]> git.ipfire.org Git - thirdparty/qemu.git/blob - hw/net/virtio-net.c
projects / thirdparty / qemu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
virtio-net: Allow qemu_announce_self to trigger virtio announcements
[thirdparty/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/iov.h"
16 #include "hw/virtio/virtio.h"
17 #include "net/net.h"
18 #include "net/checksum.h"
19 #include "net/tap.h"
20 #include "qemu/error-report.h"
21 #include "qemu/timer.h"
22 #include "hw/virtio/virtio-net.h"
23 #include "net/vhost_net.h"
24 #include "net/announce.h"
25 #include "hw/virtio/virtio-bus.h"
26 #include "qapi/error.h"
27 #include "qapi/qapi-events-net.h"
28 #include "hw/virtio/virtio-access.h"
29 #include "migration/misc.h"
30 #include "standard-headers/linux/ethtool.h"
31 #include "trace.h"
32
33 #define VIRTIO_NET_VM_VERSION 11
34
35 #define MAC_TABLE_ENTRIES 64
36 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */
37
38 /* previously fixed value */
39 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
40 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
41
42 /* for now, only allow larger queues; with virtio-1, guest can downsize */
43 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
44 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
45
46 #define VIRTIO_NET_IP4_ADDR_SIZE 8 /* ipv4 saddr + daddr */
47
48 #define VIRTIO_NET_TCP_FLAG 0x3F
49 #define VIRTIO_NET_TCP_HDR_LENGTH 0xF000
50
51 /* IPv4 max payload, 16 bits in the header */
52 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
53 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
54
55 /* header length value in ip header without option */
56 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
57
58 #define VIRTIO_NET_IP6_ADDR_SIZE 32 /* ipv6 saddr + daddr */
59 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
60
61 /* Purge coalesced packets timer interval, This value affects the performance
62 a lot, and should be tuned carefully, '300000'(300us) is the recommended
63 value to pass the WHQL test, '50000' can gain 2x netperf throughput with
64 tso/gso/gro 'off'. */
65 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
66
67 /* temporary until standard header include it */
68 #if !defined(VIRTIO_NET_HDR_F_RSC_INFO)
69
70 #define VIRTIO_NET_HDR_F_RSC_INFO 4 /* rsc_ext data in csum_ fields */
71 #define VIRTIO_NET_F_RSC_EXT 61
72
73 static inline __virtio16 *virtio_net_rsc_ext_num_packets(
74 struct virtio_net_hdr *hdr)
75 {
76 return &hdr->csum_start;
77 }
78
79 static inline __virtio16 *virtio_net_rsc_ext_num_dupacks(
80 struct virtio_net_hdr *hdr)
81 {
82 return &hdr->csum_offset;
83 }
84
85 #endif
86
87 static VirtIOFeature feature_sizes[] = {
88 {.flags = 1ULL << VIRTIO_NET_F_MAC,
89 .end = virtio_endof(struct virtio_net_config, mac)},
90 {.flags = 1ULL << VIRTIO_NET_F_STATUS,
91 .end = virtio_endof(struct virtio_net_config, status)},
92 {.flags = 1ULL << VIRTIO_NET_F_MQ,
93 .end = virtio_endof(struct virtio_net_config, max_virtqueue_pairs)},
94 {.flags = 1ULL << VIRTIO_NET_F_MTU,
95 .end = virtio_endof(struct virtio_net_config, mtu)},
96 {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX,
97 .end = virtio_endof(struct virtio_net_config, duplex)},
98 {}
99 };
100
101 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
102 {
103 VirtIONet *n = qemu_get_nic_opaque(nc);
104
105 return &n->vqs[nc->queue_index];
106 }
107
108 static int vq2q(int queue_index)
109 {
110 return queue_index / 2;
111 }
112
113 /* TODO
114 * - we could suppress RX interrupt if we were so inclined.
115 */
116
117 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
118 {
119 VirtIONet *n = VIRTIO_NET(vdev);
120 struct virtio_net_config netcfg;
121
122 virtio_stw_p(vdev, &netcfg.status, n->status);
123 virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues);
124 virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
125 memcpy(netcfg.mac, n->mac, ETH_ALEN);
126 virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed);
127 netcfg.duplex = n->net_conf.duplex;
128 memcpy(config, &netcfg, n->config_size);
129 }
130
131 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
132 {
133 VirtIONet *n = VIRTIO_NET(vdev);
134 struct virtio_net_config netcfg = {};
135
136 memcpy(&netcfg, config, n->config_size);
137
138 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
139 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
140 memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
141 memcpy(n->mac, netcfg.mac, ETH_ALEN);
142 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
143 }
144 }
145
146 static bool virtio_net_started(VirtIONet *n, uint8_t status)
147 {
148 VirtIODevice *vdev = VIRTIO_DEVICE(n);
149 return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
150 (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
151 }
152
153 static void virtio_net_announce_notify(VirtIONet *net)
154 {
155 VirtIODevice *vdev = VIRTIO_DEVICE(net);
156 trace_virtio_net_announce_notify();
157
158 net->status |= VIRTIO_NET_S_ANNOUNCE;
159 virtio_notify_config(vdev);
160 }
161
162 static void virtio_net_announce_timer(void *opaque)
163 {
164 VirtIONet *n = opaque;
165 trace_virtio_net_announce_timer(n->announce_timer.round);
166
167 n->announce_timer.round--;
168 virtio_net_announce_notify(n);
169 }
170
171 static void virtio_net_announce(NetClientState *nc)
172 {
173 VirtIONet *n = qemu_get_nic_opaque(nc);
174 VirtIODevice *vdev = VIRTIO_DEVICE(n);
175
176 /*
177 * Make sure the virtio migration announcement timer isn't running
178 * If it is, let it trigger announcement so that we do not cause
179 * confusion.
180 */
181 if (n->announce_timer.round) {
182 return;
183 }
184
185 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
186 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
187 virtio_net_announce_notify(n);
188 }
189 }
190
191 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
192 {
193 VirtIODevice *vdev = VIRTIO_DEVICE(n);
194 NetClientState *nc = qemu_get_queue(n->nic);
195 int queues = n->multiqueue ? n->max_queues : 1;
196
197 if (!get_vhost_net(nc->peer)) {
198 return;
199 }
200
201 if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
202 !!n->vhost_started) {
203 return;
204 }
205 if (!n->vhost_started) {
206 int r, i;
207
208 if (n->needs_vnet_hdr_swap) {
209 error_report("backend does not support %s vnet headers; "
210 "falling back on userspace virtio",
211 virtio_is_big_endian(vdev) ? "BE" : "LE");
212 return;
213 }
214
215 /* Any packets outstanding? Purge them to avoid touching rings
216 * when vhost is running.
217 */
218 for (i = 0; i < queues; i++) {
219 NetClientState *qnc = qemu_get_subqueue(n->nic, i);
220
221 /* Purge both directions: TX and RX. */
222 qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
223 qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
224 }
225
226 if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
227 r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
228 if (r < 0) {
229 error_report("%uBytes MTU not supported by the backend",
230 n->net_conf.mtu);
231
232 return;
233 }
234 }
235
236 n->vhost_started = 1;
237 r = vhost_net_start(vdev, n->nic->ncs, queues);
238 if (r < 0) {
239 error_report("unable to start vhost net: %d: "
240 "falling back on userspace virtio", -r);
241 n->vhost_started = 0;
242 }
243 } else {
244 vhost_net_stop(vdev, n->nic->ncs, queues);
245 n->vhost_started = 0;
246 }
247 }
248
249 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
250 NetClientState *peer,
251 bool enable)
252 {
253 if (virtio_is_big_endian(vdev)) {
254 return qemu_set_vnet_be(peer, enable);
255 } else {
256 return qemu_set_vnet_le(peer, enable);
257 }
258 }
259
260 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
261 int queues, bool enable)
262 {
263 int i;
264
265 for (i = 0; i < queues; i++) {
266 if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
267 enable) {
268 while (--i >= 0) {
269 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
270 }
271
272 return true;
273 }
274 }
275
276 return false;
277 }
278
279 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
280 {
281 VirtIODevice *vdev = VIRTIO_DEVICE(n);
282 int queues = n->multiqueue ? n->max_queues : 1;
283
284 if (virtio_net_started(n, status)) {
285 /* Before using the device, we tell the network backend about the
286 * endianness to use when parsing vnet headers. If the backend
287 * can't do it, we fallback onto fixing the headers in the core
288 * virtio-net code.
289 */
290 n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
291 queues, true);
292 } else if (virtio_net_started(n, vdev->status)) {
293 /* After using the device, we need to reset the network backend to
294 * the default (guest native endianness), otherwise the guest may
295 * lose network connectivity if it is rebooted into a different
296 * endianness.
