dict, errp);
3137 if (!n->primary_device_opts) {
3138 return false;
3139 }
3140 }
3141 n->primary_bus = n->primary_dev->parent_bus;
3142 if (!n->primary_bus) {
3143 error_setg(errp, "virtio_net: couldn't find primary bus");
3144 return false;
3145 }
3146 qdev_set_parent_bus(n->primary_dev, n->primary_bus);
3147 n->primary_should_be_hidden = false;
3148 if (!qemu_opt_set_bool(n->primary_device_opts,
3149 "partially_hotplugged", true, errp)) {
3150 return false;
3151 }
3152 hotplug_ctrl = qdev_get_hotplug_handler(n->primary_dev);
3153 if (hotplug_ctrl) {
3154 hotplug_handler_pre_plug(hotplug_ctrl, n->primary_dev, &err);
3155 if (err) {
3156 goto out;
3157 }
3158 hotplug_handler_plug(hotplug_ctrl, n->primary_dev, &err);
3159 }
3160
3161 out:
3162 error_propagate(errp, err);
3163 return !err;
3164 }
3165
3166 static void virtio_net_handle_migration_primary(VirtIONet *n,
3167 MigrationState *s)
3168 {
3169 bool should_be_hidden;
3170 Error *err = NULL;
3171
3172 should_be_hidden = atomic_read(&n->primary_should_be_hidden);
3173
3174 if (!n->primary_dev) {
3175 n->primary_dev = virtio_connect_failover_devices(n, n->qdev, &err);
3176 if (!n->primary_dev) {
3177 return;
3178 }
3179 }
3180
3181 if (migration_in_setup(s) && !should_be_hidden) {
3182 if (failover_unplug_primary(n)) {
3183 vmstate_unregister(VMSTATE_IF(n->primary_dev),
3184 qdev_get_vmsd(n->primary_dev),
3185 n->primary_dev);
3186 qapi_event_send_unplug_primary(n->primary_device_id);
3187 atomic_set(&n->primary_should_be_hidden, true);
3188 } else {
3189 warn_report("couldn't unplug primary device");
3190 }
3191 } else if (migration_has_failed(s)) {
3192 /* We already unplugged the device let's plug it back */
3193 if (!failover_replug_primary(n, &err)) {
3194 if (err) {
3195 error_report_err(err);
3196 }
3197 }
3198 }
3199 }
3200
3201 static void virtio_net_migration_state_notifier(Notifier *notifier, void *data)
3202 {
3203 MigrationState *s = data;
3204 VirtIONet *n = container_of(notifier, VirtIONet, migration_state);
3205 virtio_net_handle_migration_primary(n, s);
3206 }
3207
3208 static int virtio_net_primary_should_be_hidden(DeviceListener *listener,
3209 QemuOpts *device_opts)
3210 {
3211 VirtIONet *n = container_of(listener, VirtIONet, primary_listener);
3212 bool match_found = false;
3213 bool hide = false;
3214
3215 if (!device_opts) {
3216 return -1;
3217 }
3218 n->primary_device_dict = qemu_opts_to_qdict(device_opts,
3219 n->primary_device_dict);
3220 if (n->primary_device_dict) {
3221 g_free(n->standby_id);
3222 n->standby_id = g_strdup(qdict_get_try_str(n->primary_device_dict,
3223 "failover_pair_id"));
3224 }
3225 if (g_strcmp0(n->standby_id, n->netclient_name) == 0) {
3226 match_found = true;
3227 } else {
3228 match_found = false;
3229 hide = false;
3230 g_free(n->standby_id);
3231 n->primary_device_dict = NULL;
3232 goto out;
3233 }
3234
3235 n->primary_device_opts = device_opts;
3236
3237 /* primary_should_be_hidden is set during feature negotiation */
3238 hide = atomic_read(&n->primary_should_be_hidden);
3239
3240 if (n->primary_device_dict) {
3241 g_free(n->primary_device_id);
3242 n->primary_device_id = g_strdup(qdict_get_try_str(
3243 n->primary_device_dict, "id"));
3244 if (!n->primary_device_id) {
3245 warn_report("primary_device_id not set");
3246 }
3247 }
3248
3249 out:
3250 if (match_found && hide) {
3251 return 1;
3252 } else if (match_found && !hide) {
3253 return 0;
3254 } else {
3255 return -1;
3256 }
3257 }