]> git.proxmox.com Git - mirror_qemu.git/blob - hw/net/virtio-net.c
projects / mirror_qemu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
docs/devel/testing: Fix typo in dockerfile path
[mirror_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 "qemu/module.h"
17 #include "hw/virtio/virtio.h"
18 #include "net/net.h"
19 #include "net/checksum.h"
20 #include "net/tap.h"
21 #include "qemu/error-report.h"
22 #include "qemu/timer.h"
23 #include "hw/virtio/virtio-net.h"
24 #include "net/vhost_net.h"
25 #include "net/announce.h"
26 #include "hw/virtio/virtio-bus.h"
27 #include "qapi/error.h"
28 #include "qapi/qapi-events-net.h"
29 #include "hw/virtio/virtio-access.h"
30 #include "migration/misc.h"
31 #include "standard-headers/linux/ethtool.h"
32 #include "trace.h"
33
34 #define VIRTIO_NET_VM_VERSION 11
35
36 #define MAC_TABLE_ENTRIES 64
37 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */
38
39 /* previously fixed value */
40 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
41 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
42
43 /* for now, only allow larger queues; with virtio-1, guest can downsize */
44 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
45 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
46
47 #define VIRTIO_NET_IP4_ADDR_SIZE 8 /* ipv4 saddr + daddr */
48
49 #define VIRTIO_NET_TCP_FLAG 0x3F
50 #define VIRTIO_NET_TCP_HDR_LENGTH 0xF000
51
52 /* IPv4 max payload, 16 bits in the header */
53 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
54 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
55
56 /* header length value in ip header without option */
57 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
58
59 #define VIRTIO_NET_IP6_ADDR_SIZE 32 /* ipv6 saddr + daddr */
60 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
61
62 /* Purge coalesced packets timer interval, This value affects the performance
63 a lot, and should be tuned carefully, '300000'(300us) is the recommended
64 value to pass the WHQL test, '50000' can gain 2x netperf throughput with
65 tso/gso/gro 'off'. */
66 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
67
68 /* temporary until standard header include it */
69 #if !defined(VIRTIO_NET_HDR_F_RSC_INFO)
70
71 #define VIRTIO_NET_HDR_F_RSC_INFO 4 /* rsc_ext data in csum_ fields */
72 #define VIRTIO_NET_F_RSC_EXT 61
73
74 static inline __virtio16 *virtio_net_rsc_ext_num_packets(
75 struct virtio_net_hdr *hdr)
76 {
77 return &hdr->csum_start;
78 }
79
80 static inline __virtio16 *virtio_net_rsc_ext_num_dupacks(
81 struct virtio_net_hdr *hdr)
82 {
83 return &hdr->csum_offset;
84 }
85
86 #endif
87
88 static VirtIOFeature feature_sizes[] = {
89 {.flags = 1ULL << VIRTIO_NET_F_MAC,
90 .end = virtio_endof(struct virtio_net_config, mac)},
91 {.flags = 1ULL << VIRTIO_NET_F_STATUS,
92 .end = virtio_endof(struct virtio_net_config, status)},
93 {.flags = 1ULL << VIRTIO_NET_F_MQ,
94 .end = virtio_endof(struct virtio_net_config, max_virtqueue_pairs)},
95 {.flags = 1ULL << VIRTIO_NET_F_MTU,
96 .end = virtio_endof(struct virtio_net_config, mtu)},
97 {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX,
98 .end = virtio_endof(struct virtio_net_config, duplex)},
99 {}
100 };
101
102 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
103 {
104 VirtIONet *n = qemu_get_nic_opaque(nc);
105
106 return &n->vqs[nc->queue_index];
107 }
108
109 static int vq2q(int queue_index)
110 {
111 return queue_index / 2;
112 }
113
114 /* TODO
115 * - we could suppress RX interrupt if we were so inclined.
116 */
117
118 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
119 {
120 VirtIONet *n = VIRTIO_NET(vdev);
121 struct virtio_net_config netcfg;
122
123 virtio_stw_p(vdev, &netcfg.status, n->status);
124 virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues);
125 virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
126 memcpy(netcfg.mac, n->mac, ETH_ALEN);
127 virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed);
128 netcfg.duplex = n->net_conf.duplex;
129 memcpy(config, &netcfg, n->config_size);
130 }
131
132 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
133 {
134 VirtIONet *n = VIRTIO_NET(vdev);
135 struct virtio_net_config netcfg = {};
136
137 memcpy(&netcfg, config, n->config_size);
138
139 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
140 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
141 memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
142 memcpy(n->mac, netcfg.mac, ETH_ALEN);
143 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
144 }
145 }
146
147 static bool virtio_net_started(VirtIONet *n, uint8_t status)
148 {
149 VirtIODevice *vdev = VIRTIO_DEVICE(n);
150 return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
151 (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
152 }
153
154 static void virtio_net_announce_notify(VirtIONet *net)
155 {
156 VirtIODevice *vdev = VIRTIO_DEVICE(net);
157 trace_virtio_net_announce_notify();
158
159 net->status |= VIRTIO_NET_S_ANNOUNCE;
160 virtio_notify_config(vdev);
161 }
162
163 static void virtio_net_announce_timer(void *opaque)
164 {
165 VirtIONet *n = opaque;
166 trace_virtio_net_announce_timer(n->announce_timer.round);
167
168 n->announce_timer.round--;
169 virtio_net_announce_notify(n);
170 }
171
172 static void virtio_net_announce(NetClientState *nc)
173 {
174 VirtIONet *n = qemu_get_nic_opaque(nc);
175 VirtIODevice *vdev = VIRTIO_DEVICE(n);
176
177 /*
178 * Make sure the virtio migration announcement timer isn't running
179 * If it is, let it trigger announcement so that we do not cause
180 * confusion.
181 */
182 if (n->announce_timer.round) {
183 return;
184 }
185
186 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
187 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
188 virtio_net_announce_notify(n);
189 }
190 }
191
192 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
193 {
194 VirtIODevice *vdev = VIRTIO_DEVICE(n);
195 NetClientState *nc = qemu_get_queue(n->nic);
196 int queues = n->multiqueue ? n->max_queues : 1;
197
198 if (!get_vhost_net(nc->peer)) {
199 return;
200 }
201
202 if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
203 !!n->vhost_started) {
204 return;
205 }
206 if (!n->vhost_started) {
207 int r, i;
208
209 if (n->needs_vnet_hdr_swap) {
210 error_report("backend does not support %s vnet headers; "
211 "falling back on userspace virtio",
212 virtio_is_big_endian(vdev) ? "BE" : "LE");
213 return;
214 }
215
216 /* Any packets outstanding? Purge them to avoid touching rings
217 * when vhost is running.
218 */
219 for (i = 0; i < queues; i++) {
220 NetClientState *qnc = qemu_get_subqueue(n->nic, i);
221
222 /* Purge both directions: TX and RX. */
223 qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
224 qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
225 }
226
227 if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
228 r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
229 if (r < 0) {
230 error_report("%uBytes MTU not supported by the backend",
231 n->net_conf.mtu);
232
233 return;
234 }
235 }
236
237 n->vhost_started = 1;
238 r = vhost_net_start(vdev, n->nic->ncs, queues);
239 if (r < 0) {
240 error_report("unable to start vhost net: %d: "
241 "falling back on userspace virtio", -r);
242 n->vhost_started = 0;
243 }
244 } else {
245 vhost_net_stop(vdev, n->nic->ncs, queues);
246 n->vhost_started = 0;
247 }
248 }
249
250 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
251 NetClientState *peer,
252 bool enable)
253 {
254 if (virtio_is_big_endian(vdev)) {
255 return qemu_set_vnet_be(peer, enable);
256 } else {
257 return qemu_set_vnet_le(peer, enable);
258 }
259 }
260
261 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
262 int queues, bool enable)
263 {
264 int i;
265
266 for (i = 0; i < queues; i++) {
267 if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
268 enable) {
269 while (--i >= 0) {
270 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
271 }
272
273 return true;
274 }
275 }
276
277 return false;
278 }
279
280 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
281 {
282 VirtIODevice *vdev = VIRTIO_DEVICE(n);
283 int queues = n->multiqueue ? n->max_queues : 1;
284
285 if (virtio_net_started(n, status)) {
286 /* Before using the device, we tell the network backend about the
287 * endianness to use when parsing vnet headers. If the backend
288 * can't do it, we fallback onto fixing the headers in the core
289 * virtio-net code.
290 */
291 n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
292 queues, true);
293 } else if (virtio_net_started(n, vdev->status)) {
294 /* After using the device, we need to reset the network backend to
295 * the default (guest native endianness), otherwise the guest may
296 * lose network connectivity if it is rebooted into a different
297 * endianness.