297 */
298 virtio_net_set_vnet_endian(vdev, n->nic->ncs, queues, false);
299 }
300 }
301
302 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq)
303 {
304 unsigned int dropped = virtqueue_drop_all(vq);
305 if (dropped) {
306 virtio_notify(vdev, vq);
307 }
308 }
309
310 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
311 {
312 VirtIONet *n = VIRTIO_NET(vdev);
313 VirtIONetQueue *q;
314 int i;
315 uint8_t queue_status;
316
317 virtio_net_vnet_endian_status(n, status);
318 virtio_net_vhost_status(n, status);
319
320 for (i = 0; i < n->max_queues; i++) {
321 NetClientState *ncs = qemu_get_subqueue(n->nic, i);
322 bool queue_started;
323 q = &n->vqs[i];
324
325 if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
326 queue_status = 0;
327 } else {
328 queue_status = status;
329 }
330 queue_started =
331 virtio_net_started(n, queue_status) && !n->vhost_started;
332
333 if (queue_started) {
334 qemu_flush_queued_packets(ncs);
335 }
336
337 if (!q->tx_waiting) {
338 continue;
339 }
340
341 if (queue_started) {
342 if (q->tx_timer) {
343 timer_mod(q->tx_timer,
344 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
345 } else {
346 qemu_bh_schedule(q->tx_bh);
347 }
348 } else {
349 if (q->tx_timer) {
350 timer_del(q->tx_timer);
351 } else {
352 qemu_bh_cancel(q->tx_bh);
353 }
354 if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 &&
355 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) &&
356 vdev->vm_running) {
357 /* if tx is waiting we are likely have some packets in tx queue
358 * and disabled notification */
359 q->tx_waiting = 0;
360 virtio_queue_set_notification(q->tx_vq, 1);
361 virtio_net_drop_tx_queue_data(vdev, q->tx_vq);
362 }
363 }
364 }
365 }
366
367 static void virtio_net_set_link_status(NetClientState *nc)
368 {
369 VirtIONet *n = qemu_get_nic_opaque(nc);
370 VirtIODevice *vdev = VIRTIO_DEVICE(n);
371 uint16_t old_status = n->status;
372
373 if (nc->link_down)
374 n->status &= ~VIRTIO_NET_S_LINK_UP;
375 else
376 n->status |= VIRTIO_NET_S_LINK_UP;
377
378 if (n->status != old_status)
379 virtio_notify_config(vdev);
380
381 virtio_net_set_status(vdev, vdev->status);
382 }
383
384 static void rxfilter_notify(NetClientState *nc)
385 {
386 VirtIONet *n = qemu_get_nic_opaque(nc);
387
388 if (nc->rxfilter_notify_enabled) {
389 gchar *path = object_get_canonical_path(OBJECT(n->qdev));
390 qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
391 n->netclient_name, path);
392 g_free(path);
393
394 /* disable event notification to avoid events flooding */
395 nc->rxfilter_notify_enabled = 0;
396 }
397 }
398
399 static intList *get_vlan_table(VirtIONet *n)
400 {
401 intList *list, *entry;
402 int i, j;
403
404 list = NULL;
405 for (i = 0; i < MAX_VLAN >> 5; i++) {
406 for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
407 if (n->vlans[i] & (1U << j)) {
408 entry = g_malloc0(sizeof(*entry));
409 entry->value = (i << 5) + j;
410 entry->next = list;
411 list = entry;
412 }
413 }
414 }
415
416 return list;
417 }
418
419 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
420 {
421 VirtIONet *n = qemu_get_nic_opaque(nc);
422 VirtIODevice *vdev = VIRTIO_DEVICE(n);
423 RxFilterInfo *info;
424 strList *str_list, *entry;
425 int i;
426
427 info = g_malloc0(sizeof(*info));
428 info->name = g_strdup(nc->name);
429 info->promiscuous = n->promisc;
430
431 if (n->nouni) {
432 info->unicast = RX_STATE_NONE;
433 } else if (n->alluni) {
434 info->unicast = RX_STATE_ALL;
435 } else {
436 info->unicast = RX_STATE_NORMAL;
437 }
438
439 if (n->nomulti) {
440 info->multicast = RX_STATE_NONE;
441 } else if (n->allmulti) {
442 info->multicast = RX_STATE_ALL;
443 } else {
444 info->multicast = RX_STATE_NORMAL;
445 }
446
447 info->broadcast_allowed = n->nobcast;
448 info->multicast_overflow = n->mac_table.multi_overflow;
449 info->unicast_overflow = n->mac_table.uni_overflow;
450
451 info->main_mac = qemu_mac_strdup_printf(n->mac);
452
453 str_list = NULL;
454 for (i = 0; i < n->mac_table.first_multi; i++) {
455 entry = g_malloc0(sizeof(*entry));
456 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
457 entry->next = str_list;
458 str_list = entry;
459 }
460 info->unicast_table = str_list;
461
462 str_list = NULL;
463 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
464 entry = g_malloc0(sizeof(*entry));
465 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
466 entry->next = str_list;
467 str_list = entry;
468 }
469 info->multicast_table = str_list;
470 info->vlan_table = get_vlan_table(n);
471
472 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
473 info->vlan = RX_STATE_ALL;
474 } else if (!info->vlan_table) {
475 info->vlan = RX_STATE_NONE;
476 } else {
477 info->vlan = RX_STATE_NORMAL;
478 }
479
480 /* enable event notification after query */
481 nc->rxfilter_notify_enabled = 1;
482
483 return info;
484 }
485
486 static void virtio_net_reset(VirtIODevice *vdev)
487 {
488 VirtIONet *n = VIRTIO_NET(vdev);
489 int i;
490
491 /* Reset back to compatibility mode */
492 n->promisc = 1;
493 n->allmulti = 0;
494 n->alluni = 0;
495 n->nomulti = 0;
496 n->nouni = 0;
497 n->nobcast = 0;
498 /* multiqueue is disabled by default */
499 n->curr_queues = 1;
500 timer_del(n->announce_timer.tm);
501 n->announce_timer.round = 0;
502 n->status &= ~VIRTIO_NET_S_ANNOUNCE;
503
504 /* Flush any MAC and VLAN filter table state */
505 n->mac_table.in_use = 0;
506 n->mac_table.first_multi = 0;
507 n->mac_table.multi_overflow = 0;
508 n->mac_table.uni_overflow = 0;
509 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
510 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
511 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
512 memset(n->vlans, 0, MAX_VLAN >> 3);
513
514 /* Flush any async TX */
515 for (i = 0; i < n->max_queues; i++) {
516 NetClientState *nc = qemu_get_subqueue(n->nic, i);
517
518 if (nc->peer) {
519 qemu_flush_or_purge_queued_packets(nc->peer, true);
520 assert(!virtio_net_get_subqueue(nc)->async_tx.elem);
521 }
522 }
523 }
524
525 static void peer_test_vnet_hdr(VirtIONet *n)
526 {
527 NetClientState *nc = qemu_get_queue(n->nic);
528 if (!nc->peer) {
529 return;
530 }
531
532 n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
533 }
534
535 static int peer_has_vnet_hdr(VirtIONet *n)
536 {
537 return n->has_vnet_hdr;
538 }
539
540 static int peer_has_ufo(VirtIONet *n)
541 {
542 if (!peer_has_vnet_hdr(n))
543 return 0;
544
545 n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
546
547 return n->has_ufo;
548 }
549
550 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
551 int version_1)
552 {
553 int i;
554 NetClientState *nc;
555
556 n->mergeable_rx_bufs = mergeable_rx_bufs;
557
558 if (version_1) {
559 n->guest_hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
560 } else {
561 n->guest_hdr_len = n->mergeable_rx_bufs ?
562 sizeof(struct virtio_net_hdr_mrg_rxbuf) :
563 sizeof(struct virtio_net_hdr);
564 }
565
566 for (i = 0; i < n->max_queues; i++) {
567 nc = qemu_get_subqueue(n->nic, i);
568
569 if (peer_has_vnet_hdr(n) &&
570 qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
571 qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
572 n->host_hdr_len = n->guest_hdr_len;
573 }
574 }
575 }
576
577 static int virtio_net_max_tx_queue_size(VirtIONet *n)
578 {
579 NetClientState *peer = n->nic_conf.peers.ncs[0];
580
581 /*
582 * Backends other than vhost-user don't support max queue size.
583 */
584 if (!peer) {
585 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
586 }
587
588 if (peer->info->type != NET_CLIENT_DRIVER_VHOST_USER) {
589 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
590 }
591
592 return VIRTQUEUE_MAX_SIZE;
593 }
594
595 static int peer_attach(VirtIONet *n, int index)
596 {
597 NetClientState *nc = qemu_get_subqueue(n->nic, index);
598
599 if (!nc->peer) {
600 return 0;
601 }
602
603 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
604 vhost_set_vring_enable(nc->peer, 1);
605 }
606
607 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
608 return 0;
609 }
610
611 if (n->max_queues == 1) {
612 return 0;
613 }
614
615 return tap_enable(nc->peer);
616 }
617
618 static int peer_detach(VirtIONet *n, int index)
619 {
620 NetClientState *nc = qemu_get_subqueue(n->nic, index);
621
622 if (!nc->peer) {
623 return 0;
624 }
625
626 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
627 vhost_set_vring_enable(nc->peer, 0);
628 }
629
630 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
631 return 0;
632 }
633
634 return tap_disable(nc->peer);
635 }
636
637 static void virtio_net_set_queues(VirtIONet *n)
638 {
639 int i;
640 int r;
641
642 if (n->nic->peer_deleted) {
643 return;
644 }
645
646 for (i = 0; i < n->max_queues; i++) {
647 if (i < n->curr_queues) {
648 r = peer_attach(n, i);
649 assert(!r);
650 } else {
651 r = peer_detach(n, i);
652 assert(!r);
653 }
654 }
655 }
656
657 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
658
659 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
660 Error **errp)
661 {
662 VirtIONet *n = VIRTIO_NET(vdev);
663 NetClientState *nc = qemu_get_queue(n->nic);
664
665 /* Firstly sync all virtio-net possible supported features */
666 features |= n->host_features;
667
668 virtio_add_feature(&features, VIRTIO_NET_F_MAC);
669
670 if (!peer_has_vnet_hdr(n)) {
671 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
672 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
673 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
674 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
675
676 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
677 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
678 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
679 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
680 }
681
682 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
683 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
684 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
685 }
686
687 if (!get_vhost_net(nc->peer)) {
688 return features;
689 }
690
691 features = vhost_net_get_features(get_vhost_net(nc->peer), features);
692 vdev->backend_features = features;
693
694 if (n->mtu_bypass_backend &&
695 (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) {
696 features |= (1ULL << VIRTIO_NET_F_MTU);
697 }
698
699 return features;
700 }
701
702 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
703 {
704 uint64_t features = 0;
705
706 /* Linux kernel 2.6.25. It understood MAC (as everyone must),
707 * but also these: */
708 virtio_add_feature(&features, VIRTIO_NET_F_MAC);
709 virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
710 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
711 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
712 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
713
714 return features;
715 }
716
717 static void virtio_net_apply_guest_offloads(VirtIONet *n)
718 {
719 qemu_set_offload(qemu_get_queue(n->nic)->peer,
720 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
721 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
722 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
723 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
724 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
725 }
726
727 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
728 {
729 static const uint64_t guest_offloads_mask =
730 (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
731 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
732 (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
733 (1ULL << VIRTIO_NET_F_GUEST_ECN) |
734 (1ULL << VIRTIO_NET_F_GUEST_UFO);
735
736 return guest_offloads_mask & features;
737 }
738
739 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
740 {
741 VirtIODevice *vdev = VIRTIO_DEVICE(n);
742 return virtio_net_guest_offloads_by_features(vdev->guest_features);
743 }
744
745 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
746 {
747 VirtIONet *n = VIRTIO_NET(vdev);
748 int i;
749
750 if (n->mtu_bypass_backend &&
751 !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) {
752 features &= ~(1ULL << VIRTIO_NET_F_MTU);
753 }
754
755 virtio_net_set_multiqueue(n,
756 virtio_has_feature(features, VIRTIO_NET_F_MQ));
757
758 virtio_net_set_mrg_rx_bufs(n,
759 virtio_has_feature(features,
760 VIRTIO_NET_F_MRG_RXBUF),
