?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
virtio: introduce macro VIRTIO_CONFIG_IRQ_IDX
[qemu.git] / hw / net / virtio-net.c
blob0bfe454c23c83afe51377fab9872bcb9ec6279a6
1 /*
2 * Virtio Network Device
3 *
4 * Copyright IBM, Corp. 2007
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 */
13
14 #include "qemu/osdep.h"
15 #include "qemu/atomic.h"
16 #include "qemu/iov.h"
17 #include "qemu/log.h"
18 #include "qemu/main-loop.h"
19 #include "qemu/module.h"
20 #include "hw/virtio/virtio.h"
21 #include "net/net.h"
22 #include "net/checksum.h"
23 #include "net/tap.h"
24 #include "qemu/error-report.h"
25 #include "qemu/timer.h"
26 #include "qemu/option.h"
27 #include "qemu/option_int.h"
28 #include "qemu/config-file.h"
29 #include "qapi/qmp/qdict.h"
30 #include "hw/virtio/virtio-net.h"
31 #include "net/vhost_net.h"
32 #include "net/announce.h"
33 #include "hw/virtio/virtio-bus.h"
34 #include "qapi/error.h"
35 #include "qapi/qapi-events-net.h"
36 #include "hw/qdev-properties.h"
37 #include "qapi/qapi-types-migration.h"
38 #include "qapi/qapi-events-migration.h"
39 #include "hw/virtio/virtio-access.h"
40 #include "migration/misc.h"
41 #include "standard-headers/linux/ethtool.h"
42 #include "sysemu/sysemu.h"
43 #include "trace.h"
44 #include "monitor/qdev.h"
45 #include "hw/pci/pci.h"
46 #include "net_rx_pkt.h"
47 #include "hw/virtio/vhost.h"
48 #include "sysemu/qtest.h"
49
50 #define VIRTIO_NET_VM_VERSION 11
51
52 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */
53
54 /* previously fixed value */
55 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
56 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
57
58 /* for now, only allow larger queue_pairs; with virtio-1, guest can downsize */
59 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
60 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
61
62 #define VIRTIO_NET_IP4_ADDR_SIZE 8 /* ipv4 saddr + daddr */
63
64 #define VIRTIO_NET_TCP_FLAG 0x3F
65 #define VIRTIO_NET_TCP_HDR_LENGTH 0xF000
66
67 /* IPv4 max payload, 16 bits in the header */
68 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header))
69 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535
70
71 /* header length value in ip header without option */
72 #define VIRTIO_NET_IP4_HEADER_LENGTH 5
73
74 #define VIRTIO_NET_IP6_ADDR_SIZE 32 /* ipv6 saddr + daddr */
75 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD
76
77 /* Purge coalesced packets timer interval, This value affects the performance
78 a lot, and should be tuned carefully, '300000'(300us) is the recommended
79 value to pass the WHQL test, '50000' can gain 2x netperf throughput with
80 tso/gso/gro 'off'. */
81 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000
82
83 #define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \
84 VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \
85 VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \
86 VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \
87 VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \
88 VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \
89 VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \
90 VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \
91 VIRTIO_NET_RSS_HASH_TYPE_UDP_EX)
92
93 static const VirtIOFeature feature_sizes[] = {
94 {.flags = 1ULL << VIRTIO_NET_F_MAC,
95 .end = endof(struct virtio_net_config, mac)},
96 {.flags = 1ULL << VIRTIO_NET_F_STATUS,
97 .end = endof(struct virtio_net_config, status)},
98 {.flags = 1ULL << VIRTIO_NET_F_MQ,
99 .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
100 {.flags = 1ULL << VIRTIO_NET_F_MTU,
101 .end = endof(struct virtio_net_config, mtu)},
102 {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX,
103 .end = endof(struct virtio_net_config, duplex)},
104 {.flags = (1ULL << VIRTIO_NET_F_RSS) | (1ULL << VIRTIO_NET_F_HASH_REPORT),
105 .end = endof(struct virtio_net_config, supported_hash_types)},
106 {}
107 };
108
109 static const VirtIOConfigSizeParams cfg_size_params = {
110 .min_size = endof(struct virtio_net_config, mac),
111 .max_size = sizeof(struct virtio_net_config),
112 .feature_sizes = feature_sizes
113 };
114
115 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
116 {
117 VirtIONet *n = qemu_get_nic_opaque(nc);
118
119 return &n->vqs[nc->queue_index];
120 }
121
122 static int vq2q(int queue_index)
123 {
124 return queue_index / 2;
125 }
126
127 static void flush_or_purge_queued_packets(NetClientState *nc)
128 {
129 if (!nc->peer) {
130 return;
131 }
132
133 qemu_flush_or_purge_queued_packets(nc->peer, true);
134 assert(!virtio_net_get_subqueue(nc)->async_tx.elem);
135 }
136
137 /* TODO
138 * - we could suppress RX interrupt if we were so inclined.