298 */
299 virtio_net_set_vnet_endian(vdev, n->nic->ncs, queues, false);
300 }
301 }
302
303 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq)
304 {
305 unsigned int dropped = virtqueue_drop_all(vq);
306 if (dropped) {
307 virtio_notify(vdev, vq);
308 }
309 }
310
311 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
312 {
313 VirtIONet *n = VIRTIO_NET(vdev);
314 VirtIONetQueue *q;
315 int i;
316 uint8_t queue_status;
317
318 virtio_net_vnet_endian_status(n, status);
319 virtio_net_vhost_status(n, status);
320
321 for (i = 0; i < n->max_queues; i++) {
322 NetClientState *ncs = qemu_get_subqueue(n->nic, i);
323 bool queue_started;
324 q = &n->vqs[i];
325
326 if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
327 queue_status = 0;
328 } else {
329 queue_status = status;
330 }
331 queue_started =
332 virtio_net_started(n, queue_status) && !n->vhost_started;
333
334 if (queue_started) {
335 qemu_flush_queued_packets(ncs);
336 }
337
338 if (!q->tx_waiting) {
339 continue;
340 }
341
342 if (queue_started) {
343 if (q->tx_timer) {
344 timer_mod(q->tx_timer,
345 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
346 } else {
347 qemu_bh_schedule(q->tx_bh);
348 }
349 } else {
350 if (q->tx_timer) {
351 timer_del(q->tx_timer);
352 } else {
353 qemu_bh_cancel(q->tx_bh);
354 }
355 if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 &&
356 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) &&
357 vdev->vm_running) {
358 /* if tx is waiting we are likely have some packets in tx queue
359 * and disabled notification */
360 q->tx_waiting = 0;
361 virtio_queue_set_notification(q->tx_vq, 1);
362 virtio_net_drop_tx_queue_data(vdev, q->tx_vq);
363 }
364 }
365 }
366 }
367
368 static void virtio_net_set_link_status(NetClientState *nc)
369 {
370 VirtIONet *n = qemu_get_nic_opaque(nc);
371 VirtIODevice *vdev = VIRTIO_DEVICE(n);
372 uint16_t old_status = n->status;
373
374 if (nc->link_down)
375 n->status &= ~VIRTIO_NET_S_LINK_UP;
376 else
377 n->status |= VIRTIO_NET_S_LINK_UP;
378
379 if (n->status != old_status)
380 virtio_notify_config(vdev);
381
382 virtio_net_set_status(vdev, vdev->status);
383 }
384
385 static void rxfilter_notify(NetClientState *nc)
386 {
387 VirtIONet *n = qemu_get_nic_opaque(nc);
388
389 if (nc->rxfilter_notify_enabled) {
390 gchar *path = object_get_canonical_path(OBJECT(n->qdev));
391 qapi_event_send_nic_rx_filter_changed(!!n->netclient_name,
392 n->netclient_name, path);
393 g_free(path);
394
395 /* disable event notification to avoid events flooding */
396 nc->rxfilter_notify_enabled = 0;
397 }
398 }
399
400 static intList *get_vlan_table(VirtIONet *n)
401 {
402 intList *list, *entry;
403 int i, j;
404
405 list = NULL;
406 for (i = 0; i < MAX_VLAN >> 5; i++) {
407 for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
408 if (n->vlans[i] & (1U << j)) {
409 entry = g_malloc0(sizeof(*entry));
410 entry->value = (i << 5) + j;
411 entry->next = list;
412 list = entry;
413 }
414 }
415 }
416
417 return list;
418 }
419
420 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
421 {
422 VirtIONet *n = qemu_get_nic_opaque(nc);
423 VirtIODevice *vdev = VIRTIO_DEVICE(n);
424 RxFilterInfo *info;
425 strList *str_list, *entry;
426 int i;
427
428 info = g_malloc0(sizeof(*info));
429 info->name = g_strdup(nc->name);
430 info->promiscuous = n->promisc;
431
432 if (n->nouni) {
433 info->unicast = RX_STATE_NONE;
434 } else if (n->alluni) {
435 info->unicast = RX_STATE_ALL;
436 } else {
437 info->unicast = RX_STATE_NORMAL;
438 }
439
440 if (n->nomulti) {
441 info->multicast = RX_STATE_NONE;
442 } else if (n->allmulti) {
443 info->multicast = RX_STATE_ALL;
444 } else {
445 info->multicast = RX_STATE_NORMAL;
446 }
447
448 info->broadcast_allowed = n->nobcast;
449 info->multicast_overflow = n->mac_table.multi_overflow;
450 info->unicast_overflow = n->mac_table.uni_overflow;
451
452 info->main_mac = qemu_mac_strdup_printf(n->mac);
453
454 str_list = NULL;
455 for (i = 0; i < n->mac_table.first_multi; i++) {
456 entry = g_malloc0(sizeof(*entry));
457 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
458 entry->next = str_list;
459 str_list = entry;
460 }
461 info->unicast_table = str_list;
462
463 str_list = NULL;
464 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
465 entry = g_malloc0(sizeof(*entry));
466 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN);
467 entry->next = str_list;
468 str_list = entry;
469 }
470 info->multicast_table = str_list;
471 info->vlan_table = get_vlan_table(n);
472
473 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
474 info->vlan = RX_STATE_ALL;
475 } else if (!info->vlan_table) {
476 info->vlan = RX_STATE_NONE;
477 } else {
478 info->vlan = RX_STATE_NORMAL;
479 }
480
481 /* enable event notification after query */
482 nc->rxfilter_notify_enabled = 1;
483
484 return info;
485 }
486
487 static void virtio_net_reset(VirtIODevice *vdev)
488 {
489 VirtIONet *n = VIRTIO_NET(vdev);
490 int i;
491
492 /* Reset back to compatibility mode */
493 n->promisc = 1;
494 n->allmulti = 0;
495 n->alluni = 0;
496 n->nomulti = 0;
497 n->nouni = 0;
498 n->nobcast = 0;
499 /* multiqueue is disabled by default */
500 n->curr_queues = 1;
501 timer_del(n->announce_timer.tm);
502 n->announce_timer.round = 0;
503 n->status &= ~VIRTIO_NET_S_ANNOUNCE;
504
505 /* Flush any MAC and VLAN filter table state */
506 n->mac_table.in_use = 0;
507 n->mac_table.first_multi = 0;
508 n->mac_table.multi_overflow = 0;
509 n->mac_table.uni_overflow = 0;
510 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
511 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
512 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
513 memset(n->vlans, 0, MAX_VLAN >> 3);
514
515 /* Flush any async TX */
516 for (i = 0; i < n->max_queues; i++) {
517 NetClientState *nc = qemu_get_subqueue(n->nic, i);
518
519 if (nc->peer) {
520 qemu_flush_or_purge_queued_packets(nc->peer, true);
521 assert(!virtio_net_get_subqueue(nc)->async_tx.elem);
522 }
523 }
524 }
525
526 static void peer_test_vnet_hdr(VirtIONet *n)
527 {
528 NetClientState *nc = qemu_get_queue(n->nic);
529 if (!nc->peer) {
530 return;
531 }
532
533 n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
534 }
535
536 static int peer_has_vnet_hdr(VirtIONet *n)
537 {
538 return n->has_vnet_hdr;
539 }
540
541 static int peer_has_ufo(VirtIONet *n)
542 {
543 if (!peer_has_vnet_hdr(n))
544 return 0;
545
546 n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
547
548 return n->has_ufo;
549 }
550
551 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
552 int version_1)
553 {
554 int i;
555 NetClientState *nc;
556
557 n->mergeable_rx_bufs = mergeable_rx_bufs;
558
559 if (version_1) {
560 n->guest_hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
561 } else {
562 n->guest_hdr_len = n->mergeable_rx_bufs ?
563 sizeof(struct virtio_net_hdr_mrg_rxbuf) :
564 sizeof(struct virtio_net_hdr);
565 }
566
567 for (i = 0; i < n->max_queues; i++) {
568 nc = qemu_get_subqueue(n->nic, i);
569
570 if (peer_has_vnet_hdr(n) &&
571 qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
572 qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
573 n->host_hdr_len = n->guest_hdr_len;
574 }
575 }
576 }
577
578 static int virtio_net_max_tx_queue_size(VirtIONet *n)
579 {
580 NetClientState *peer = n->nic_conf.peers.ncs[0];
581
582 /*
583 * Backends other than vhost-user don't support max queue size.
584 */
585 if (!peer) {
586 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
587 }
588
589 if (peer->info->type != NET_CLIENT_DRIVER_VHOST_USER) {
590 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
591 }
592
593 return VIRTQUEUE_MAX_SIZE;
594 }
595
596 static int peer_attach(VirtIONet *n, int index)
597 {
598 NetClientState *nc = qemu_get_subqueue(n->nic, index);
599
600 if (!nc->peer) {
601 return 0;
602 }
603
604 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
605 vhost_set_vring_enable(nc->peer, 1);
606 }
607
608 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
609 return 0;
610 }
611
612 if (n->max_queues == 1) {
613 return 0;
614 }
615
616 return tap_enable(nc->peer);
617 }
618
619 static int peer_detach(VirtIONet *n, int index)
620 {
621 NetClientState *nc = qemu_get_subqueue(n->nic, index);
622
623 if (!nc->peer) {
624 return 0;
625 }
626
627 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
628 vhost_set_vring_enable(nc->peer, 0);
629 }
630
631 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
632 return 0;
633 }
634
635 return tap_disable(nc->peer);
636 }
637
638 static void virtio_net_set_queues(VirtIONet *n)
639 {
640 int i;
641 int r;
642
643 if (n->nic->peer_deleted) {
644 return;
645 }
646
647 for (i = 0; i < n->max_queues; i++) {
648 if (i < n->curr_queues) {
649 r = peer_attach(n, i);
650 assert(!r);
651 } else {
652 r = peer_detach(n, i);
653 assert(!r);
654 }
655 }
656 }
657
658 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
659
660 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
661 Error **errp)
662 {
663 VirtIONet *n = VIRTIO_NET(vdev);
664 NetClientState *nc = qemu_get_queue(n->nic);
665
666 /* Firstly sync all virtio-net possible supported features */
667 features |= n->host_features;
668
669 virtio_add_feature(&features, VIRTIO_NET_F_MAC);
670
671 if (!peer_has_vnet_hdr(n)) {
672 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
673 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
674 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
675 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
676
677 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
678 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
679 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
680 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
681 }
682
683 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
684 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
685 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
686 }
687
688 if (!get_vhost_net(nc->peer)) {
689 return features;
690 }
691
692 features = vhost_net_get_features(get_vhost_net(nc->peer), features);
693 vdev->backend_features = features;
694
695 if (n->mtu_bypass_backend &&
696 (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) {
697 features |= (1ULL << VIRTIO_NET_F_MTU);
698 }
699
700 return features;
701 }
702
703 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
704 {
705 uint64_t features = 0;
706
707 /* Linux kernel 2.6.25. It understood MAC (as everyone must),
708 * but also these: */
709 virtio_add_feature(&features, VIRTIO_NET_F_MAC);
710 virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
711 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
712 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
713 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
714
715 return features;
716 }
717
718 static void virtio_net_apply_guest_offloads(VirtIONet *n)
719 {
720 qemu_set_offload(qemu_get_queue(n->nic)->peer,
721 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
722 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
723 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
724 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
725 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
726 }
727
728 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
729 {
730 static const uint64_t guest_offloads_mask =
731 (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
732 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
733 (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
734 (1ULL << VIRTIO_NET_F_GUEST_ECN) |
735 (1ULL << VIRTIO_NET_F_GUEST_UFO);
736
737 return guest_offloads_mask & features;
738 }
739
740 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
741 {
742 VirtIODevice *vdev = VIRTIO_DEVICE(n);
743 return virtio_net_guest_offloads_by_features(vdev->guest_features);
744 }
745
746 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
747 {
748 VirtIONet *n = VIRTIO_NET(vdev);
749 int i;
750
751 if (n->mtu_bypass_backend &&
752 !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) {
753 features &= ~(1ULL << VIRTIO_NET_F_MTU);
754 }
755
756 virtio_net_set_multiqueue(n,
757 virtio_has_feature(features, VIRTIO_NET_F_MQ));
758
759 virtio_net_set_mrg_rx_bufs(n,
760 virtio_has_feature(features,
761 VIRTIO_NET_F_MRG_RXBUF),