761 virtio_has_feature(features,
762 VIRTIO_F_VERSION_1));
763
764 n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
765 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4);
766 n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
767 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6);
768
769 if (n->has_vnet_hdr) {
770 n->curr_guest_offloads =
771 virtio_net_guest_offloads_by_features(features);
772 virtio_net_apply_guest_offloads(n);
773 }
774
775 for (i = 0; i < n->max_queues; i++) {
776 NetClientState *nc = qemu_get_subqueue(n->nic, i);
777
778 if (!get_vhost_net(nc->peer)) {
779 continue;
780 }
781 vhost_net_ack_features(get_vhost_net(nc->peer), features);
782 }
783
784 if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
785 memset(n->vlans, 0, MAX_VLAN >> 3);
786 } else {
787 memset(n->vlans, 0xff, MAX_VLAN >> 3);
788 }
789 }
790
791 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
792 struct iovec *iov, unsigned int iov_cnt)
793 {
794 uint8_t on;
795 size_t s;
796 NetClientState *nc = qemu_get_queue(n->nic);
797
798 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
799 if (s != sizeof(on)) {
800 return VIRTIO_NET_ERR;
801 }
802
803 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
804 n->promisc = on;
805 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
806 n->allmulti = on;
807 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
808 n->alluni = on;
809 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
810 n->nomulti = on;
811 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
812 n->nouni = on;
813 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
814 n->nobcast = on;
815 } else {
816 return VIRTIO_NET_ERR;
817 }
818
819 rxfilter_notify(nc);
820
821 return VIRTIO_NET_OK;
822 }
823
824 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
825 struct iovec *iov, unsigned int iov_cnt)
826 {
827 VirtIODevice *vdev = VIRTIO_DEVICE(n);
828 uint64_t offloads;
829 size_t s;
830
831 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
832 return VIRTIO_NET_ERR;
833 }
834
835 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
836 if (s != sizeof(offloads)) {
837 return VIRTIO_NET_ERR;
838 }
839
840 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
841 uint64_t supported_offloads;
842
843 offloads = virtio_ldq_p(vdev, &offloads);
844
845 if (!n->has_vnet_hdr) {
846 return VIRTIO_NET_ERR;
847 }
848
849 n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
850 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4);
851 n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
852 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6);
853 virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT);
854
855 supported_offloads = virtio_net_supported_guest_offloads(n);
856 if (offloads & ~supported_offloads) {
857 return VIRTIO_NET_ERR;
858 }
859
860 n->curr_guest_offloads = offloads;
861 virtio_net_apply_guest_offloads(n);
862
863 return VIRTIO_NET_OK;
864 } else {
865 return VIRTIO_NET_ERR;
866 }
867 }
868
869 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
870 struct iovec *iov, unsigned int iov_cnt)
871 {
872 VirtIODevice *vdev = VIRTIO_DEVICE(n);
873 struct virtio_net_ctrl_mac mac_data;
874 size_t s;
875 NetClientState *nc = qemu_get_queue(n->nic);
876
877 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
878 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
879 return VIRTIO_NET_ERR;
880 }
881 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
882 assert(s == sizeof(n->mac));
883 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
884 rxfilter_notify(nc);
885
886 return VIRTIO_NET_OK;
887 }
888
889 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
890 return VIRTIO_NET_ERR;
891 }
892
893 int in_use = 0;
894 int first_multi = 0;
895 uint8_t uni_overflow = 0;
896 uint8_t multi_overflow = 0;
897 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
898
899 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
900 sizeof(mac_data.entries));
901 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
902 if (s != sizeof(mac_data.entries)) {
903 goto error;
904 }
905 iov_discard_front(&iov, &iov_cnt, s);
906
907 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
908 goto error;
909 }
910
911 if (mac_data.entries <= MAC_TABLE_ENTRIES) {
912 s = iov_to_buf(iov, iov_cnt, 0, macs,
913 mac_data.entries * ETH_ALEN);
914 if (s != mac_data.entries * ETH_ALEN) {
915 goto error;
916 }
917 in_use += mac_data.entries;
918 } else {
919 uni_overflow = 1;
920 }
921
922 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
923
924 first_multi = in_use;
925
926 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
927 sizeof(mac_data.entries));
928 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
929 if (s != sizeof(mac_data.entries)) {
930 goto error;
931 }
932
933 iov_discard_front(&iov, &iov_cnt, s);
934
935 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
936 goto error;
937 }
938
939 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
940 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
941 mac_data.entries * ETH_ALEN);
942 if (s != mac_data.entries * ETH_ALEN) {
943 goto error;
944 }
945 in_use += mac_data.entries;
946 } else {
947 multi_overflow = 1;
948 }
949
950 n->mac_table.in_use = in_use;
951 n->mac_table.first_multi = first_multi;
952 n->mac_table.uni_overflow = uni_overflow;
953 n->mac_table.multi_overflow = multi_overflow;
954 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
955 g_free(macs);
956 rxfilter_notify(nc);
957
958 return VIRTIO_NET_OK;
959
960 error:
961 g_free(macs);
962 return VIRTIO_NET_ERR;
963 }
964
965 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
966 struct iovec *iov, unsigned int iov_cnt)
967 {
968 VirtIODevice *vdev = VIRTIO_DEVICE(n);
969 uint16_t vid;
970 size_t s;
971 NetClientState *nc = qemu_get_queue(n->nic);
972
973 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
974 vid = virtio_lduw_p(vdev, &vid);
975 if (s != sizeof(vid)) {
976 return VIRTIO_NET_ERR;
977 }
978
979 if (vid >= MAX_VLAN)
980 return VIRTIO_NET_ERR;
981
982 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
983 n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
984 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
985 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
986 else
987 return VIRTIO_NET_ERR;
988
989 rxfilter_notify(nc);
990
991 return VIRTIO_NET_OK;
992 }
993
994 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
995 struct iovec *iov, unsigned int iov_cnt)
996 {
997 trace_virtio_net_handle_announce(n->announce_timer.round);
998 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
999 n->status & VIRTIO_NET_S_ANNOUNCE) {
1000 n->status &= ~VIRTIO_NET_S_ANNOUNCE;
1001 if (n->announce_timer.round) {
1002 qemu_announce_timer_step(&n->announce_timer);
1003 }
1004 return VIRTIO_NET_OK;
1005 } else {
1006 return VIRTIO_NET_ERR;
1007 }
1008 }
1009
1010 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
1011 struct iovec *iov, unsigned int iov_cnt)
1012 {
1013 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1014 struct virtio_net_ctrl_mq mq;
1015 size_t s;
1016 uint16_t queues;
1017
1018 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
1019 if (s != sizeof(mq)) {
1020 return VIRTIO_NET_ERR;
1021 }
1022
1023 if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
1024 return VIRTIO_NET_ERR;
1025 }
1026
1027 queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
1028
1029 if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1030 queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1031 queues > n->max_queues ||
1032 !n->multiqueue) {
1033 return VIRTIO_NET_ERR;
1034 }
1035
1036 n->curr_queues = queues;
1037 /* stop the backend before changing the number of queues to avoid handling a
1038 * disabled queue */
1039 virtio_net_set_status(vdev, vdev->status);
1040 virtio_net_set_queues(n);
1041
1042 return VIRTIO_NET_OK;
1043 }
1044
1045 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
1046 {
1047 VirtIONet *n = VIRTIO_NET(vdev);
1048 struct virtio_net_ctrl_hdr ctrl;
1049 virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
1050 VirtQueueElement *elem;
1051 size_t s;
1052 struct iovec *iov, *iov2;
1053 unsigned int iov_cnt;
1054
1055 for (;;) {
1056 elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
1057 if (!elem) {
1058 break;
1059 }
1060 if (iov_size(elem->in_sg, elem->in_num) < sizeof(status) ||
1061 iov_size(elem->out_sg, elem->out_num) < sizeof(ctrl)) {
1062 virtio_error(vdev, "virtio-net ctrl missing headers");
1063 virtqueue_detach_element(vq, elem, 0);
1064 g_free(elem);
1065 break;
1066 }
1067
1068 iov_cnt = elem->out_num;
1069 iov2 = iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num);
1070 s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
1071 iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
1072 if (s != sizeof(ctrl)) {
1073 status = VIRTIO_NET_ERR;
1074 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
1075 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
1076 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
1077 status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
1078 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
1079 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
1080 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
1081 status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
1082 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
1083 status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
1084 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
1085 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
1086 }
1087
1088 s = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, sizeof(status));
1089 assert(s == sizeof(status));
1090
1091 virtqueue_push(vq, elem, sizeof(status));
1092 virtio_notify(vdev, vq);
1093 g_free(iov2);
1094 g_free(elem);
1095 }
1096 }
1097
1098 /* RX */
1099
1100 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
1101 {
1102 VirtIONet *n = VIRTIO_NET(vdev);
1103 int queue_index = vq2q(virtio_get_queue_index(vq));
1104
1105 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
1106 }
1107
1108 static int virtio_net_can_receive(NetClientState *nc)
1109 {
1110 VirtIONet *n = qemu_get_nic_opaque(nc);
1111 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1112 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1113
1114 if (!vdev->vm_running) {
1115 return 0;
1116 }
1117
1118 if (nc->queue_index >= n->curr_queues) {
1119 return 0;
1120 }
1121
1122 if (!virtio_queue_ready(q->rx_vq) ||
1123 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1124 return 0;
1125 }
1126
1127 return 1;
1128 }
1129
1130 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1131 {
1132 VirtIONet *n = q->n;
1133 if (virtio_queue_empty(q->rx_vq) ||
1134 (n->mergeable_rx_bufs &&
1135 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1136 virtio_queue_set_notification(q->rx_vq, 1);
1137
1138 /* To avoid a race condition where the guest has made some buffers
1139 * available after the above check but before notification was
1140 * enabled, check for available buffers again.
1141 */
1142 if (virtio_queue_empty(q->rx_vq) ||
1143 (n->mergeable_rx_bufs &&
1144 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1145 return 0;
1146 }
1147 }
1148
1149 virtio_queue_set_notification(q->rx_vq, 0);
1150 return 1;
1151 }
1152
1153 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1154 {
1155 virtio_tswap16s(vdev, &hdr->hdr_len);
1156 virtio_tswap16s(vdev, &hdr->gso_size);
1157 virtio_tswap16s(vdev, &hdr->csum_start);
1158 virtio_tswap16s(vdev, &hdr->csum_offset);
1159 }
1160
1161 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1162 * it never finds out that the packets don't have valid checksums. This
1163 * causes dhclient to get upset. Fedora's carried a patch for ages to
1164 * fix this with Xen but it hasn't appeared in an upstream release of
1165 * dhclient yet.
1166 *
1167 * To avoid breaking existing guests, we catch udp packets and add
1168 * checksums. This is terrible but it's better than hacking the guest
1169 * kernels.
1170 *
1171 * N.B. if we introduce a zero-copy API, this operation is no longer free so
1172 * we should provide a mechanism to disable it to avoid polluting the host
1173 * cache.