762 virtio_has_feature(features,
763 VIRTIO_F_VERSION_1));
764
765 n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
766 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4);
767 n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
768 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6);
769
770 if (n->has_vnet_hdr) {
771 n->curr_guest_offloads =
772 virtio_net_guest_offloads_by_features(features);
773 virtio_net_apply_guest_offloads(n);
774 }
775
776 for (i = 0; i < n->max_queues; i++) {
777 NetClientState *nc = qemu_get_subqueue(n->nic, i);
778
779 if (!get_vhost_net(nc->peer)) {
780 continue;
781 }
782 vhost_net_ack_features(get_vhost_net(nc->peer), features);
783 }
784
785 if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
786 memset(n->vlans, 0, MAX_VLAN >> 3);
787 } else {
788 memset(n->vlans, 0xff, MAX_VLAN >> 3);
789 }
790 }
791
792 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
793 struct iovec *iov, unsigned int iov_cnt)
794 {
795 uint8_t on;
796 size_t s;
797 NetClientState *nc = qemu_get_queue(n->nic);
798
799 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
800 if (s != sizeof(on)) {
801 return VIRTIO_NET_ERR;
802 }
803
804 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
805 n->promisc = on;
806 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
807 n->allmulti = on;
808 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
809 n->alluni = on;
810 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
811 n->nomulti = on;
812 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
813 n->nouni = on;
814 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
815 n->nobcast = on;
816 } else {
817 return VIRTIO_NET_ERR;
818 }
819
820 rxfilter_notify(nc);
821
822 return VIRTIO_NET_OK;
823 }
824
825 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
826 struct iovec *iov, unsigned int iov_cnt)
827 {
828 VirtIODevice *vdev = VIRTIO_DEVICE(n);
829 uint64_t offloads;
830 size_t s;
831
832 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
833 return VIRTIO_NET_ERR;
834 }
835
836 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
837 if (s != sizeof(offloads)) {
838 return VIRTIO_NET_ERR;
839 }
840
841 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
842 uint64_t supported_offloads;
843
844 offloads = virtio_ldq_p(vdev, &offloads);
845
846 if (!n->has_vnet_hdr) {
847 return VIRTIO_NET_ERR;
848 }
849
850 n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
851 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4);
852 n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
853 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6);
854 virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT);
855
856 supported_offloads = virtio_net_supported_guest_offloads(n);
857 if (offloads & ~supported_offloads) {
858 return VIRTIO_NET_ERR;
859 }
860
861 n->curr_guest_offloads = offloads;
862 virtio_net_apply_guest_offloads(n);
863
864 return VIRTIO_NET_OK;
865 } else {
866 return VIRTIO_NET_ERR;
867 }
868 }
869
870 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
871 struct iovec *iov, unsigned int iov_cnt)
872 {
873 VirtIODevice *vdev = VIRTIO_DEVICE(n);
874 struct virtio_net_ctrl_mac mac_data;
875 size_t s;
876 NetClientState *nc = qemu_get_queue(n->nic);
877
878 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
879 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
880 return VIRTIO_NET_ERR;
881 }
882 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
883 assert(s == sizeof(n->mac));
884 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
885 rxfilter_notify(nc);
886
887 return VIRTIO_NET_OK;
888 }
889
890 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
891 return VIRTIO_NET_ERR;
892 }
893
894 int in_use = 0;
895 int first_multi = 0;
896 uint8_t uni_overflow = 0;
897 uint8_t multi_overflow = 0;
898 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
899
900 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
901 sizeof(mac_data.entries));
902 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
903 if (s != sizeof(mac_data.entries)) {
904 goto error;
905 }
906 iov_discard_front(&iov, &iov_cnt, s);
907
908 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
909 goto error;
910 }
911
912 if (mac_data.entries <= MAC_TABLE_ENTRIES) {
913 s = iov_to_buf(iov, iov_cnt, 0, macs,
914 mac_data.entries * ETH_ALEN);
915 if (s != mac_data.entries * ETH_ALEN) {
916 goto error;
917 }
918 in_use += mac_data.entries;
919 } else {
920 uni_overflow = 1;
921 }
922
923 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
924
925 first_multi = in_use;
926
927 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
928 sizeof(mac_data.entries));
929 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
930 if (s != sizeof(mac_data.entries)) {
931 goto error;
932 }
933
934 iov_discard_front(&iov, &iov_cnt, s);
935
936 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
937 goto error;
938 }
939
940 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
941 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
942 mac_data.entries * ETH_ALEN);
943 if (s != mac_data.entries * ETH_ALEN) {
944 goto error;
945 }
946 in_use += mac_data.entries;
947 } else {
948 multi_overflow = 1;
949 }
950
951 n->mac_table.in_use = in_use;
952 n->mac_table.first_multi = first_multi;
953 n->mac_table.uni_overflow = uni_overflow;
954 n->mac_table.multi_overflow = multi_overflow;
955 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
956 g_free(macs);
957 rxfilter_notify(nc);
958
959 return VIRTIO_NET_OK;
960
961 error:
962 g_free(macs);
963 return VIRTIO_NET_ERR;
964 }
965
966 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
967 struct iovec *iov, unsigned int iov_cnt)
968 {
969 VirtIODevice *vdev = VIRTIO_DEVICE(n);
970 uint16_t vid;
971 size_t s;
972 NetClientState *nc = qemu_get_queue(n->nic);
973
974 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
975 vid = virtio_lduw_p(vdev, &vid);
976 if (s != sizeof(vid)) {
977 return VIRTIO_NET_ERR;
978 }
979
980 if (vid >= MAX_VLAN)
981 return VIRTIO_NET_ERR;
982
983 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
984 n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
985 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
986 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
987 else
988 return VIRTIO_NET_ERR;
989
990 rxfilter_notify(nc);
991
992 return VIRTIO_NET_OK;
993 }
994
995 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
996 struct iovec *iov, unsigned int iov_cnt)
997 {
998 trace_virtio_net_handle_announce(n->announce_timer.round);
999 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
1000 n->status & VIRTIO_NET_S_ANNOUNCE) {
1001 n->status &= ~VIRTIO_NET_S_ANNOUNCE;
1002 if (n->announce_timer.round) {
1003 qemu_announce_timer_step(&n->announce_timer);
1004 }
1005 return VIRTIO_NET_OK;
1006 } else {
1007 return VIRTIO_NET_ERR;
1008 }
1009 }
1010
1011 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
1012 struct iovec *iov, unsigned int iov_cnt)
1013 {
1014 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1015 struct virtio_net_ctrl_mq mq;
1016 size_t s;
1017 uint16_t queues;
1018
1019 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
1020 if (s != sizeof(mq)) {
1021 return VIRTIO_NET_ERR;
1022 }
1023
1024 if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
1025 return VIRTIO_NET_ERR;
1026 }
1027
1028 queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
1029
1030 if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1031 queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1032 queues > n->max_queues ||
1033 !n->multiqueue) {
1034 return VIRTIO_NET_ERR;
1035 }
1036
1037 n->curr_queues = queues;
1038 /* stop the backend before changing the number of queues to avoid handling a
1039 * disabled queue */
1040 virtio_net_set_status(vdev, vdev->status);
1041 virtio_net_set_queues(n);
1042
1043 return VIRTIO_NET_OK;
1044 }
1045
1046 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
1047 {
1048 VirtIONet *n = VIRTIO_NET(vdev);
1049 struct virtio_net_ctrl_hdr ctrl;
1050 virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
1051 VirtQueueElement *elem;
1052 size_t s;
1053 struct iovec *iov, *iov2;
1054 unsigned int iov_cnt;
1055
1056 for (;;) {
1057 elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
1058 if (!elem) {
1059 break;
1060 }
1061 if (iov_size(elem->in_sg, elem->in_num) < sizeof(status) ||
1062 iov_size(elem->out_sg, elem->out_num) < sizeof(ctrl)) {
1063 virtio_error(vdev, "virtio-net ctrl missing headers");
1064 virtqueue_detach_element(vq, elem, 0);
1065 g_free(elem);
1066 break;
1067 }
1068
1069 iov_cnt = elem->out_num;
1070 iov2 = iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num);
1071 s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
1072 iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
1073 if (s != sizeof(ctrl)) {
1074 status = VIRTIO_NET_ERR;
1075 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
1076 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
1077 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
1078 status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
1079 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
1080 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
1081 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
1082 status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt);
1083 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
1084 status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
1085 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
1086 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
1087 }
1088
1089 s = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, sizeof(status));
1090 assert(s == sizeof(status));
1091
1092 virtqueue_push(vq, elem, sizeof(status));
1093 virtio_notify(vdev, vq);
1094 g_free(iov2);
1095 g_free(elem);
1096 }
1097 }
1098
1099 /* RX */
1100
1101 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
1102 {
1103 VirtIONet *n = VIRTIO_NET(vdev);
1104 int queue_index = vq2q(virtio_get_queue_index(vq));
1105
1106 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
1107 }
1108
1109 static int virtio_net_can_receive(NetClientState *nc)
1110 {
1111 VirtIONet *n = qemu_get_nic_opaque(nc);
1112 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1113 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1114
1115 if (!vdev->vm_running) {
1116 return 0;
1117 }
1118
1119 if (nc->queue_index >= n->curr_queues) {
1120 return 0;
1121 }
1122
1123 if (!virtio_queue_ready(q->rx_vq) ||
1124 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1125 return 0;
1126 }
1127
1128 return 1;
1129 }
1130
1131 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1132 {
1133 VirtIONet *n = q->n;
1134 if (virtio_queue_empty(q->rx_vq) ||
1135 (n->mergeable_rx_bufs &&
1136 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1137 virtio_queue_set_notification(q->rx_vq, 1);
1138
1139 /* To avoid a race condition where the guest has made some buffers
1140 * available after the above check but before notification was
1141 * enabled, check for available buffers again.
1142 */
1143 if (virtio_queue_empty(q->rx_vq) ||
1144 (n->mergeable_rx_bufs &&
1145 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1146 return 0;
1147 }
1148 }
1149
1150 virtio_queue_set_notification(q->rx_vq, 0);
1151 return 1;
1152 }
1153
1154 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1155 {
1156 virtio_tswap16s(vdev, &hdr->hdr_len);
1157 virtio_tswap16s(vdev, &hdr->gso_size);
1158 virtio_tswap16s(vdev, &hdr->csum_start);
1159 virtio_tswap16s(vdev, &hdr->csum_offset);
1160 }
1161
1162 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1163 * it never finds out that the packets don't have valid checksums. This
1164 * causes dhclient to get upset. Fedora's carried a patch for ages to
1165 * fix this with Xen but it hasn't appeared in an upstream release of
1166 * dhclient yet.
1167 *
1168 * To avoid breaking existing guests, we catch udp packets and add
1169 * checksums. This is terrible but it's better than hacking the guest
1170 * kernels.
1171 *
1172 * N.B. if we introduce a zero-copy API, this operation is no longer free so
1173 * we should provide a mechanism to disable it to avoid polluting the host
1174 * cache.
1175 */
1176 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1177 uint8_t *buf, size_t size)
1178 {
1179 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1180 (size > 27 && size < 1500) && /* normal sized MTU */
1181 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1182 (buf[23] == 17) && /* ip.protocol == UDP */
1183 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1184 net_checksum_calculate(buf, size);
1185 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1186 }
1187 }
1188
1189 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1190 const void *buf, size_t size)
1191 {
1192 if (n->has_vnet_hdr) {
1193 /* FIXME this cast is evil */
1194 void *wbuf = (void *)buf;
1195 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1196 size - n->host_hdr_len);
1197
1198 if (n->needs_vnet_hdr_swap) {
1199 virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1200 }
1201 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1202 } else {
1203 struct virtio_net_hdr hdr = {
1204 .flags = 0,
1205 .gso_type = VIRTIO_NET_HDR_GSO_NONE
1206 };
1207 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1208 }
1209 }
1210
1211 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1212 {
1213 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1214 static const uint8_t vlan[] = {0x81, 0x00};
1215 uint8_t *ptr = (uint8_t *)buf;
1216 int i;
1217
1218 if (n->promisc)
1219 return 1;
1220
1221 ptr += n->host_hdr_len;
1222
1223 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1224 int vid = lduw_be_p(ptr + 14) & 0xfff;
1225 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1226 return 0;
1227 }
1228
1229 if (ptr[0] & 1) { // multicast
1230 if (!memcmp(ptr, bcast, sizeof(bcast))) {
1231 return !n->nobcast;
1232 } else if (n->nomulti) {
1233 return 0;
1234 } else if (n->allmulti || n->mac_table.multi_overflow) {
1235 return 1;
1236 }
1237
1238 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1239 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1240 return 1;
1241 }
1242 }
1243 } else { // unicast
1244 if (n->nouni) {
1245 return 0;
1246 } else if (n->alluni || n->mac_table.uni_overflow) {
1247 return 1;
1248 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1249 return 1;
1250 }
1251
1252 for (i = 0; i < n->mac_table.first_multi; i++) {
1253 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1254 return 1;
1255 }
1256 }
1257 }
1258
1259 return 0;
1260 }
1261
1262 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
1263 size_t size)
1264 {
1265 VirtIONet *n = qemu_get_nic_opaque(nc);
1266 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1267 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1268 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1269 struct virtio_net_hdr_mrg_rxbuf mhdr;
1270 unsigned mhdr_cnt = 0;
1271 size_t offset, i, guest_offset;
1272
1273 if (!virtio_net_can_receive(nc)) {
1274 return -1;
1275 }
1276
1277 /* hdr_len refers to the header we supply to the guest */
1278 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1279 return 0;
1280 }
1281
1282 if (!receive_filter(n, buf, size))
1283 return size;
1284
1285 offset = i = 0;
1286
1287 while (offset < size) {
1288 VirtQueueElement *elem;
1289 int len, total;
1290 const struct iovec *sg;
1291
1292 total = 0;
1293
1294 elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1295 if (!elem) {
1296 if (i) {
1297 virtio_error(vdev, "virtio-net unexpected empty queue: "
1298 "i %zd mergeable %d offset %zd, size %zd, "
1299 "guest hdr len %zd, host hdr len %zd "
1300 "guest features 0x%" PRIx64,
1301 i, n->mergeable_rx_bufs, offset, size,
1302 n->guest_hdr_len, n->host_hdr_len,
1303 vdev->guest_features);
1304 }
1305 return -1;
1306 }
1307
1308 if (elem->in_num < 1) {
1309 virtio_error(vdev,
1310 "virtio-net receive queue contains no in buffers");
1311 virtqueue_detach_element(q->rx_vq, elem, 0);
1312 g_free(elem);
1313 return -1;
1314 }
1315
1316 sg = elem->in_sg;
1317 if (i == 0) {
1318 assert(offset == 0);
1319 if (n->mergeable_rx_bufs) {
1320 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1321 sg, elem->in_num,
1322 offsetof(typeof(mhdr), num_buffers),
1323 sizeof(mhdr.num_buffers));
1324 }
1325
1326 receive_header(n, sg, elem->in_num, buf, size);
1327 offset = n->host_hdr_len;
1328 total += n->guest_hdr_len;
1329 guest_offset = n->guest_hdr_len;
1330 } else {
1331 guest_offset = 0;
1332 }
1333
1334 /* copy in packet. ugh */
1335 len = iov_from_buf(sg, elem->in_num, guest_offset,
1336 buf + offset, size - offset);
1337 total += len;
1338 offset += len;
1339 /* If buffers can't be merged, at this point we
1340 * must have consumed the complete packet.
1341 * Otherwise, drop it. */
1342 if (!n->mergeable_rx_bufs && offset < size) {
1343 virtqueue_unpop(q->rx_vq, elem, total);
1344 g_free(elem);
1345 return size;
1346 }
1347
1348 /* signal other side */
1349 virtqueue_fill(q->rx_vq, elem, total, i++);
1350 g_free(elem);
1351 }
1352
1353 if (mhdr_cnt) {
1354 virtio_stw_p(vdev, &mhdr.num_buffers, i);
1355 iov_from_buf(mhdr_sg, mhdr_cnt,
1356 0,
1357 &mhdr.num_buffers, sizeof mhdr.num_buffers);
1358 }
1359
1360 virtqueue_flush(q->rx_vq, i);
1361 virtio_notify(vdev, q->rx_vq);
1362
1363 return size;
1364 }
1365
1366 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf,
1367 size_t size)
1368 {
1369 ssize_t r;
1370
1371 rcu_read_lock();
1372 r = virtio_net_receive_rcu(nc, buf, size);
1373 rcu_read_unlock();
1374 return r;
1375 }
1376
1377 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain,
1378 const uint8_t *buf,
1379 VirtioNetRscUnit *unit)
1380 {
1381 uint16_t ip_hdrlen;
1382 struct ip_header *ip;
1383
1384 ip = (struct ip_header *)(buf + chain->n->guest_hdr_len
1385 + sizeof(struct eth_header));
1386 unit->ip = (void *)ip;
1387 ip_hdrlen = (ip->ip_ver_len & 0xF) << 2;
1388 unit->ip_plen = &ip->ip_len;
1389 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen);
1390 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1391 unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen;
1392 }
1393
1394 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain,
1395 const uint8_t *buf,
1396 VirtioNetRscUnit *unit)
1397 {
1398 struct ip6_header *ip6;
1399
1400 ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len
1401 + sizeof(struct eth_header));
1402 unit->ip = ip6;
1403 unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
1404 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip)\
1405 + sizeof(struct ip6_header));
1406 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1407
1408 /* There is a difference between payload lenght in ipv4 and v6,
1409 ip header is excluded in ipv6 */
1410 unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen;
1411 }
1412
1413 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain,
1414 VirtioNetRscSeg *seg)
1415 {
1416 int ret;
1417 struct virtio_net_hdr *h;
1418
1419 h = (struct virtio_net_hdr *)seg->buf;
1420 h->flags = 0;
1421 h->gso_type = VIRTIO_NET_HDR_GSO_NONE;
1422
1423 if (seg->is_coalesced) {
1424 *virtio_net_rsc_ext_num_packets(h) = seg->packets;
1425 *virtio_net_rsc_ext_num_dupacks(h) = seg->dup_ack;
1426 h->flags = VIRTIO_NET_HDR_F_RSC_INFO;
1427 if (chain->proto == ETH_P_IP) {
1428 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1429 } else {
1430 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1431 }
1432 }
1433
1434 ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size);
1435 QTAILQ_REMOVE(&chain->buffers, seg, next);
1436 g_free(seg->buf);
1437 g_free(seg);
1438
1439 return ret;
1440 }
1441
1442 static void virtio_net_rsc_purge(void *opq)
1443 {
1444 VirtioNetRscSeg *seg, *rn;
1445 VirtioNetRscChain *chain = (VirtioNetRscChain *)opq;
1446
1447 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) {
1448 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
1449 chain->stat.purge_failed++;
1450 continue;
1451 }
1452 }
1453
1454 chain->stat.timer++;
1455 if (!QTAILQ_EMPTY(&chain->buffers)) {
1456 timer_mod(chain->drain_timer,
1457 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
1458 }
1459 }
1460
1461 static void virtio_net_rsc_cleanup(VirtIONet *n)
1462 {
1463 VirtioNetRscChain *chain, *rn_chain;
1464 VirtioNetRscSeg *seg, *rn_seg;
1465
1466 QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) {
1467 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) {
1468 QTAILQ_REMOVE(&chain->buffers, seg, next);
1469 g_free(seg->buf);
1470 g_free(seg);
1471 }
1472
1473 timer_del(chain->drain_timer);
1474 timer_free(chain->drain_timer);
1475 QTAILQ_REMOVE(&n->rsc_chains, chain, next);
1476 g_free(chain);
1477 }
1478 }
1479
1480 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain,
1481 NetClientState *nc,
1482 const uint8_t *buf, size_t size)
1483 {
1484 uint16_t hdr_len;
1485 VirtioNetRscSeg *seg;
1486
1487 hdr_len = chain->n->guest_hdr_len;
1488 seg = g_malloc(sizeof(VirtioNetRscSeg));
1489 seg->buf = g_malloc(hdr_len + sizeof(struct eth_header)
1490 + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD);
1491 memcpy(seg->buf, buf, size);
1492 seg->size = size;
1493 seg->packets = 1;
1494 seg->dup_ack = 0;
1495 seg->is_coalesced = 0;
1496 seg->nc = nc;
1497
1498 QTAILQ_INSERT_TAIL(&chain->buffers, seg, next);
1499 chain->stat.cache++;
1500
1501 switch (chain->proto) {
1502 case ETH_P_IP:
1503 virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
1504 break;
1505 case ETH_P_IPV6:
1506 virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit);
1507 break;
1508 default:
1509 g_assert_not_reached();
1510 }
1511 }
1512
1513 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain,
1514 VirtioNetRscSeg *seg,
1515 const uint8_t *buf,
1516 struct tcp_header *n_tcp,
1517 struct tcp_header *o_tcp)
1518 {
1519 uint32_t nack, oack;
1520 uint16_t nwin, owin;
1521
1522 nack = htonl(n_tcp->th_ack);
1523 nwin = htons(n_tcp->th_win);
1524 oack = htonl(o_tcp->th_ack);
1525 owin = htons(o_tcp->th_win);
1526
1527 if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) {
1528 chain->stat.ack_out_of_win++;
1529 return RSC_FINAL;
1530 } else if (nack == oack) {
1531 /* duplicated ack or window probe */
1532 if (nwin == owin) {
1533 /* duplicated ack, add dup ack count due to whql test up to 1 */
1534 chain->stat.dup_ack++;
1535 return RSC_FINAL;
1536 } else {
1537 /* Coalesce window update */
1538 o_tcp->th_win = n_tcp->th_win;
1539 chain->stat.win_update++;
1540 return RSC_COALESCE;
1541 }
1542 } else {
1543 /* pure ack, go to 'C', finalize*/
1544 chain->stat.pure_ack++;
1545 return RSC_FINAL;
1546 }
1547 }
1548
1549 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain,
1550 VirtioNetRscSeg *seg,
1551 const uint8_t *buf,
1552 VirtioNetRscUnit *n_unit)
1553 {
1554 void *data;
1555 uint16_t o_ip_len;
1556 uint32_t nseq, oseq;
1557 VirtioNetRscUnit *o_unit;
1558
1559 o_unit = &seg->unit;