1174 */
1175 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1176 uint8_t *buf, size_t size)
1177 {
1178 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1179 (size > 27 && size < 1500) && /* normal sized MTU */
1180 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1181 (buf[23] == 17) && /* ip.protocol == UDP */
1182 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1183 net_checksum_calculate(buf, size);
1184 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1185 }
1186 }
1187
1188 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1189 const void *buf, size_t size)
1190 {
1191 if (n->has_vnet_hdr) {
1192 /* FIXME this cast is evil */
1193 void *wbuf = (void *)buf;
1194 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1195 size - n->host_hdr_len);
1196
1197 if (n->needs_vnet_hdr_swap) {
1198 virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1199 }
1200 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1201 } else {
1202 struct virtio_net_hdr hdr = {
1203 .flags = 0,
1204 .gso_type = VIRTIO_NET_HDR_GSO_NONE
1205 };
1206 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1207 }
1208 }
1209
1210 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1211 {
1212 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1213 static const uint8_t vlan[] = {0x81, 0x00};
1214 uint8_t *ptr = (uint8_t *)buf;
1215 int i;
1216
1217 if (n->promisc)
1218 return 1;
1219
1220 ptr += n->host_hdr_len;
1221
1222 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1223 int vid = lduw_be_p(ptr + 14) & 0xfff;
1224 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1225 return 0;
1226 }
1227
1228 if (ptr[0] & 1) { // multicast
1229 if (!memcmp(ptr, bcast, sizeof(bcast))) {
1230 return !n->nobcast;
1231 } else if (n->nomulti) {
1232 return 0;
1233 } else if (n->allmulti || n->mac_table.multi_overflow) {
1234 return 1;
1235 }
1236
1237 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1238 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1239 return 1;
1240 }
1241 }
1242 } else { // unicast
1243 if (n->nouni) {
1244 return 0;
1245 } else if (n->alluni || n->mac_table.uni_overflow) {
1246 return 1;
1247 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1248 return 1;
1249 }
1250
1251 for (i = 0; i < n->mac_table.first_multi; i++) {
1252 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1253 return 1;
1254 }
1255 }
1256 }
1257
1258 return 0;
1259 }
1260
1261 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
1262 size_t size)
1263 {
1264 VirtIONet *n = qemu_get_nic_opaque(nc);
1265 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1266 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1267 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1268 struct virtio_net_hdr_mrg_rxbuf mhdr;
1269 unsigned mhdr_cnt = 0;
1270 size_t offset, i, guest_offset;
1271
1272 if (!virtio_net_can_receive(nc)) {
1273 return -1;
1274 }
1275
1276 /* hdr_len refers to the header we supply to the guest */
1277 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1278 return 0;
1279 }
1280
1281 if (!receive_filter(n, buf, size))
1282 return size;
1283
1284 offset = i = 0;
1285
1286 while (offset < size) {
1287 VirtQueueElement *elem;
1288 int len, total;
1289 const struct iovec *sg;
1290
1291 total = 0;
1292
1293 elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1294 if (!elem) {
1295 if (i) {
1296 virtio_error(vdev, "virtio-net unexpected empty queue: "
1297 "i %zd mergeable %d offset %zd, size %zd, "
1298 "guest hdr len %zd, host hdr len %zd "
1299 "guest features 0x%" PRIx64,
1300 i, n->mergeable_rx_bufs, offset, size,
1301 n->guest_hdr_len, n->host_hdr_len,
1302 vdev->guest_features);
1303 }
1304 return -1;
1305 }
1306
1307 if (elem->in_num < 1) {
1308 virtio_error(vdev,
1309 "virtio-net receive queue contains no in buffers");
1310 virtqueue_detach_element(q->rx_vq, elem, 0);
1311 g_free(elem);
1312 return -1;
1313 }
1314
1315 sg = elem->in_sg;
1316 if (i == 0) {
1317 assert(offset == 0);
1318 if (n->mergeable_rx_bufs) {
1319 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1320 sg, elem->in_num,
1321 offsetof(typeof(mhdr), num_buffers),
1322 sizeof(mhdr.num_buffers));
1323 }
1324
1325 receive_header(n, sg, elem->in_num, buf, size);
1326 offset = n->host_hdr_len;
1327 total += n->guest_hdr_len;
1328 guest_offset = n->guest_hdr_len;
1329 } else {
1330 guest_offset = 0;
1331 }
1332
1333 /* copy in packet. ugh */
1334 len = iov_from_buf(sg, elem->in_num, guest_offset,
1335 buf + offset, size - offset);
1336 total += len;
1337 offset += len;
1338 /* If buffers can't be merged, at this point we
1339 * must have consumed the complete packet.
1340 * Otherwise, drop it. */
1341 if (!n->mergeable_rx_bufs && offset < size) {
1342 virtqueue_unpop(q->rx_vq, elem, total);
1343 g_free(elem);
1344 return size;
1345 }
1346
1347 /* signal other side */
1348 virtqueue_fill(q->rx_vq, elem, total, i++);
1349 g_free(elem);
1350 }
1351
1352 if (mhdr_cnt) {
1353 virtio_stw_p(vdev, &mhdr.num_buffers, i);
1354 iov_from_buf(mhdr_sg, mhdr_cnt,
1355 0,
1356 &mhdr.num_buffers, sizeof mhdr.num_buffers);
1357 }
1358
1359 virtqueue_flush(q->rx_vq, i);
1360 virtio_notify(vdev, q->rx_vq);
1361
1362 return size;
1363 }
1364
1365 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf,
1366 size_t size)
1367 {
1368 ssize_t r;
1369
1370 rcu_read_lock();
1371 r = virtio_net_receive_rcu(nc, buf, size);
1372 rcu_read_unlock();
1373 return r;
1374 }
1375
1376 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain,
1377 const uint8_t *buf,
1378 VirtioNetRscUnit *unit)
1379 {
1380 uint16_t ip_hdrlen;
1381 struct ip_header *ip;
1382
1383 ip = (struct ip_header *)(buf + chain->n->guest_hdr_len
1384 + sizeof(struct eth_header));
1385 unit->ip = (void *)ip;
1386 ip_hdrlen = (ip->ip_ver_len & 0xF) << 2;
1387 unit->ip_plen = &ip->ip_len;
1388 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen);
1389 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1390 unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen;
1391 }
1392
1393 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain,
1394 const uint8_t *buf,
1395 VirtioNetRscUnit *unit)
1396 {
1397 struct ip6_header *ip6;
1398
1399 ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len
1400 + sizeof(struct eth_header));
1401 unit->ip = ip6;
1402 unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
1403 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip)\
1404 + sizeof(struct ip6_header));
1405 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1406
1407 /* There is a difference between payload lenght in ipv4 and v6,
1408 ip header is excluded in ipv6 */
1409 unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen;
1410 }
1411
1412 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain,
1413 VirtioNetRscSeg *seg)
1414 {
1415 int ret;
1416 struct virtio_net_hdr *h;
1417
1418 h = (struct virtio_net_hdr *)seg->buf;
1419 h->flags = 0;
1420 h->gso_type = VIRTIO_NET_HDR_GSO_NONE;
1421
1422 if (seg->is_coalesced) {
1423 *virtio_net_rsc_ext_num_packets(h) = seg->packets;
1424 *virtio_net_rsc_ext_num_dupacks(h) = seg->dup_ack;
1425 h->flags = VIRTIO_NET_HDR_F_RSC_INFO;
1426 if (chain->proto == ETH_P_IP) {
1427 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1428 } else {
1429 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1430 }
1431 }
1432
1433 ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size);
1434 QTAILQ_REMOVE(&chain->buffers, seg, next);
1435 g_free(seg->buf);
1436 g_free(seg);
1437
1438 return ret;
1439 }
1440
1441 static void virtio_net_rsc_purge(void *opq)
1442 {
1443 VirtioNetRscSeg *seg, *rn;
1444 VirtioNetRscChain *chain = (VirtioNetRscChain *)opq;
1445
1446 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) {
1447 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
1448 chain->stat.purge_failed++;
1449 continue;
1450 }
1451 }
1452
1453 chain->stat.timer++;
1454 if (!QTAILQ_EMPTY(&chain->buffers)) {
1455 timer_mod(chain->drain_timer,
1456 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
1457 }
1458 }
1459
1460 static void virtio_net_rsc_cleanup(VirtIONet *n)
1461 {
1462 VirtioNetRscChain *chain, *rn_chain;
1463 VirtioNetRscSeg *seg, *rn_seg;
1464
1465 QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) {
1466 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) {
1467 QTAILQ_REMOVE(&chain->buffers, seg, next);
1468 g_free(seg->buf);
1469 g_free(seg);
1470 }
1471
1472 timer_del(chain->drain_timer);
1473 timer_free(chain->drain_timer);
1474 QTAILQ_REMOVE(&n->rsc_chains, chain, next);
1475 g_free(chain);
1476 }
1477 }
1478
1479 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain,
1480 NetClientState *nc,
1481 const uint8_t *buf, size_t size)
1482 {
1483 uint16_t hdr_len;
1484 VirtioNetRscSeg *seg;
1485
1486 hdr_len = chain->n->guest_hdr_len;
1487 seg = g_malloc(sizeof(VirtioNetRscSeg));
1488 seg->buf = g_malloc(hdr_len + sizeof(struct eth_header)
1489 + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD);
1490 memcpy(seg->buf, buf, size);
1491 seg->size = size;
1492 seg->packets = 1;
1493 seg->dup_ack = 0;
1494 seg->is_coalesced = 0;
1495 seg->nc = nc;
1496
1497 QTAILQ_INSERT_TAIL(&chain->buffers, seg, next);
1498 chain->stat.cache++;
1499
1500 switch (chain->proto) {
1501 case ETH_P_IP:
1502 virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
1503 break;
1504 case ETH_P_IPV6:
1505 virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit);
1506 break;
1507 default:
1508 g_assert_not_reached();
1509 }
1510 }
1511
1512 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain,
1513 VirtioNetRscSeg *seg,
1514 const uint8_t *buf,
1515 struct tcp_header *n_tcp,
1516 struct tcp_header *o_tcp)
1517 {
1518 uint32_t nack, oack;
1519 uint16_t nwin, owin;
1520
1521 nack = htonl(n_tcp->th_ack);
1522 nwin = htons(n_tcp->th_win);
1523 oack = htonl(o_tcp->th_ack);
1524 owin = htons(o_tcp->th_win);
1525
1526 if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) {
1527 chain->stat.ack_out_of_win++;
1528 return RSC_FINAL;
1529 } else if (nack == oack) {
1530 /* duplicated ack or window probe */
1531 if (nwin == owin) {
1532 /* duplicated ack, add dup ack count due to whql test up to 1 */
1533 chain->stat.dup_ack++;
1534 return RSC_FINAL;
1535 } else {
1536 /* Coalesce window update */
1537 o_tcp->th_win = n_tcp->th_win;
1538 chain->stat.win_update++;
1539 return RSC_COALESCE;
1540 }
1541 } else {
1542 /* pure ack, go to 'C', finalize*/
1543 chain->stat.pure_ack++;
1544 return RSC_FINAL;
1545 }
1546 }
1547
1548 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain,
1549 VirtioNetRscSeg *seg,
1550 const uint8_t *buf,
1551 VirtioNetRscUnit *n_unit)
1552 {
1553 void *data;
1554 uint16_t o_ip_len;
1555 uint32_t nseq, oseq;
1556 VirtioNetRscUnit *o_unit;
1557
1558 o_unit = &seg->unit;