1560 o_ip_len = htons(*o_unit->ip_plen);
1561 nseq = htonl(n_unit->tcp->th_seq);
1562 oseq = htonl(o_unit->tcp->th_seq);
1563
1564 /* out of order or retransmitted. */
1565 if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) {
1566 chain->stat.data_out_of_win++;
1567 return RSC_FINAL;
1568 }
1569
1570 data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen;
1571 if (nseq == oseq) {
1572 if ((o_unit->payload == 0) && n_unit->payload) {
1573 /* From no payload to payload, normal case, not a dup ack or etc */
1574 chain->stat.data_after_pure_ack++;
1575 goto coalesce;
1576 } else {
1577 return virtio_net_rsc_handle_ack(chain, seg, buf,
1578 n_unit->tcp, o_unit->tcp);
1579 }
1580 } else if ((nseq - oseq) != o_unit->payload) {
1581 /* Not a consistent packet, out of order */
1582 chain->stat.data_out_of_order++;
1583 return RSC_FINAL;
1584 } else {
1585 coalesce:
1586 if ((o_ip_len + n_unit->payload) > chain->max_payload) {
1587 chain->stat.over_size++;
1588 return RSC_FINAL;
1589 }
1590
1591 /* Here comes the right data, the payload length in v4/v6 is different,
1592 so use the field value to update and record the new data len */
1593 o_unit->payload += n_unit->payload; /* update new data len */
1594
1595 /* update field in ip header */
1596 *o_unit->ip_plen = htons(o_ip_len + n_unit->payload);
1597
1598 /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
1599 for windows guest, while this may change the behavior for linux
1600 guest (only if it uses RSC feature). */
1601 o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags;
1602
1603 o_unit->tcp->th_ack = n_unit->tcp->th_ack;
1604 o_unit->tcp->th_win = n_unit->tcp->th_win;
1605
1606 memmove(seg->buf + seg->size, data, n_unit->payload);
1607 seg->size += n_unit->payload;
1608 seg->packets++;
1609 chain->stat.coalesced++;
1610 return RSC_COALESCE;
1611 }
1612 }
1613
1614 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain,
1615 VirtioNetRscSeg *seg,
1616 const uint8_t *buf, size_t size,
1617 VirtioNetRscUnit *unit)
1618 {
1619 struct ip_header *ip1, *ip2;
1620
1621 ip1 = (struct ip_header *)(unit->ip);
1622 ip2 = (struct ip_header *)(seg->unit.ip);
1623 if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst)
1624 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
1625 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
1626 chain->stat.no_match++;
1627 return RSC_NO_MATCH;
1628 }
1629
1630 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
1631 }
1632
1633 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain,
1634 VirtioNetRscSeg *seg,
1635 const uint8_t *buf, size_t size,
1636 VirtioNetRscUnit *unit)
1637 {
1638 struct ip6_header *ip1, *ip2;
1639
1640 ip1 = (struct ip6_header *)(unit->ip);
1641 ip2 = (struct ip6_header *)(seg->unit.ip);
1642 if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address))
1643 || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address))
1644 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
1645 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
1646 chain->stat.no_match++;
1647 return RSC_NO_MATCH;
1648 }
1649
1650 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
1651 }
1652
1653 /* Packets with 'SYN' should bypass, other flag should be sent after drain
1654 * to prevent out of order */
1655 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain,
1656 struct tcp_header *tcp)
1657 {
1658 uint16_t tcp_hdr;
1659 uint16_t tcp_flag;
1660
1661 tcp_flag = htons(tcp->th_offset_flags);
1662 tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10;
1663 tcp_flag &= VIRTIO_NET_TCP_FLAG;
1664 tcp_flag = htons(tcp->th_offset_flags) & 0x3F;
1665 if (tcp_flag & TH_SYN) {
1666 chain->stat.tcp_syn++;
1667 return RSC_BYPASS;
1668 }
1669
1670 if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) {
1671 chain->stat.tcp_ctrl_drain++;
1672 return RSC_FINAL;
1673 }
1674
1675 if (tcp_hdr > sizeof(struct tcp_header)) {
1676 chain->stat.tcp_all_opt++;
1677 return RSC_FINAL;
1678 }
1679
1680 return RSC_CANDIDATE;
1681 }
1682
1683 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain,
1684 NetClientState *nc,
1685 const uint8_t *buf, size_t size,
1686 VirtioNetRscUnit *unit)
1687 {
1688 int ret;
1689 VirtioNetRscSeg *seg, *nseg;
1690
1691 if (QTAILQ_EMPTY(&chain->buffers)) {
1692 chain->stat.empty_cache++;
1693 virtio_net_rsc_cache_buf(chain, nc, buf, size);
1694 timer_mod(chain->drain_timer,
1695 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
1696 return size;
1697 }
1698
1699 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
1700 if (chain->proto == ETH_P_IP) {
1701 ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
1702 } else {
1703 ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit);
1704 }
1705
1706 if (ret == RSC_FINAL) {
1707 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
1708 /* Send failed */
1709 chain->stat.final_failed++;
1710 return 0;
1711 }
1712
1713 /* Send current packet */
1714 return virtio_net_do_receive(nc, buf, size);
1715 } else if (ret == RSC_NO_MATCH) {
1716 continue;
1717 } else {
1718 /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
1719 seg->is_coalesced = 1;
1720 return size;
1721 }
1722 }
1723
1724 chain->stat.no_match_cache++;
1725 virtio_net_rsc_cache_buf(chain, nc, buf, size);
1726 return size;
1727 }
1728
1729 /* Drain a connection data, this is to avoid out of order segments */
1730 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain,
1731 NetClientState *nc,
1732 const uint8_t *buf, size_t size,
1733 uint16_t ip_start, uint16_t ip_size,
1734 uint16_t tcp_port)
1735 {
1736 VirtioNetRscSeg *seg, *nseg;
1737 uint32_t ppair1, ppair2;
1738
1739 ppair1 = *(uint32_t *)(buf + tcp_port);
1740 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
1741 ppair2 = *(uint32_t *)(seg->buf + tcp_port);
1742 if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size)
1743 || (ppair1 != ppair2)) {
1744 continue;
1745 }
1746 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
1747 chain->stat.drain_failed++;
1748 }
1749
1750 break;
1751 }
1752
1753 return virtio_net_do_receive(nc, buf, size);
1754 }
1755
1756 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain,
1757 struct ip_header *ip,
1758 const uint8_t *buf, size_t size)
1759 {
1760 uint16_t ip_len;
1761
1762 /* Not an ipv4 packet */
1763 if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) {
1764 chain->stat.ip_option++;
1765 return RSC_BYPASS;
1766 }
1767
1768 /* Don't handle packets with ip option */
1769 if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) {
1770 chain->stat.ip_option++;
1771 return RSC_BYPASS;
1772 }
1773
1774 if (ip->ip_p != IPPROTO_TCP) {
1775 chain->stat.bypass_not_tcp++;
1776 return RSC_BYPASS;
1777 }
1778
1779 /* Don't handle packets with ip fragment */
1780 if (!(htons(ip->ip_off) & IP_DF)) {
1781 chain->stat.ip_frag++;
1782 return RSC_BYPASS;
1783 }
1784
1785 /* Don't handle packets with ecn flag */
1786 if (IPTOS_ECN(ip->ip_tos)) {
1787 chain->stat.ip_ecn++;
1788 return RSC_BYPASS;
1789 }
1790
1791 ip_len = htons(ip->ip_len);
1792 if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header))
1793 || ip_len > (size - chain->n->guest_hdr_len -
1794 sizeof(struct eth_header))) {
1795 chain->stat.ip_hacked++;
1796 return RSC_BYPASS;
1797 }
1798
1799 return RSC_CANDIDATE;
1800 }
1801
1802 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain,
1803 NetClientState *nc,
1804 const uint8_t *buf, size_t size)
1805 {
1806 int32_t ret;
1807 uint16_t hdr_len;
1808 VirtioNetRscUnit unit;
1809
1810 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
1811
1812 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)
1813 + sizeof(struct tcp_header))) {
1814 chain->stat.bypass_not_tcp++;
1815 return virtio_net_do_receive(nc, buf, size);
1816 }
1817
1818 virtio_net_rsc_extract_unit4(chain, buf, &unit);
1819 if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size)
1820 != RSC_CANDIDATE) {
1821 return virtio_net_do_receive(nc, buf, size);
1822 }
1823
1824 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
1825 if (ret == RSC_BYPASS) {
1826 return virtio_net_do_receive(nc, buf, size);
1827 } else if (ret == RSC_FINAL) {
1828 return virtio_net_rsc_drain_flow(chain, nc, buf, size,
1829 ((hdr_len + sizeof(struct eth_header)) + 12),
1830 VIRTIO_NET_IP4_ADDR_SIZE,
1831 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header));
1832 }
1833
1834 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
1835 }
1836
1837 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain,
1838 struct ip6_header *ip6,
1839 const uint8_t *buf, size_t size)
1840 {
1841 uint16_t ip_len;
1842
1843 if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4)
1844 != IP_HEADER_VERSION_6) {
1845 return RSC_BYPASS;
1846 }
1847
1848 /* Both option and protocol is checked in this */
1849 if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) {
1850 chain->stat.bypass_not_tcp++;
1851 return RSC_BYPASS;
1852 }
1853
1854 ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
1855 if (ip_len < sizeof(struct tcp_header) ||
1856 ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header)
1857 - sizeof(struct ip6_header))) {
1858 chain->stat.ip_hacked++;
1859 return RSC_BYPASS;
1860 }
1861
1862 /* Don't handle packets with ecn flag */
1863 if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) {
1864 chain->stat.ip_ecn++;
1865 return RSC_BYPASS;
1866 }
1867
1868 return RSC_CANDIDATE;
1869 }
1870
1871 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc,
1872 const uint8_t *buf, size_t size)
1873 {
1874 int32_t ret;
1875 uint16_t hdr_len;
1876 VirtioNetRscChain *chain;
1877 VirtioNetRscUnit unit;
1878
1879 chain = (VirtioNetRscChain *)opq;
1880 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
1881
1882 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header)
1883 + sizeof(tcp_header))) {
1884 return virtio_net_do_receive(nc, buf, size);
1885 }
1886
1887 virtio_net_rsc_extract_unit6(chain, buf, &unit);
1888 if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain,
1889 unit.ip, buf, size)) {
1890 return virtio_net_do_receive(nc, buf, size);
1891 }
1892
1893 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
1894 if (ret == RSC_BYPASS) {
1895 return virtio_net_do_receive(nc, buf, size);
1896 } else if (ret == RSC_FINAL) {
1897 return virtio_net_rsc_drain_flow(chain, nc, buf, size,
1898 ((hdr_len + sizeof(struct eth_header)) + 8),
1899 VIRTIO_NET_IP6_ADDR_SIZE,
1900 hdr_len + sizeof(struct eth_header)
1901 + sizeof(struct ip6_header));
1902 }
1903
1904 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
1905 }
1906
1907 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n,
1908 NetClientState *nc,
1909 uint16_t proto)
1910 {
1911 VirtioNetRscChain *chain;
1912
1913 if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) {