1559 o_ip_len = htons(*o_unit->ip_plen);
1560 nseq = htonl(n_unit->tcp->th_seq);
1561 oseq = htonl(o_unit->tcp->th_seq);
1562
1563 /* out of order or retransmitted. */
1564 if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) {
1565 chain->stat.data_out_of_win++;
1566 return RSC_FINAL;
1567 }
1568
1569 data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen;
1570 if (nseq == oseq) {
1571 if ((o_unit->payload == 0) && n_unit->payload) {
1572 /* From no payload to payload, normal case, not a dup ack or etc */
1573 chain->stat.data_after_pure_ack++;
1574 goto coalesce;
1575 } else {
1576 return virtio_net_rsc_handle_ack(chain, seg, buf,
1577 n_unit->tcp, o_unit->tcp);
1578 }
1579 } else if ((nseq - oseq) != o_unit->payload) {
1580 /* Not a consistent packet, out of order */
1581 chain->stat.data_out_of_order++;
1582 return RSC_FINAL;
1583 } else {
1584 coalesce:
1585 if ((o_ip_len + n_unit->payload) > chain->max_payload) {
1586 chain->stat.over_size++;
1587 return RSC_FINAL;
1588 }
1589
1590 /* Here comes the right data, the payload length in v4/v6 is different,
1591 so use the field value to update and record the new data len */
1592 o_unit->payload += n_unit->payload; /* update new data len */
1593
1594 /* update field in ip header */
1595 *o_unit->ip_plen = htons(o_ip_len + n_unit->payload);
1596
1597 /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
1598 for windows guest, while this may change the behavior for linux
1599 guest (only if it uses RSC feature). */
1600 o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags;
1601
1602 o_unit->tcp->th_ack = n_unit->tcp->th_ack;
1603 o_unit->tcp->th_win = n_unit->tcp->th_win;
1604
1605 memmove(seg->buf + seg->size, data, n_unit->payload);
1606 seg->size += n_unit->payload;
1607 seg->packets++;
1608 chain->stat.coalesced++;
1609 return RSC_COALESCE;
1610 }
1611 }
1612
1613 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain,
1614 VirtioNetRscSeg *seg,
1615 const uint8_t *buf, size_t size,
1616 VirtioNetRscUnit *unit)
1617 {
1618 struct ip_header *ip1, *ip2;
1619
1620 ip1 = (struct ip_header *)(unit->ip);
1621 ip2 = (struct ip_header *)(seg->unit.ip);
1622 if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst)
1623 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
1624 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
1625 chain->stat.no_match++;
1626 return RSC_NO_MATCH;
1627 }
1628
1629 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
1630 }
1631
1632 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain,
1633 VirtioNetRscSeg *seg,
1634 const uint8_t *buf, size_t size,
1635 VirtioNetRscUnit *unit)
1636 {
1637 struct ip6_header *ip1, *ip2;
1638
1639 ip1 = (struct ip6_header *)(unit->ip);
1640 ip2 = (struct ip6_header *)(seg->unit.ip);
1641 if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address))
1642 || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address))
1643 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
1644 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
1645 chain->stat.no_match++;
1646 return RSC_NO_MATCH;
1647 }
1648
1649 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
1650 }
1651
1652 /* Packets with 'SYN' should bypass, other flag should be sent after drain
1653 * to prevent out of order */
1654 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain,
1655 struct tcp_header *tcp)
1656 {
1657 uint16_t tcp_hdr;
1658 uint16_t tcp_flag;
1659
1660 tcp_flag = htons(tcp->th_offset_flags);
1661 tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10;
1662 tcp_flag &= VIRTIO_NET_TCP_FLAG;
1663 tcp_flag = htons(tcp->th_offset_flags) & 0x3F;
1664 if (tcp_flag & TH_SYN) {
1665 chain->stat.tcp_syn++;
1666 return RSC_BYPASS;
1667 }
1668
1669 if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) {
1670 chain->stat.tcp_ctrl_drain++;
1671 return RSC_FINAL;
1672 }
1673
1674 if (tcp_hdr > sizeof(struct tcp_header)) {
1675 chain->stat.tcp_all_opt++;
1676 return RSC_FINAL;
1677 }
1678
1679 return RSC_CANDIDATE;
1680 }
1681
1682 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain,
1683 NetClientState *nc,
1684 const uint8_t *buf, size_t size,
1685 VirtioNetRscUnit *unit)
1686 {
1687 int ret;
1688 VirtioNetRscSeg *seg, *nseg;
1689
1690 if (QTAILQ_EMPTY(&chain->buffers)) {
1691 chain->stat.empty_cache++;
1692 virtio_net_rsc_cache_buf(chain, nc, buf, size);
1693 timer_mod(chain->drain_timer,
1694 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
1695 return size;
1696 }
1697
1698 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
1699 if (chain->proto == ETH_P_IP) {
1700 ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
1701 } else {
1702 ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit);
1703 }
1704
1705 if (ret == RSC_FINAL) {
1706 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
1707 /* Send failed */
1708 chain->stat.final_failed++;
1709 return 0;
1710 }
1711
1712 /* Send current packet */
1713 return virtio_net_do_receive(nc, buf, size);
1714 } else if (ret == RSC_NO_MATCH) {
1715 continue;
1716 } else {
1717 /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
1718 seg->is_coalesced = 1;
1719 return size;
1720 }
1721 }
1722
1723 chain->stat.no_match_cache++;
1724 virtio_net_rsc_cache_buf(chain, nc, buf, size);
1725 return size;
1726 }
1727
1728 /* Drain a connection data, this is to avoid out of order segments */
1729 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain,
1730 NetClientState *nc,
1731 const uint8_t *buf, size_t size,
1732 uint16_t ip_start, uint16_t ip_size,
1733 uint16_t tcp_port)
1734 {
1735 VirtioNetRscSeg *seg, *nseg;
1736 uint32_t ppair1, ppair2;
1737
1738 ppair1 = *(uint32_t *)(buf + tcp_port);
1739 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
1740 ppair2 = *(uint32_t *)(seg->buf + tcp_port);
1741 if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size)
1742 || (ppair1 != ppair2)) {
1743 continue;
1744 }
1745 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
1746 chain->stat.drain_failed++;
1747 }
1748
1749 break;
1750 }
1751
1752 return virtio_net_do_receive(nc, buf, size);
1753 }
1754
1755 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain,
1756 struct ip_header *ip,
1757 const uint8_t *buf, size_t size)
1758 {
1759 uint16_t ip_len;
1760
1761 /* Not an ipv4 packet */
1762 if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) {
1763 chain->stat.ip_option++;
1764 return RSC_BYPASS;
1765 }
1766
1767 /* Don't handle packets with ip option */
1768 if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) {
1769 chain->stat.ip_option++;
1770 return RSC_BYPASS;
1771 }
1772
1773 if (ip->ip_p != IPPROTO_TCP) {
1774 chain->stat.bypass_not_tcp++;
1775 return RSC_BYPASS;
1776 }
1777
1778 /* Don't handle packets with ip fragment */
1779 if (!(htons(ip->ip_off) & IP_DF)) {
1780 chain->stat.ip_frag++;
1781 return RSC_BYPASS;
1782 }
1783
1784 /* Don't handle packets with ecn flag */
1785 if (IPTOS_ECN(ip->ip_tos)) {
1786 chain->stat.ip_ecn++;
1787 return RSC_BYPASS;
1788 }
1789
1790 ip_len = htons(ip->ip_len);
1791 if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header))
1792 || ip_len > (size - chain->n->guest_hdr_len -
1793 sizeof(struct eth_header))) {
1794 chain->stat.ip_hacked++;
1795 return RSC_BYPASS;
1796 }
1797
1798 return RSC_CANDIDATE;
1799 }
1800
1801 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain,
1802 NetClientState *nc,
1803 const uint8_t *buf, size_t size)
1804 {
1805 int32_t ret;
1806 uint16_t hdr_len;
1807 VirtioNetRscUnit unit;
1808
1809 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
1810
1811 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)
1812 + sizeof(struct tcp_header))) {
1813 chain->stat.bypass_not_tcp++;
1814 return virtio_net_do_receive(nc, buf, size);
1815 }
1816
1817 virtio_net_rsc_extract_unit4(chain, buf, &unit);
1818 if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size)
1819 != RSC_CANDIDATE) {
1820 return virtio_net_do_receive(nc, buf, size);
1821 }
1822
1823 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
1824 if (ret == RSC_BYPASS) {
1825 return virtio_net_do_receive(nc, buf, size);
1826 } else if (ret == RSC_FINAL) {
1827 return virtio_net_rsc_drain_flow(chain, nc, buf, size,
1828 ((hdr_len + sizeof(struct eth_header)) + 12),
1829 VIRTIO_NET_IP4_ADDR_SIZE,
1830 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header));
1831 }
1832
1833 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
1834 }
1835
1836 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain,
1837 struct ip6_header *ip6,
1838 const uint8_t *buf, size_t size)
1839 {
1840 uint16_t ip_len;
1841
1842 if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4)
1843 != IP_HEADER_VERSION_6) {
1844 return RSC_BYPASS;
1845 }
1846
1847 /* Both option and protocol is checked in this */
1848 if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) {
1849 chain->stat.bypass_not_tcp++;
1850 return RSC_BYPASS;
1851 }
1852
1853 ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
1854 if (ip_len < sizeof(struct tcp_header) ||
1855 ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header)
1856 - sizeof(struct ip6_header))) {
1857 chain->stat.ip_hacked++;
1858 return RSC_BYPASS;
1859 }
1860
1861 /* Don't handle packets with ecn flag */
1862 if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) {
1863 chain->stat.ip_ecn++;
1864 return RSC_BYPASS;
1865 }
1866
1867 return RSC_CANDIDATE;
1868 }
1869
1870 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc,
1871 const uint8_t *buf, size_t size)
1872 {
1873 int32_t ret;
1874 uint16_t hdr_len;
1875 VirtioNetRscChain *chain;
1876 VirtioNetRscUnit unit;
1877
1878 chain = (VirtioNetRscChain *)opq;
1879 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
1880
1881 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header)
1882 + sizeof(tcp_header))) {
1883 return virtio_net_do_receive(nc, buf, size);
1884 }
1885
1886 virtio_net_rsc_extract_unit6(chain, buf, &unit);
1887 if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain,
1888 unit.ip, buf, size)) {
1889 return virtio_net_do_receive(nc, buf, size);
1890 }
1891
1892 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
1893 if (ret == RSC_BYPASS) {
1894 return virtio_net_do_receive(nc, buf, size);
1895 } else if (ret == RSC_FINAL) {
1896 return virtio_net_rsc_drain_flow(chain, nc, buf, size,
1897 ((hdr_len + sizeof(struct eth_header)) + 8),
1898 VIRTIO_NET_IP6_ADDR_SIZE,
1899 hdr_len + sizeof(struct eth_header)
1900 + sizeof(struct ip6_header));
1901 }
1902
1903 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
1904 }
1905
1906 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n,
1907 NetClientState *nc,
1908 uint16_t proto)
1909 {
1910 VirtioNetRscChain *chain;
1911
1912 if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) {
1913 return NULL;
1914 }
1915
1916 QTAILQ_FOREACH(chain, &n->rsc_chains, next) {
1917 if (chain->proto == proto) {
1918 return chain;
1919 }
1920 }
1921
1922 chain = g_malloc(sizeof(*chain));
1923 chain->n = n;
1924 chain->proto = proto;
1925 if (proto == (uint16_t)ETH_P_IP) {
1926 chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
1927 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1928 } else {
1929 chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD;
1930 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1931 }
1932 chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST,
1933 virtio_net_rsc_purge, chain);
1934 memset(&chain->stat, 0, sizeof(chain->stat));
1935
1936 QTAILQ_INIT(&chain->buffers);
1937 QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next);
1938
1939 return chain;
1940 }
1941
1942 static ssize_t virtio_net_rsc_receive(NetClientState *nc,
1943 const uint8_t *buf,
1944 size_t size)
1945 {
1946 uint16_t proto;
1947 VirtioNetRscChain *chain;
1948 struct eth_header *eth;
1949 VirtIONet *n;
1950
1951 n = qemu_get_nic_opaque(nc);
1952 if (size < (n->host_hdr_len + sizeof(struct eth_header))) {
1953 return virtio_net_do_receive(nc, buf, size);
1954 }
1955
1956 eth = (struct eth_header *)(buf + n->guest_hdr_len);
1957 proto = htons(eth->h_proto);
1958
1959 chain = virtio_net_rsc_lookup_chain(n, nc, proto);
1960 if (chain) {
1961 chain->stat.received++;
1962 if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) {
1963 return virtio_net_rsc_receive4(chain, nc, buf, size);
1964 } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) {
1965 return virtio_net_rsc_receive6(chain, nc, buf, size);
1966 }
1967 }
1968 return virtio_net_do_receive(nc, buf, size);
1969 }
1970
1971 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf,
1972 size_t size)
1973 {
1974 VirtIONet *n = qemu_get_nic_opaque(nc);
1975 if ((n->rsc4_enabled || n->rsc6_enabled)) {
1976 return virtio_net_rsc_receive(nc, buf, size);
1977 } else {
1978 return virtio_net_do_receive(nc, buf, size);
1979 }
1980 }
1981
1982 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1983
1984 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1985 {
1986 VirtIONet *n = qemu_get_nic_opaque(nc);
1987 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1988 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1989
1990 virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
1991 virtio_notify(vdev, q->tx_vq);
1992
1993 g_free(q->async_tx.elem);
1994 q->async_tx.elem = NULL;
1995
1996 virtio_queue_set_notification(q->tx_vq, 1);
1997 virtio_net_flush_tx(q);
1998 }
1999
2000 /* TX */
2001 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
2002 {
2003 VirtIONet *n = q->n;
2004 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2005 VirtQueueElement *elem;
2006 int32_t num_packets = 0;
2007 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
2008 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2009 return num_packets;
2010 }
2011
2012 if (q->async_tx.elem) {
2013 virtio_queue_set_notification(q->tx_vq, 0);
2014 return num_packets;
2015 }
2016
2017 for (;;) {
2018 ssize_t ret;
2019 unsigned int out_num;
2020 struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
2021 struct virtio_net_hdr_mrg_rxbuf mhdr;
2022
2023 elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
2024 if (!elem) {
2025 break;
2026 }
2027
2028 out_num = elem->out_num;
2029 out_sg = elem->out_sg;
2030 if (out_num < 1) {
2031 virtio_error(vdev, "virtio-net header not in first element");
2032 virtqueue_detach_element(q->tx_vq, elem, 0);
2033 g_free(elem);
2034 return -EINVAL;
2035 }
2036
2037 if (n->has_vnet_hdr) {
2038 if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
2039 n->guest_hdr_len) {
2040 virtio_error(vdev, "virtio-net header incorrect");
2041 virtqueue_detach_element(q->tx_vq, elem, 0);
2042 g_free(elem);
2043 return -EINVAL;
2044 }
2045 if (n->needs_vnet_hdr_swap) {
2046 virtio_net_hdr_swap(vdev, (void *) &mhdr);
2047 sg2[0].iov_base = &mhdr;
2048 sg2[0].iov_len = n->guest_hdr_len;
2049 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
2050 out_sg, out_num,
2051 n->guest_hdr_len, -1);
2052 if (out_num == VIRTQUEUE_MAX_SIZE) {
2053 goto drop;
2054 }
2055 out_num += 1;
2056 out_sg = sg2;
2057 }
2058 }
2059 /*
2060 * If host wants to see the guest header as is, we can
2061 * pass it on unchanged. Otherwise, copy just the parts
2062 * that host is interested in.
2063 */
2064 assert(n->host_hdr_len <= n->guest_hdr_len);
2065 if (n->host_hdr_len != n->guest_hdr_len) {
2066 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
2067 out_sg, out_num,
2068 0, n->host_hdr_len);
2069 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
2070 out_sg, out_num,
2071 n->guest_hdr_len, -1);
2072 out_num = sg_num;
2073 out_sg = sg;
2074 }
2075
2076 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
2077 out_sg, out_num, virtio_net_tx_complete);
2078 if (ret == 0) {
2079 virtio_queue_set_notification(q->tx_vq, 0);
2080 q->async_tx.elem = elem;
2081 return -EBUSY;
2082 }
2083
2084 drop:
2085 virtqueue_push(q->tx_vq, elem, 0);
2086 virtio_notify(vdev, q->tx_vq);
2087 g_free(elem);
2088
2089 if (++num_packets >= n->tx_burst) {
2090 break;
2091 }
2092 }
2093 return num_packets;
2094 }
2095
2096 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
2097 {
2098 VirtIONet *n = VIRTIO_NET(vdev);
2099 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2100
2101 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2102 virtio_net_drop_tx_queue_data(vdev, vq);
2103 return;
2104 }
2105
2106 /* This happens when device was stopped but VCPU wasn't. */
2107 if (!vdev->vm_running) {
2108 q->tx_waiting = 1;
2109 return;
2110 }
2111
2112 if (q->tx_waiting) {
2113 virtio_queue_set_notification(vq, 1);
2114 timer_del(q->tx_timer);
2115 q->tx_waiting = 0;
2116 if (virtio_net_flush_tx(q) == -EINVAL) {
2117 return;
2118 }
2119 } else {
2120 timer_mod(q->tx_timer,
2121 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2122 q->tx_waiting = 1;
2123 virtio_queue_set_notification(vq, 0);
2124 }
2125 }
2126
2127 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
2128 {
2129 VirtIONet *n = VIRTIO_NET(vdev);
2130 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2131
2132 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2133 virtio_net_drop_tx_queue_data(vdev, vq);
2134 return;
2135 }
2136
2137 if (unlikely(q->tx_waiting)) {
2138 return;
2139 }
2140 q->tx_waiting = 1;
2141 /* This happens when device was stopped but VCPU wasn't. */
2142 if (!vdev->vm_running) {
2143 return;
2144 }
2145 virtio_queue_set_notification(vq, 0);
2146 qemu_bh_schedule(q->tx_bh);
2147 }
2148
2149 static void virtio_net_tx_timer(void *opaque)
2150 {
2151 VirtIONetQueue *q = opaque;
2152 VirtIONet *n = q->n;
2153 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2154 /* This happens when device was stopped but BH wasn't. */
2155 if (!vdev->vm_running) {
2156 /* Make sure tx waiting is set, so we'll run when restarted. */
2157 assert(q->tx_waiting);
2158 return;
2159 }
2160
2161 q->tx_waiting = 0;
2162
2163 /* Just in case the driver is not ready on more */
2164 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2165 return;
2166 }
2167
2168 virtio_queue_set_notification(q->tx_vq, 1);
2169 virtio_net_flush_tx(q);
2170 }
2171
2172 static void virtio_net_tx_bh(void *opaque)
2173 {
2174 VirtIONetQueue *q = opaque;
2175 VirtIONet *n = q->n;
2176 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2177 int32_t ret;
2178
2179 /* This happens when device was stopped but BH wasn't. */
2180 if (!vdev->vm_running) {
2181 /* Make sure tx waiting is set, so we'll run when restarted. */
2182 assert(q->tx_waiting);
2183 return;
2184 }
2185
2186 q->tx_waiting = 0;
2187
2188 /* Just in case the driver is not ready on more */
2189 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
2190 return;
2191 }
2192
2193 ret = virtio_net_flush_tx(q);
2194 if (ret == -EBUSY || ret == -EINVAL) {
2195 return; /* Notification re-enable handled by tx_complete or device
2196 * broken */
2197 }
2198
2199 /* If we flush a full burst of packets, assume there are
2200 * more coming and immediately reschedule */
2201 if (ret >= n->tx_burst) {
2202 qemu_bh_schedule(q->tx_bh);
2203 q->tx_waiting = 1;
2204 return;
2205 }
2206
2207 /* If less than a full burst, re-enable notification and flush
2208 * anything that may have come in while we weren't looking. If
2209 * we find something, assume the guest is still active and reschedule */
2210 virtio_queue_set_notification(q->tx_vq, 1);
2211 ret = virtio_net_flush_tx(q);
2212 if (ret == -EINVAL) {
2213 return;
2214 } else if (ret > 0) {
2215 virtio_queue_set_notification(q->tx_vq, 0);
2216 qemu_bh_schedule(q->tx_bh);
2217 q->tx_waiting = 1;
2218 }
2219 }
2220
2221 static void virtio_net_add_queue(VirtIONet *n, int index)
2222 {
2223 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2224
2225 n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
2226 virtio_net_handle_rx);
2227
2228 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
2229 n->vqs[index].tx_vq =
2230 virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2231 virtio_net_handle_tx_timer);
2232 n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
2233 virtio_net_tx_timer,
2234 &n->vqs[index]);
2235 } else {
2236 n->vqs[index].tx_vq =
2237 virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2238 virtio_net_handle_tx_bh);
2239 n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
2240 }
2241
2242 n->vqs[index].tx_waiting = 0;
2243 n->vqs[index].n = n;
2244 }
2245
2246 static void virtio_net_del_queue(VirtIONet *n, int index)
2247 {
2248 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2249 VirtIONetQueue *q = &n->vqs[index];
2250 NetClientState *nc = qemu_get_subqueue(n->nic, index);
2251
2252 qemu_purge_queued_packets(nc);
2253
2254 virtio_del_queue(vdev, index * 2);
2255 if (q->tx_timer) {
2256 timer_del(q->tx_timer);
2257 timer_free(q->tx_timer);
2258 q->tx_timer = NULL;
2259 } else {
2260 qemu_bh_delete(q->tx_bh);
2261 q->tx_bh = NULL;
2262 }
2263 q->tx_waiting = 0;
2264 virtio_del_queue(vdev, index * 2 + 1);
2265 }
2266
2267 static void virtio_net_change_num_queues(VirtIONet *n, int new_max_queues)
2268 {
2269 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2270 int old_num_queues = virtio_get_num_queues(vdev);
2271 int new_num_queues = new_max_queues * 2 + 1;
2272 int i;
2273
2274 assert(old_num_queues >= 3);
2275 assert(old_num_queues % 2 == 1);
2276
2277 if (old_num_queues == new_num_queues) {
2278 return;
2279 }
2280
2281 /*
2282 * We always need to remove and add ctrl vq if
2283 * old_num_queues != new_num_queues. Remove ctrl_vq first,
2284 * and then we only enter one of the following too loops.