1914 return NULL;
1915 }
1916
1917 QTAILQ_FOREACH(chain, &n->rsc_chains, next) {
1918 if (chain->proto == proto) {
1919 return chain;
1920 }
1921 }
1922
1923 chain = g_malloc(sizeof(*chain));
1924 chain->n = n;
1925 chain->proto = proto;
1926 if (proto == (uint16_t)ETH_P_IP) {
1927 chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
1928 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1929 } else {
1930 chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD;
1931 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1932 }
1933 chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST,
1934 virtio_net_rsc_purge, chain);
1935 memset(&chain->stat, 0, sizeof(chain->stat));
1936
1937 QTAILQ_INIT(&chain->buffers);
1938 QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next);
1939
1940 return chain;
1941 }
1942
1943 static ssize_t virtio_net_rsc_receive(NetClientState *nc,
1944 const uint8_t *buf,
1945 size_t size)
1946 {
1947 uint16_t proto;
1948 VirtioNetRscChain *chain;
1949 struct eth_header *eth;
1950 VirtIONet *n;
1951
1952 n = qemu_get_nic_opaque(nc);
1953 if (size < (n->host_hdr_len + sizeof(struct eth_header))) {
1954 return virtio_net_do_receive(nc, buf, size);
1955 }
1956
1957 eth = (struct eth_header *)(buf + n->guest_hdr_len);
1958 proto = htons(eth->h_proto);
1959
1960 chain = virtio_net_rsc_lookup_chain(n, nc, proto);
1961 if (chain) {
1962 chain->stat.received++;
1963 if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) {
1964 return virtio_net_rsc_receive4(chain, nc, buf, size);
1965 } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) {
1966 return virtio_net_rsc_receive6(chain, nc, buf, size);
1967 }
1968 }
1969 return virtio_net_do_receive(nc, buf, size);
1970 }
1971
1972 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf,
1973 size_t size)
1974 {
1975 VirtIONet *n = qemu_get_nic_opaque(nc);
1976 if ((n->rsc4_enabled || n->rsc6_enabled)) {
1977 return virtio_net_rsc_receive(nc, buf, size);
1978 } else {
1979 return virtio_net_do_receive(nc, buf, size);
1980 }
1981 }
1982
1983 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
1984
1985 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
1986 {
1987 VirtIONet *n = qemu_get_nic_opaque(nc);
1988 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1989 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1990
1991 virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
1992 virtio_notify(vdev, q->tx_vq);
1993
1994 g_free(q->async_tx.elem);
1995 q->async_tx.elem = NULL;
1996
1997 virtio_queue_set_notification(q->tx_vq, 1);
1998 virtio_net_flush_tx(q);
1999 }
2000
2001 /* TX */
2002 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
2003 {
2004 VirtIONet *n = q->n;
2005 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2006 VirtQueueElement *elem;
2007 int32_t num_packets = 0;
2008 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
2009 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2010 return num_packets;
2011 }
2012
2013 if (q->async_tx.elem) {
2014 virtio_queue_set_notification(q->tx_vq, 0);
2015 return num_packets;
2016 }
2017
2018 for (;;) {
2019 ssize_t ret;
2020 unsigned int out_num;
2021 struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
2022 struct virtio_net_hdr_mrg_rxbuf mhdr;
2023
2024 elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
2025 if (!elem) {
2026 break;
2027 }
2028
2029 out_num = elem->out_num;
2030 out_sg = elem->out_sg;
2031 if (out_num < 1) {
2032 virtio_error(vdev, "virtio-net header not in first element");
2033 virtqueue_detach_element(q->tx_vq, elem, 0);
2034 g_free(elem);
2035 return -EINVAL;
2036 }
2037
2038 if (n->has_vnet_hdr) {
2039 if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
2040 n->guest_hdr_len) {
2041 virtio_error(vdev, "virtio-net header incorrect");
2042 virtqueue_detach_element(q->tx_vq, elem, 0);
2043 g_free(elem);
2044 return -EINVAL;
2045 }
2046 if (n->needs_vnet_hdr_swap) {
2047 virtio_net_hdr_swap(vdev, (void *) &mhdr);
2048 sg2[0].iov_base = &mhdr;
2049 sg2[0].iov_len = n->guest_hdr_len;
2050 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
2051 out_sg, out_num,
2052 n->guest_hdr_len, -1);
2053 if (out_num == VIRTQUEUE_MAX_SIZE) {
2054 goto drop;
2055 }
2056 out_num += 1;
2057 out_sg = sg2;
2058 }
2059 }
2060 /*
2061 * If host wants to see the guest header as is, we can
2062 * pass it on unchanged. Otherwise, copy just the parts
2063 * that host is interested in.
2064 */
2065 assert(n->host_hdr_len <= n->guest_hdr_len);
2066 if (n->host_hdr_len != n->guest_hdr_len) {
2067 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
2068 out_sg, out_num,
2069 0, n->host_hdr_len);
2070 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
2071 out_sg, out_num,
2072 n->guest_hdr_len, -1);
2073 out_num = sg_num;
2074 out_sg = sg;
2075 }
2076
2077 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
2078 out_sg, out_num, virtio_net_tx_complete);
2079 if (ret == 0) {
2080 virtio_queue_set_notification(q->tx_vq, 0);
2081 q->async_tx.elem = elem;
2082 return -EBUSY;
2083 }
2084
2085 drop:
2086 virtqueue_push(q->tx_vq, elem, 0);
2087 virtio_notify(vdev, q->tx_vq);
2088 g_free(elem);
2089
2090 if (++num_packets >= n->tx_burst) {
2091 break;
2092 }
2093 }
2094 return num_packets;
2095 }
2096
2097 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
2098 {
2099 VirtIONet *n = VIRTIO_NET(vdev);
2100 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2101
2102 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2103 virtio_net_drop_tx_queue_data(vdev, vq);
2104 return;
2105 }
2106
2107 /* This happens when device was stopped but VCPU wasn't. */
2108 if (!vdev->vm_running) {
2109 q->tx_waiting = 1;
2110 return;
2111 }
2112
2113 if (q->tx_waiting) {
2114 virtio_queue_set_notification(vq, 1);
2115 timer_del(q->tx_timer);
2116 q->tx_waiting = 0;
2117 if (virtio_net_flush_tx(q) == -EINVAL) {
2118 return;
2119 }
2120 } else {
2121 timer_mod(q->tx_timer,
2122 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2123 q->tx_waiting = 1;
2124 virtio_queue_set_notification(vq, 0);
2125 }
2126 }
2127
2128 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
2129 {
2130 VirtIONet *n = VIRTIO_NET(vdev);
2131 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2132
2133 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2134 virtio_net_drop_tx_queue_data(vdev, vq);
2135 return;
2136 }
2137
2138 if (unlikely(q->tx_waiting)) {
2139 return;
2140 }
2141 q->tx_waiting = 1;
2142 /* This happens when device was stopped but VCPU wasn't. */
2143 if (!vdev->vm_running) {
2144 return;
2145 }
2146 virtio_queue_set_notification(vq, 0);
2147 qemu_bh_schedule(q->tx_bh);
2148 }
2149
2150 static void virtio_net_tx_timer(void *opaque)
2151 {
2152 VirtIONetQueue *q = opaque;
2153 VirtIONet *n = q->n;
2154 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2155 /* This happens when device was stopped but BH wasn't. */
2156 if (!vdev->vm_running) {
2157 /* Make sure tx waiting is set, so we'll run when restarted. */
2158 assert(q->tx_waiting);
2159 return;
2160 }
2161
2162 q->tx_waiting = 0;
2163
2164 /* Just in case the driver is not ready on more */
2165 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2166 return;
2167 }
2168
2169 virtio_queue_set_notification(q->tx_vq, 1);
2170 virtio_net_flush_tx(q);
2171 }
2172
2173 static void virtio_net_tx_bh(void *opaque)
2174 {
2175 VirtIONetQueue *q = opaque;
2176 VirtIONet *n = q->n;
2177 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2178 int32_t ret;
2179
2180 /* This happens when device was stopped but BH wasn't. */
2181 if (!vdev->vm_running) {
2182 /* Make sure tx waiting is set, so we'll run when restarted. */
2183 assert(q->tx_waiting);
2184 return;
2185 }
2186
2187 q->tx_waiting = 0;
2188
2189 /* Just in case the driver is not ready on more */
2190 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
2191 return;
2192 }
2193
2194 ret = virtio_net_flush_tx(q);
2195 if (ret == -EBUSY || ret == -EINVAL) {
2196 return; /* Notification re-enable handled by tx_complete or device
2197 * broken */
2198 }
2199
2200 /* If we flush a full burst of packets, assume there are
2201 * more coming and immediately reschedule */
2202 if (ret >= n->tx_burst) {
2203 qemu_bh_schedule(q->tx_bh);
2204 q->tx_waiting = 1;
2205 return;
2206 }
2207
2208 /* If less than a full burst, re-enable notification and flush
2209 * anything that may have come in while we weren't looking. If
2210 * we find something, assume the guest is still active and reschedule */
2211 virtio_queue_set_notification(q->tx_vq, 1);
2212 ret = virtio_net_flush_tx(q);
2213 if (ret == -EINVAL) {
2214 return;
2215 } else if (ret > 0) {
2216 virtio_queue_set_notification(q->tx_vq, 0);
2217 qemu_bh_schedule(q->tx_bh);
2218 q->tx_waiting = 1;
2219 }
2220 }
2221
2222 static void virtio_net_add_queue(VirtIONet *n, int index)
2223 {
2224 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2225
2226 n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
2227 virtio_net_handle_rx);
2228
2229 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
2230 n->vqs[index].tx_vq =
2231 virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2232 virtio_net_handle_tx_timer);
2233 n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
2234 virtio_net_tx_timer,
2235 &n->vqs[index]);
2236 } else {
2237 n->vqs[index].tx_vq =
2238 virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2239 virtio_net_handle_tx_bh);
2240 n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
2241 }
2242
2243 n->vqs[index].tx_waiting = 0;
2244 n->vqs[index].n = n;
2245 }
2246
2247 static void virtio_net_del_queue(VirtIONet *n, int index)
2248 {
2249 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2250 VirtIONetQueue *q = &n->vqs[index];
2251 NetClientState *nc = qemu_get_subqueue(n->nic, index);
2252
2253 qemu_purge_queued_packets(nc);
2254
2255 virtio_del_queue(vdev, index * 2);
2256 if (q->tx_timer) {
2257 timer_del(q->tx_timer);
2258 timer_free(q->tx_timer);
2259 q->tx_timer = NULL;
2260 } else {
2261 qemu_bh_delete(q->tx_bh);
2262 q->tx_bh = NULL;
2263 }
2264 q->tx_waiting = 0;
2265 virtio_del_queue(vdev, index * 2 + 1);
2266 }
2267
2268 static void virtio_net_change_num_queues(VirtIONet *n, int new_max_queues)
2269 {
2270 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2271 int old_num_queues = virtio_get_num_queues(vdev);
2272 int new_num_queues = new_max_queues * 2 + 1;
2273 int i;
2274
2275 assert(old_num_queues >= 3);
2276 assert(old_num_queues % 2 == 1);
2277
2278 if (old_num_queues == new_num_queues) {
2279 return;
2280 }
2281
2282 /*
2283 * We always need to remove and add ctrl vq if
2284 * old_num_queues != new_num_queues. Remove ctrl_vq first,
2285 * and then we only enter one of the following two loops.