2285 */
2286 virtio_del_queue(vdev, old_num_queues - 1);
2287
2288 for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
2289 /* new_num_queues < old_num_queues */
2290 virtio_net_del_queue(n, i / 2);
2291 }
2292
2293 for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
2294 /* new_num_queues > old_num_queues */
2295 virtio_net_add_queue(n, i / 2);
2296 }
2297
2298 /* add ctrl_vq last */
2299 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
2300 }
2301
2302 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
2303 {
2304 int max = multiqueue ? n->max_queues : 1;
2305
2306 n->multiqueue = multiqueue;
2307 virtio_net_change_num_queues(n, max);
2308
2309 virtio_net_set_queues(n);
2310 }
2311
2312 static int virtio_net_post_load_device(void *opaque, int version_id)
2313 {
2314 VirtIONet *n = opaque;
2315 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2316 int i, link_down;
2317
2318 trace_virtio_net_post_load_device();
2319 virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
2320 virtio_vdev_has_feature(vdev,
2321 VIRTIO_F_VERSION_1));
2322
2323 /* MAC_TABLE_ENTRIES may be different from the saved image */
2324 if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
2325 n->mac_table.in_use = 0;
2326 }
2327
2328 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
2329 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
2330 }
2331
2332 if (peer_has_vnet_hdr(n)) {
2333 virtio_net_apply_guest_offloads(n);
2334 }
2335
2336 virtio_net_set_queues(n);
2337
2338 /* Find the first multicast entry in the saved MAC filter */
2339 for (i = 0; i < n->mac_table.in_use; i++) {
2340 if (n->mac_table.macs[i * ETH_ALEN] & 1) {
2341 break;
2342 }
2343 }
2344 n->mac_table.first_multi = i;
2345
2346 /* nc.link_down can't be migrated, so infer link_down according
2347 * to link status bit in n->status */
2348 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
2349 for (i = 0; i < n->max_queues; i++) {
2350 qemu_get_subqueue(n->nic, i)->link_down = link_down;
2351 }
2352
2353 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
2354 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
2355 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
2356 QEMU_CLOCK_VIRTUAL,
2357 virtio_net_announce_timer, n);
2358 if (n->announce_timer.round) {
2359 timer_mod(n->announce_timer.tm,
2360 qemu_clock_get_ms(n->announce_timer.type));
2361 } else {
2362 qemu_announce_timer_del(&n->announce_timer);
2363 }
2364 }
2365
2366 return 0;
2367 }
2368
2369 /* tx_waiting field of a VirtIONetQueue */
2370 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = {
2371 .name = "virtio-net-queue-tx_waiting",
2372 .fields = (VMStateField[]) {
2373 VMSTATE_UINT32(tx_waiting, VirtIONetQueue),
2374 VMSTATE_END_OF_LIST()
2375 },
2376 };
2377
2378 static bool max_queues_gt_1(void *opaque, int version_id)
2379 {
2380 return VIRTIO_NET(opaque)->max_queues > 1;
2381 }
2382
2383 static bool has_ctrl_guest_offloads(void *opaque, int version_id)
2384 {
2385 return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque),
2386 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
2387 }
2388
2389 static bool mac_table_fits(void *opaque, int version_id)
2390 {
2391 return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES;
2392 }
2393
2394 static bool mac_table_doesnt_fit(void *opaque, int version_id)
2395 {
2396 return !mac_table_fits(opaque, version_id);
2397 }
2398
2399 /* This temporary type is shared by all the WITH_TMP methods
2400 * although only some fields are used by each.
2401 */
2402 struct VirtIONetMigTmp {
2403 VirtIONet *parent;
2404 VirtIONetQueue *vqs_1;
2405 uint16_t curr_queues_1;
2406 uint8_t has_ufo;
2407 uint32_t has_vnet_hdr;
2408 };
2409
2410 /* The 2nd and subsequent tx_waiting flags are loaded later than
2411 * the 1st entry in the queues and only if there's more than one
2412 * entry. We use the tmp mechanism to calculate a temporary
2413 * pointer and count and also validate the count.
2414 */
2415
2416 static int virtio_net_tx_waiting_pre_save(void *opaque)
2417 {
2418 struct VirtIONetMigTmp *tmp = opaque;
2419
2420 tmp->vqs_1 = tmp->parent->vqs + 1;
2421 tmp->curr_queues_1 = tmp->parent->curr_queues - 1;
2422 if (tmp->parent->curr_queues == 0) {
2423 tmp->curr_queues_1 = 0;
2424 }
2425
2426 return 0;
2427 }
2428
2429 static int virtio_net_tx_waiting_pre_load(void *opaque)
2430 {
2431 struct VirtIONetMigTmp *tmp = opaque;
2432
2433 /* Reuse the pointer setup from save */
2434 virtio_net_tx_waiting_pre_save(opaque);
2435
2436 if (tmp->parent->curr_queues > tmp->parent->max_queues) {
2437 error_report("virtio-net: curr_queues %x > max_queues %x",
2438 tmp->parent->curr_queues, tmp->parent->max_queues);
2439
2440 return -EINVAL;
2441 }
2442
2443 return 0; /* all good */
2444 }
2445
2446 static const VMStateDescription vmstate_virtio_net_tx_waiting = {
2447 .name = "virtio-net-tx_waiting",
2448 .pre_load = virtio_net_tx_waiting_pre_load,
2449 .pre_save = virtio_net_tx_waiting_pre_save,
2450 .fields = (VMStateField[]) {
2451 VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp,
2452 curr_queues_1,
2453 vmstate_virtio_net_queue_tx_waiting,
2454 struct VirtIONetQueue),
2455 VMSTATE_END_OF_LIST()
2456 },
2457 };
2458
2459 /* the 'has_ufo' flag is just tested; if the incoming stream has the
2460 * flag set we need to check that we have it
2461 */
2462 static int virtio_net_ufo_post_load(void *opaque, int version_id)
2463 {
2464 struct VirtIONetMigTmp *tmp = opaque;
2465
2466 if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) {
2467 error_report("virtio-net: saved image requires TUN_F_UFO support");
2468 return -EINVAL;
2469 }
2470
2471 return 0;
2472 }
2473
2474 static int virtio_net_ufo_pre_save(void *opaque)
2475 {
2476 struct VirtIONetMigTmp *tmp = opaque;
2477
2478 tmp->has_ufo = tmp->parent->has_ufo;
2479
2480 return 0;
2481 }
2482
2483 static const VMStateDescription vmstate_virtio_net_has_ufo = {
2484 .name = "virtio-net-ufo",
2485 .post_load = virtio_net_ufo_post_load,
2486 .pre_save = virtio_net_ufo_pre_save,
2487 .fields = (VMStateField[]) {
2488 VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp),
2489 VMSTATE_END_OF_LIST()
2490 },
2491 };
2492
2493 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
2494 * flag set we need to check that we have it
2495 */
2496 static int virtio_net_vnet_post_load(void *opaque, int version_id)
2497 {
2498 struct VirtIONetMigTmp *tmp = opaque;
2499
2500 if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) {
2501 error_report("virtio-net: saved image requires vnet_hdr=on");
2502 return -EINVAL;
2503 }
2504
2505 return 0;
2506 }
2507
2508 static int virtio_net_vnet_pre_save(void *opaque)
2509 {
2510 struct VirtIONetMigTmp *tmp = opaque;
2511
2512 tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr;
2513
2514 return 0;
2515 }
2516
2517 static const VMStateDescription vmstate_virtio_net_has_vnet = {
2518 .name = "virtio-net-vnet",
2519 .post_load = virtio_net_vnet_post_load,
2520 .pre_save = virtio_net_vnet_pre_save,
2521 .fields = (VMStateField[]) {
2522 VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp),
2523 VMSTATE_END_OF_LIST()
2524 },
2525 };
2526
2527 static const VMStateDescription vmstate_virtio_net_device = {
2528 .name = "virtio-net-device",
2529 .version_id = VIRTIO_NET_VM_VERSION,
2530 .minimum_version_id = VIRTIO_NET_VM_VERSION,
2531 .post_load = virtio_net_post_load_device,
2532 .fields = (VMStateField[]) {
2533 VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN),
2534 VMSTATE_STRUCT_POINTER(vqs, VirtIONet,
2535 vmstate_virtio_net_queue_tx_waiting,
2536 VirtIONetQueue),
2537 VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet),
2538 VMSTATE_UINT16(status, VirtIONet),
2539 VMSTATE_UINT8(promisc, VirtIONet),
2540 VMSTATE_UINT8(allmulti, VirtIONet),
2541 VMSTATE_UINT32(mac_table.in_use, VirtIONet),
2542
2543 /* Guarded pair: If it fits we load it, else we throw it away
2544 * - can happen if source has a larger MAC table.; post-load
2545 * sets flags in this case.
2546 */
2547 VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet,
2548 0, mac_table_fits, mac_table.in_use,
2549 ETH_ALEN),
2550 VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0,
2551 mac_table.in_use, ETH_ALEN),
2552
2553 /* Note: This is an array of uint32's that's always been saved as a
2554 * buffer; hold onto your endiannesses; it's actually used as a bitmap
2555 * but based on the uint.
2556 */
2557 VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3),
2558 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
2559 vmstate_virtio_net_has_vnet),
2560 VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet),
2561 VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet),
2562 VMSTATE_UINT8(alluni, VirtIONet),
2563 VMSTATE_UINT8(nomulti, VirtIONet),
2564 VMSTATE_UINT8(nouni, VirtIONet),
2565 VMSTATE_UINT8(nobcast, VirtIONet),
2566 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
2567 vmstate_virtio_net_has_ufo),
2568 VMSTATE_SINGLE_TEST(max_queues, VirtIONet, max_queues_gt_1, 0,
2569 vmstate_info_uint16_equal, uint16_t),
2570 VMSTATE_UINT16_TEST(curr_queues, VirtIONet, max_queues_gt_1),
2571 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
2572 vmstate_virtio_net_tx_waiting),
2573 VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet,
2574 has_ctrl_guest_offloads),
2575 VMSTATE_END_OF_LIST()
2576 },
2577 };
2578
2579 static NetClientInfo net_virtio_info = {
2580 .type = NET_CLIENT_DRIVER_NIC,
2581 .size = sizeof(NICState),
2582 .can_receive = virtio_net_can_receive,
2583 .receive = virtio_net_receive,
2584 .link_status_changed = virtio_net_set_link_status,
2585 .query_rx_filter = virtio_net_query_rxfilter,
2586 .announce = virtio_net_announce,
2587 };
2588
2589 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
2590 {
2591 VirtIONet *n = VIRTIO_NET(vdev);
2592 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
2593 assert(n->vhost_started);
2594 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
2595 }
2596
2597 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
2598 bool mask)
2599 {
2600 VirtIONet *n = VIRTIO_NET(vdev);
2601 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
2602 assert(n->vhost_started);
2603 vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
2604 vdev, idx, mask);
2605 }
2606
2607 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
2608 {
2609 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
2610
2611 n->config_size = virtio_feature_get_config_size(feature_sizes,
2612 host_features);
2613 }
2614
2615 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
2616 const char *type)
2617 {
2618 /*
2619 * The name can be NULL, the netclient name will be type.x.