2286 */
2287 virtio_del_queue(vdev, old_num_queues - 1);
2288
2289 for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
2290 /* new_num_queues < old_num_queues */
2291 virtio_net_del_queue(n, i / 2);
2292 }
2293
2294 for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
2295 /* new_num_queues > old_num_queues */
2296 virtio_net_add_queue(n, i / 2);
2297 }
2298
2299 /* add ctrl_vq last */
2300 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
2301 }
2302
2303 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
2304 {
2305 int max = multiqueue ? n->max_queues : 1;
2306
2307 n->multiqueue = multiqueue;
2308 virtio_net_change_num_queues(n, max);
2309
2310 virtio_net_set_queues(n);
2311 }
2312
2313 static int virtio_net_post_load_device(void *opaque, int version_id)
2314 {
2315 VirtIONet *n = opaque;
2316 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2317 int i, link_down;
2318
2319 trace_virtio_net_post_load_device();
2320 virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
2321 virtio_vdev_has_feature(vdev,
2322 VIRTIO_F_VERSION_1));
2323
2324 /* MAC_TABLE_ENTRIES may be different from the saved image */
2325 if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
2326 n->mac_table.in_use = 0;
2327 }
2328
2329 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
2330 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
2331 }
2332
2333 if (peer_has_vnet_hdr(n)) {
2334 virtio_net_apply_guest_offloads(n);
2335 }
2336
2337 virtio_net_set_queues(n);
2338
2339 /* Find the first multicast entry in the saved MAC filter */
2340 for (i = 0; i < n->mac_table.in_use; i++) {
2341 if (n->mac_table.macs[i * ETH_ALEN] & 1) {
2342 break;
2343 }
2344 }
2345 n->mac_table.first_multi = i;
2346
2347 /* nc.link_down can't be migrated, so infer link_down according
2348 * to link status bit in n->status */
2349 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
2350 for (i = 0; i < n->max_queues; i++) {
2351 qemu_get_subqueue(n->nic, i)->link_down = link_down;
2352 }
2353
2354 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
2355 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
2356 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
2357 QEMU_CLOCK_VIRTUAL,
2358 virtio_net_announce_timer, n);
2359 if (n->announce_timer.round) {
2360 timer_mod(n->announce_timer.tm,
2361 qemu_clock_get_ms(n->announce_timer.type));
2362 } else {
2363 qemu_announce_timer_del(&n->announce_timer);
2364 }
2365 }
2366
2367 return 0;
2368 }
2369
2370 /* tx_waiting field of a VirtIONetQueue */
2371 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = {
2372 .name = "virtio-net-queue-tx_waiting",
2373 .fields = (VMStateField[]) {
2374 VMSTATE_UINT32(tx_waiting, VirtIONetQueue),
2375 VMSTATE_END_OF_LIST()
2376 },
2377 };
2378
2379 static bool max_queues_gt_1(void *opaque, int version_id)
2380 {
2381 return VIRTIO_NET(opaque)->max_queues > 1;
2382 }
2383
2384 static bool has_ctrl_guest_offloads(void *opaque, int version_id)
2385 {
2386 return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque),
2387 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
2388 }
2389
2390 static bool mac_table_fits(void *opaque, int version_id)
2391 {
2392 return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES;
2393 }
2394
2395 static bool mac_table_doesnt_fit(void *opaque, int version_id)
2396 {
2397 return !mac_table_fits(opaque, version_id);
2398 }
2399
2400 /* This temporary type is shared by all the WITH_TMP methods
2401 * although only some fields are used by each.
2402 */
2403 struct VirtIONetMigTmp {
2404 VirtIONet *parent;
2405 VirtIONetQueue *vqs_1;
2406 uint16_t curr_queues_1;
2407 uint8_t has_ufo;
2408 uint32_t has_vnet_hdr;
2409 };
2410
2411 /* The 2nd and subsequent tx_waiting flags are loaded later than
2412 * the 1st entry in the queues and only if there's more than one
2413 * entry. We use the tmp mechanism to calculate a temporary
2414 * pointer and count and also validate the count.
2415 */
2416
2417 static int virtio_net_tx_waiting_pre_save(void *opaque)
2418 {
2419 struct VirtIONetMigTmp *tmp = opaque;
2420
2421 tmp->vqs_1 = tmp->parent->vqs + 1;
2422 tmp->curr_queues_1 = tmp->parent->curr_queues - 1;
2423 if (tmp->parent->curr_queues == 0) {
2424 tmp->curr_queues_1 = 0;
2425 }
2426
2427 return 0;
2428 }
2429
2430 static int virtio_net_tx_waiting_pre_load(void *opaque)
2431 {
2432 struct VirtIONetMigTmp *tmp = opaque;
2433
2434 /* Reuse the pointer setup from save */
2435 virtio_net_tx_waiting_pre_save(opaque);
2436
2437 if (tmp->parent->curr_queues > tmp->parent->max_queues) {
2438 error_report("virtio-net: curr_queues %x > max_queues %x",
2439 tmp->parent->curr_queues, tmp->parent->max_queues);
2440
2441 return -EINVAL;
2442 }
2443
2444 return 0; /* all good */
2445 }
2446
2447 static const VMStateDescription vmstate_virtio_net_tx_waiting = {
2448 .name = "virtio-net-tx_waiting",
2449 .pre_load = virtio_net_tx_waiting_pre_load,
2450 .pre_save = virtio_net_tx_waiting_pre_save,
2451 .fields = (VMStateField[]) {
2452 VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp,
2453 curr_queues_1,
2454 vmstate_virtio_net_queue_tx_waiting,
2455 struct VirtIONetQueue),
2456 VMSTATE_END_OF_LIST()
2457 },
2458 };
2459
2460 /* the 'has_ufo' flag is just tested; if the incoming stream has the
2461 * flag set we need to check that we have it
2462 */
2463 static int virtio_net_ufo_post_load(void *opaque, int version_id)
2464 {
2465 struct VirtIONetMigTmp *tmp = opaque;
2466
2467 if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) {
2468 error_report("virtio-net: saved image requires TUN_F_UFO support");
2469 return -EINVAL;
2470 }
2471
2472 return 0;
2473 }
2474
2475 static int virtio_net_ufo_pre_save(void *opaque)
2476 {
2477 struct VirtIONetMigTmp *tmp = opaque;
2478
2479 tmp->has_ufo = tmp->parent->has_ufo;
2480
2481 return 0;
2482 }
2483
2484 static const VMStateDescription vmstate_virtio_net_has_ufo = {
2485 .name = "virtio-net-ufo",
2486 .post_load = virtio_net_ufo_post_load,
2487 .pre_save = virtio_net_ufo_pre_save,
2488 .fields = (VMStateField[]) {
2489 VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp),
2490 VMSTATE_END_OF_LIST()
2491 },
2492 };
2493
2494 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
2495 * flag set we need to check that we have it
2496 */
2497 static int virtio_net_vnet_post_load(void *opaque, int version_id)
2498 {
2499 struct VirtIONetMigTmp *tmp = opaque;
2500
2501 if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) {
2502 error_report("virtio-net: saved image requires vnet_hdr=on");
2503 return -EINVAL;
2504 }
2505
2506 return 0;
2507 }
2508
2509 static int virtio_net_vnet_pre_save(void *opaque)
2510 {
2511 struct VirtIONetMigTmp *tmp = opaque;
2512
2513 tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr;
2514
2515 return 0;
2516 }
2517
2518 static const VMStateDescription vmstate_virtio_net_has_vnet = {
2519 .name = "virtio-net-vnet",
2520 .post_load = virtio_net_vnet_post_load,
2521 .pre_save = virtio_net_vnet_pre_save,
2522 .fields = (VMStateField[]) {
2523 VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp),
2524 VMSTATE_END_OF_LIST()
2525 },
2526 };
2527
2528 static const VMStateDescription vmstate_virtio_net_device = {
2529 .name = "virtio-net-device",
2530 .version_id = VIRTIO_NET_VM_VERSION,
2531 .minimum_version_id = VIRTIO_NET_VM_VERSION,
2532 .post_load = virtio_net_post_load_device,
2533 .fields = (VMStateField[]) {
2534 VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN),
2535 VMSTATE_STRUCT_POINTER(vqs, VirtIONet,
2536 vmstate_virtio_net_queue_tx_waiting,
2537 VirtIONetQueue),
2538 VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet),
2539 VMSTATE_UINT16(status, VirtIONet),
2540 VMSTATE_UINT8(promisc, VirtIONet),
2541 VMSTATE_UINT8(allmulti, VirtIONet),
2542 VMSTATE_UINT32(mac_table.in_use, VirtIONet),
2543
2544 /* Guarded pair: If it fits we load it, else we throw it away
2545 * - can happen if source has a larger MAC table.; post-load
2546 * sets flags in this case.
2547 */
2548 VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet,
2549 0, mac_table_fits, mac_table.in_use,
2550 ETH_ALEN),
2551 VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0,
2552 mac_table.in_use, ETH_ALEN),
2553
2554 /* Note: This is an array of uint32's that's always been saved as a
2555 * buffer; hold onto your endiannesses; it's actually used as a bitmap
2556 * but based on the uint.
2557 */
2558 VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3),
2559 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
2560 vmstate_virtio_net_has_vnet),
2561 VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet),
2562 VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet),
2563 VMSTATE_UINT8(alluni, VirtIONet),
2564 VMSTATE_UINT8(nomulti, VirtIONet),
2565 VMSTATE_UINT8(nouni, VirtIONet),
2566 VMSTATE_UINT8(nobcast, VirtIONet),
2567 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
2568 vmstate_virtio_net_has_ufo),
2569 VMSTATE_SINGLE_TEST(max_queues, VirtIONet, max_queues_gt_1, 0,
2570 vmstate_info_uint16_equal, uint16_t),
2571 VMSTATE_UINT16_TEST(curr_queues, VirtIONet, max_queues_gt_1),
2572 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
2573 vmstate_virtio_net_tx_waiting),
2574 VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet,
2575 has_ctrl_guest_offloads),
2576 VMSTATE_END_OF_LIST()
2577 },
2578 };
2579
2580 static NetClientInfo net_virtio_info = {
2581 .type = NET_CLIENT_DRIVER_NIC,
2582 .size = sizeof(NICState),
2583 .can_receive = virtio_net_can_receive,
2584 .receive = virtio_net_receive,
2585 .link_status_changed = virtio_net_set_link_status,
2586 .query_rx_filter = virtio_net_query_rxfilter,
2587 .announce = virtio_net_announce,
2588 };
2589
2590 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
2591 {
2592 VirtIONet *n = VIRTIO_NET(vdev);
2593 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
2594 assert(n->vhost_started);
2595 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
2596 }
2597
2598 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
2599 bool mask)
2600 {
2601 VirtIONet *n = VIRTIO_NET(vdev);
2602 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
2603 assert(n->vhost_started);
2604 vhost_net_virtqueue_mask(get_vhost_net(nc->peer),
2605 vdev, idx, mask);
2606 }
2607
2608 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
2609 {
2610 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
2611
2612 n->config_size = virtio_feature_get_config_size(feature_sizes,
2613 host_features);
2614 }
2615
2616 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
2617 const char *type)
2618 {
2619 /*
2620 * The name can be NULL, the netclient name will be type.x.