2620 */
2621 assert(type != NULL);
2622
2623 g_free(n->netclient_name);
2624 g_free(n->netclient_type);
2625 n->netclient_name = g_strdup(name);
2626 n->netclient_type = g_strdup(type);
2627 }
2628
2629 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
2630 {
2631 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2632 VirtIONet *n = VIRTIO_NET(dev);
2633 NetClientState *nc;
2634 int i;
2635
2636 if (n->net_conf.mtu) {
2637 n->host_features |= (1ULL << VIRTIO_NET_F_MTU);
2638 }
2639
2640 if (n->net_conf.duplex_str) {
2641 if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) {
2642 n->net_conf.duplex = DUPLEX_HALF;
2643 } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) {
2644 n->net_conf.duplex = DUPLEX_FULL;
2645 } else {
2646 error_setg(errp, "'duplex' must be 'half' or 'full'");
2647 }
2648 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
2649 } else {
2650 n->net_conf.duplex = DUPLEX_UNKNOWN;
2651 }
2652
2653 if (n->net_conf.speed < SPEED_UNKNOWN) {
2654 error_setg(errp, "'speed' must be between 0 and INT_MAX");
2655 } else if (n->net_conf.speed >= 0) {
2656 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
2657 }
2658
2659 virtio_net_set_config_size(n, n->host_features);
2660 virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
2661
2662 /*
2663 * We set a lower limit on RX queue size to what it always was.
2664 * Guests that want a smaller ring can always resize it without
2665 * help from us (using virtio 1 and up).
2666 */
2667 if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
2668 n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
2669 !is_power_of_2(n->net_conf.rx_queue_size)) {
2670 error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
2671 "must be a power of 2 between %d and %d.",
2672 n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
2673 VIRTQUEUE_MAX_SIZE);
2674 virtio_cleanup(vdev);
2675 return;
2676 }
2677
2678 if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE ||
2679 n->net_conf.tx_queue_size > VIRTQUEUE_MAX_SIZE ||
2680 !is_power_of_2(n->net_conf.tx_queue_size)) {
2681 error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), "
2682 "must be a power of 2 between %d and %d",
2683 n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE,
2684 VIRTQUEUE_MAX_SIZE);
2685 virtio_cleanup(vdev);
2686 return;
2687 }
2688
2689 n->max_queues = MAX(n->nic_conf.peers.queues, 1);
2690 if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) {
2691 error_setg(errp, "Invalid number of queues (= %" PRIu32 "), "
2692 "must be a positive integer less than %d.",
2693 n->max_queues, (VIRTIO_QUEUE_MAX - 1) / 2);
2694 virtio_cleanup(vdev);
2695 return;
2696 }
2697 n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
2698 n->curr_queues = 1;
2699 n->tx_timeout = n->net_conf.txtimer;
2700
2701 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
2702 && strcmp(n->net_conf.tx, "bh")) {
2703 warn_report("virtio-net: "
2704 "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
2705 n->net_conf.tx);
2706 error_printf("Defaulting to \"bh\"");
2707 }
2708
2709 n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n),
2710 n->net_conf.tx_queue_size);
2711
2712 for (i = 0; i < n->max_queues; i++) {
2713 virtio_net_add_queue(n, i);
2714 }
2715
2716 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
2717 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
2718 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
2719 n->status = VIRTIO_NET_S_LINK_UP;
2720 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
2721 QEMU_CLOCK_VIRTUAL,
2722 virtio_net_announce_timer, n);
2723 n->announce_timer.round = 0;
2724
2725 if (n->netclient_type) {
2726 /*
2727 * Happen when virtio_net_set_netclient_name has been called.
2728 */
2729 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
2730 n->netclient_type, n->netclient_name, n);
2731 } else {
2732 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
2733 object_get_typename(OBJECT(dev)), dev->id, n);
2734 }
2735
2736 peer_test_vnet_hdr(n);
2737 if (peer_has_vnet_hdr(n)) {
2738 for (i = 0; i < n->max_queues; i++) {
2739 qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
2740 }
2741 n->host_hdr_len = sizeof(struct virtio_net_hdr);
2742 } else {
2743 n->host_hdr_len = 0;
2744 }
2745
2746 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
2747
2748 n->vqs[0].tx_waiting = 0;
2749 n->tx_burst = n->net_conf.txburst;
2750 virtio_net_set_mrg_rx_bufs(n, 0, 0);
2751 n->promisc = 1; /* for compatibility */
2752
2753 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
2754
2755 n->vlans = g_malloc0(MAX_VLAN >> 3);
2756
2757 nc = qemu_get_queue(n->nic);
2758 nc->rxfilter_notify_enabled = 1;
2759
2760 QTAILQ_INIT(&n->rsc_chains);
2761 n->qdev = dev;
2762 }
2763
2764 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
2765 {
2766 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2767 VirtIONet *n = VIRTIO_NET(dev);
2768 int i, max_queues;
2769
2770 /* This will stop vhost backend if appropriate. */
2771 virtio_net_set_status(vdev, 0);
2772
2773 g_free(n->netclient_name);
2774 n->netclient_name = NULL;
2775 g_free(n->netclient_type);
2776 n->netclient_type = NULL;
2777
2778 g_free(n->mac_table.macs);
2779 g_free(n->vlans);
2780
2781 max_queues = n->multiqueue ? n->max_queues : 1;
2782 for (i = 0; i < max_queues; i++) {
2783 virtio_net_del_queue(n, i);
2784 }
2785
2786 qemu_announce_timer_del(&n->announce_timer);
2787 g_free(n->vqs);
2788 qemu_del_nic(n->nic);
2789 virtio_net_rsc_cleanup(n);
2790 virtio_cleanup(vdev);
2791 }
2792
2793 static void virtio_net_instance_init(Object *obj)
2794 {
2795 VirtIONet *n = VIRTIO_NET(obj);
2796
2797 /*
2798 * The default config_size is sizeof(struct virtio_net_config).
2799 * Can be overriden with virtio_net_set_config_size.
2800 */
2801 n->config_size = sizeof(struct virtio_net_config);
2802 device_add_bootindex_property(obj, &n->nic_conf.bootindex,
2803 "bootindex", "/ethernet-phy@0",
2804 DEVICE(n), NULL);
2805 }
2806
2807 static int virtio_net_pre_save(void *opaque)
2808 {
2809 VirtIONet *n = opaque;
2810
2811 /* At this point, backend must be stopped, otherwise
2812 * it might keep writing to memory. */
2813 assert(!n->vhost_started);
2814
2815 return 0;
2816 }
2817
2818 static const VMStateDescription vmstate_virtio_net = {
2819 .name = "virtio-net",
2820 .minimum_version_id = VIRTIO_NET_VM_VERSION,
2821 .version_id = VIRTIO_NET_VM_VERSION,
2822 .fields = (VMStateField[]) {
2823 VMSTATE_VIRTIO_DEVICE,
2824 VMSTATE_END_OF_LIST()
2825 },
2826 .pre_save = virtio_net_pre_save,
2827 };
2828
2829 static Property virtio_net_properties[] = {
2830 DEFINE_PROP_BIT64("csum", VirtIONet, host_features,
2831 VIRTIO_NET_F_CSUM, true),
2832 DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features,
2833 VIRTIO_NET_F_GUEST_CSUM, true),
2834 DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
2835 DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features,
2836 VIRTIO_NET_F_GUEST_TSO4, true),
2837 DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features,
2838 VIRTIO_NET_F_GUEST_TSO6, true),
2839 DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features,
2840 VIRTIO_NET_F_GUEST_ECN, true),
2841 DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features,
2842 VIRTIO_NET_F_GUEST_UFO, true),
2843 DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features,
2844 VIRTIO_NET_F_GUEST_ANNOUNCE, true),
2845 DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features,
2846 VIRTIO_NET_F_HOST_TSO4, true),
2847 DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features,
2848 VIRTIO_NET_F_HOST_TSO6, true),
2849 DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features,
2850 VIRTIO_NET_F_HOST_ECN, true),
2851 DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features,
2852 VIRTIO_NET_F_HOST_UFO, true),
2853 DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features,
2854 VIRTIO_NET_F_MRG_RXBUF, true),
2855 DEFINE_PROP_BIT64("status", VirtIONet, host_features,
2856 VIRTIO_NET_F_STATUS, true),
2857 DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features,
2858 VIRTIO_NET_F_CTRL_VQ, true),
2859 DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features,
2860 VIRTIO_NET_F_CTRL_RX, true),
2861 DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features,
2862 VIRTIO_NET_F_CTRL_VLAN, true),
2863 DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features,
2864 VIRTIO_NET_F_CTRL_RX_EXTRA, true),
2865 DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features,
2866 VIRTIO_NET_F_CTRL_MAC_ADDR, true),
2867 DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features,
2868 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
2869 DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
2870 DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features,
2871 VIRTIO_NET_F_RSC_EXT, false),
2872 DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout,
2873 VIRTIO_NET_RSC_DEFAULT_INTERVAL),
2874 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
2875 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
2876 TX_TIMER_INTERVAL),
2877 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
2878 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
2879 DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
2880 VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
2881 DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size,
2882 VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE),
2883 DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
2884 DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend,
2885 true),
2886 DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN),
2887 DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str),
2888 DEFINE_PROP_END_OF_LIST(),
2889 };
2890
2891 static void virtio_net_class_init(ObjectClass *klass, void *data)
2892 {
2893 DeviceClass *dc = DEVICE_CLASS(klass);
2894 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
2895
2896 dc->props = virtio_net_properties;
2897 dc->vmsd = &vmstate_virtio_net;
2898 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
2899 vdc->realize = virtio_net_device_realize;
2900 vdc->unrealize = virtio_net_device_unrealize;
2901 vdc->get_config = virtio_net_get_config;
2902 vdc->set_config = virtio_net_set_config;
2903 vdc->get_features = virtio_net_get_features;
2904 vdc->set_features = virtio_net_set_features;
2905 vdc->bad_features = virtio_net_bad_features;
2906 vdc->reset = virtio_net_reset;
2907 vdc->set_status = virtio_net_set_status;
2908 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
2909 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
2910 vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
2911 vdc->vmsd = &vmstate_virtio_net_device;
2912 }
2913
2914 static const TypeInfo virtio_net_info = {
2915 .name = TYPE_VIRTIO_NET,
2916 .parent = TYPE_VIRTIO_DEVICE,
2917 .instance_size = sizeof(VirtIONet),
2918 .instance_init = virtio_net_instance_init,
2919 .class_init = virtio_net_class_init,
2920 };
2921
2922 static void virtio_register_types(void)
2923 {
2924 type_register_static(&virtio_net_info);
2925 }
2926
2927 type_init(virtio_register_types)
Mirror of https://git.qemu.org/git/qemu.git