2621 */
2622 assert(type != NULL);
2623
2624 g_free(n->netclient_name);
2625 g_free(n->netclient_type);
2626 n->netclient_name = g_strdup(name);
2627 n->netclient_type = g_strdup(type);
2628 }
2629
2630 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
2631 {
2632 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2633 VirtIONet *n = VIRTIO_NET(dev);
2634 NetClientState *nc;
2635 int i;
2636
2637 if (n->net_conf.mtu) {
2638 n->host_features |= (1ULL << VIRTIO_NET_F_MTU);
2639 }
2640
2641 if (n->net_conf.duplex_str) {
2642 if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) {
2643 n->net_conf.duplex = DUPLEX_HALF;
2644 } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) {
2645 n->net_conf.duplex = DUPLEX_FULL;
2646 } else {
2647 error_setg(errp, "'duplex' must be 'half' or 'full'");
2648 }
2649 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
2650 } else {
2651 n->net_conf.duplex = DUPLEX_UNKNOWN;
2652 }
2653
2654 if (n->net_conf.speed < SPEED_UNKNOWN) {
2655 error_setg(errp, "'speed' must be between 0 and INT_MAX");
2656 } else if (n->net_conf.speed >= 0) {
2657 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
2658 }
2659
2660 virtio_net_set_config_size(n, n->host_features);
2661 virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size);
2662
2663 /*
2664 * We set a lower limit on RX queue size to what it always was.
2665 * Guests that want a smaller ring can always resize it without
2666 * help from us (using virtio 1 and up).
2667 */
2668 if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
2669 n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
2670 !is_power_of_2(n->net_conf.rx_queue_size)) {
2671 error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
2672 "must be a power of 2 between %d and %d.",
2673 n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
2674 VIRTQUEUE_MAX_SIZE);
2675 virtio_cleanup(vdev);
2676 return;
2677 }
2678
2679 if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE ||
2680 n->net_conf.tx_queue_size > VIRTQUEUE_MAX_SIZE ||
2681 !is_power_of_2(n->net_conf.tx_queue_size)) {
2682 error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), "
2683 "must be a power of 2 between %d and %d",
2684 n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE,
2685 VIRTQUEUE_MAX_SIZE);
2686 virtio_cleanup(vdev);
2687 return;
2688 }
2689
2690 n->max_queues = MAX(n->nic_conf.peers.queues, 1);
2691 if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) {
2692 error_setg(errp, "Invalid number of queues (= %" PRIu32 "), "
2693 "must be a positive integer less than %d.",
2694 n->max_queues, (VIRTIO_QUEUE_MAX - 1) / 2);
2695 virtio_cleanup(vdev);
2696 return;
2697 }
2698 n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
2699 n->curr_queues = 1;
2700 n->tx_timeout = n->net_conf.txtimer;
2701
2702 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
2703 && strcmp(n->net_conf.tx, "bh")) {
2704 warn_report("virtio-net: "
2705 "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
2706 n->net_conf.tx);
2707 error_printf("Defaulting to \"bh\"");
2708 }
2709
2710 n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n),
2711 n->net_conf.tx_queue_size);
2712
2713 for (i = 0; i < n->max_queues; i++) {
2714 virtio_net_add_queue(n, i);
2715 }
2716
2717 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
2718 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
2719 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
2720 n->status = VIRTIO_NET_S_LINK_UP;
2721 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
2722 QEMU_CLOCK_VIRTUAL,
2723 virtio_net_announce_timer, n);
2724 n->announce_timer.round = 0;
2725
2726 if (n->netclient_type) {
2727 /*
2728 * Happen when virtio_net_set_netclient_name has been called.
2729 */
2730 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
2731 n->netclient_type, n->netclient_name, n);
2732 } else {
2733 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
2734 object_get_typename(OBJECT(dev)), dev->id, n);
2735 }
2736
2737 peer_test_vnet_hdr(n);
2738 if (peer_has_vnet_hdr(n)) {
2739 for (i = 0; i < n->max_queues; i++) {
2740 qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
2741 }
2742 n->host_hdr_len = sizeof(struct virtio_net_hdr);
2743 } else {
2744 n->host_hdr_len = 0;
2745 }
2746
2747 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
2748
2749 n->vqs[0].tx_waiting = 0;
2750 n->tx_burst = n->net_conf.txburst;
2751 virtio_net_set_mrg_rx_bufs(n, 0, 0);
2752 n->promisc = 1; /* for compatibility */
2753
2754 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
2755
2756 n->vlans = g_malloc0(MAX_VLAN >> 3);
2757
2758 nc = qemu_get_queue(n->nic);
2759 nc->rxfilter_notify_enabled = 1;
2760
2761 QTAILQ_INIT(&n->rsc_chains);
2762 n->qdev = dev;
2763 }
2764
2765 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp)
2766 {
2767 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2768 VirtIONet *n = VIRTIO_NET(dev);
2769 int i, max_queues;
2770
2771 /* This will stop vhost backend if appropriate. */
2772 virtio_net_set_status(vdev, 0);
2773
2774 g_free(n->netclient_name);
2775 n->netclient_name = NULL;
2776 g_free(n->netclient_type);
2777 n->netclient_type = NULL;
2778
2779 g_free(n->mac_table.macs);
2780 g_free(n->vlans);
2781
2782 max_queues = n->multiqueue ? n->max_queues : 1;
2783 for (i = 0; i < max_queues; i++) {
2784 virtio_net_del_queue(n, i);
2785 }
2786
2787 qemu_announce_timer_del(&n->announce_timer);
2788 g_free(n->vqs);
2789 qemu_del_nic(n->nic);
2790 virtio_net_rsc_cleanup(n);
2791 virtio_cleanup(vdev);
2792 }
2793
2794 static void virtio_net_instance_init(Object *obj)
2795 {
2796 VirtIONet *n = VIRTIO_NET(obj);
2797
2798 /*
2799 * The default config_size is sizeof(struct virtio_net_config).
2800 * Can be overriden with virtio_net_set_config_size.
2801 */
2802 n->config_size = sizeof(struct virtio_net_config);
2803 device_add_bootindex_property(obj, &n->nic_conf.bootindex,
2804 "bootindex", "/ethernet-phy@0",
2805 DEVICE(n), NULL);
2806 }
2807
2808 static int virtio_net_pre_save(void *opaque)
2809 {
2810 VirtIONet *n = opaque;
2811
2812 /* At this point, backend must be stopped, otherwise
2813 * it might keep writing to memory. */
2814 assert(!n->vhost_started);
2815
2816 return 0;
2817 }
2818
2819 static const VMStateDescription vmstate_virtio_net = {
2820 .name = "virtio-net",
2821 .minimum_version_id = VIRTIO_NET_VM_VERSION,
2822 .version_id = VIRTIO_NET_VM_VERSION,
2823 .fields = (VMStateField[]) {
2824 VMSTATE_VIRTIO_DEVICE,
2825 VMSTATE_END_OF_LIST()
2826 },
2827 .pre_save = virtio_net_pre_save,
2828 };
2829
2830 static Property virtio_net_properties[] = {
2831 DEFINE_PROP_BIT64("csum", VirtIONet, host_features,
2832 VIRTIO_NET_F_CSUM, true),
2833 DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features,
2834 VIRTIO_NET_F_GUEST_CSUM, true),
2835 DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
2836 DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features,
2837 VIRTIO_NET_F_GUEST_TSO4, true),
2838 DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features,
2839 VIRTIO_NET_F_GUEST_TSO6, true),
2840 DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features,
2841 VIRTIO_NET_F_GUEST_ECN, true),
2842 DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features,
2843 VIRTIO_NET_F_GUEST_UFO, true),
2844 DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features,
2845 VIRTIO_NET_F_GUEST_ANNOUNCE, true),
2846 DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features,
2847 VIRTIO_NET_F_HOST_TSO4, true),
2848 DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features,
2849 VIRTIO_NET_F_HOST_TSO6, true),
2850 DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features,
2851 VIRTIO_NET_F_HOST_ECN, true),
2852 DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features,
2853 VIRTIO_NET_F_HOST_UFO, true),
2854 DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features,
2855 VIRTIO_NET_F_MRG_RXBUF, true),
2856 DEFINE_PROP_BIT64("status", VirtIONet, host_features,
2857 VIRTIO_NET_F_STATUS, true),
2858 DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features,
2859 VIRTIO_NET_F_CTRL_VQ, true),
2860 DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features,
2861 VIRTIO_NET_F_CTRL_RX, true),
2862 DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features,
2863 VIRTIO_NET_F_CTRL_VLAN, true),
2864 DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features,
2865 VIRTIO_NET_F_CTRL_RX_EXTRA, true),
2866 DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features,
2867 VIRTIO_NET_F_CTRL_MAC_ADDR, true),
2868 DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features,
2869 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
2870 DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
2871 DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features,
2872 VIRTIO_NET_F_RSC_EXT, false),
2873 DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout,
2874 VIRTIO_NET_RSC_DEFAULT_INTERVAL),
2875 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
2876 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
2877 TX_TIMER_INTERVAL),
2878 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
2879 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
2880 DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
2881 VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
2882 DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size,
2883 VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE),
2884 DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
2885 DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend,
2886 true),
2887 DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN),
2888 DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str),
2889 DEFINE_PROP_END_OF_LIST(),
2890 };
2891
2892 static void virtio_net_class_init(ObjectClass *klass, void *data)
2893 {
2894 DeviceClass *dc = DEVICE_CLASS(klass);
2895 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
2896
2897 dc->props = virtio_net_properties;
2898 dc->vmsd = &vmstate_virtio_net;
2899 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
2900 vdc->realize = virtio_net_device_realize;
2901 vdc->unrealize = virtio_net_device_unrealize;
2902 vdc->get_config = virtio_net_get_config;
2903 vdc->set_config = virtio_net_set_config;
2904 vdc->get_features = virtio_net_get_features;
2905 vdc->set_features = virtio_net_set_features;
2906 vdc->bad_features = virtio_net_bad_features;
2907 vdc->reset = virtio_net_reset;
2908 vdc->set_status = virtio_net_set_status;
2909 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
2910 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
2911 vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
2912 vdc->vmsd = &vmstate_virtio_net_device;
2913 }
2914
2915 static const TypeInfo virtio_net_info = {
2916 .name = TYPE_VIRTIO_NET,
2917 .parent = TYPE_VIRTIO_DEVICE,
2918 .instance_size = sizeof(VirtIONet),
2919 .instance_init = virtio_net_instance_init,
2920 .class_init = virtio_net_class_init,
2921 };
2922
2923 static void virtio_register_types(void)
2924 {
2925 type_register_static(&virtio_net_info);
2926 }
2927
2928 type_init(virtio_register_types)
QEMU mirror