139 */
140
141 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
142 {
143 VirtIONet *n = VIRTIO_NET(vdev);
144 struct virtio_net_config netcfg;
145 NetClientState *nc = qemu_get_queue(n->nic);
146 static const MACAddr zero = { .a = { 0, 0, 0, 0, 0, 0 } };
147
148 int ret = 0;
149 memset(&netcfg, 0 , sizeof(struct virtio_net_config));
150 virtio_stw_p(vdev, &netcfg.status, n->status);
151 virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queue_pairs);
152 virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu);
153 memcpy(netcfg.mac, n->mac, ETH_ALEN);
154 virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed);
155 netcfg.duplex = n->net_conf.duplex;
156 netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE;
157 virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length,
158 virtio_host_has_feature(vdev, VIRTIO_NET_F_RSS) ?
159 VIRTIO_NET_RSS_MAX_TABLE_LEN : 1);
160 virtio_stl_p(vdev, &netcfg.supported_hash_types,
161 VIRTIO_NET_RSS_SUPPORTED_HASHES);
162 memcpy(config, &netcfg, n->config_size);
163
164 /*
165 * Is this VDPA? No peer means not VDPA: there's no way to
166 * disconnect/reconnect a VDPA peer.
167 */
168 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
169 ret = vhost_net_get_config(get_vhost_net(nc->peer), (uint8_t *)&netcfg,
170 n->config_size);
171 if (ret == -1) {
172 return;
173 }
174
175 /*
176 * Some NIC/kernel combinations present 0 as the mac address. As that
177 * is not a legal address, try to proceed with the address from the
178 * QEMU command line in the hope that the address has been configured
179 * correctly elsewhere - just not reported by the device.
180 */
181 if (memcmp(&netcfg.mac, &zero, sizeof(zero)) == 0) {
182 info_report("Zero hardware mac address detected. Ignoring.");
183 memcpy(netcfg.mac, n->mac, ETH_ALEN);
184 }
185
186 netcfg.status |= virtio_tswap16(vdev,
187 n->status & VIRTIO_NET_S_ANNOUNCE);
188 memcpy(config, &netcfg, n->config_size);
189 }
190 }
191
192 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
193 {
194 VirtIONet *n = VIRTIO_NET(vdev);
195 struct virtio_net_config netcfg = {};
196 NetClientState *nc = qemu_get_queue(n->nic);
197
198 memcpy(&netcfg, config, n->config_size);
199
200 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
201 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) &&
202 memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
203 memcpy(n->mac, netcfg.mac, ETH_ALEN);
204 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
205 }
206
207 /*
208 * Is this VDPA? No peer means not VDPA: there's no way to
209 * disconnect/reconnect a VDPA peer.
210 */
211 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
212 vhost_net_set_config(get_vhost_net(nc->peer),
213 (uint8_t *)&netcfg, 0, n->config_size,
214 VHOST_SET_CONFIG_TYPE_MASTER);
215 }
216 }
217
218 static bool virtio_net_started(VirtIONet *n, uint8_t status)
219 {
220 VirtIODevice *vdev = VIRTIO_DEVICE(n);
221 return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
222 (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
223 }
224
225 static void virtio_net_announce_notify(VirtIONet *net)
226 {
227 VirtIODevice *vdev = VIRTIO_DEVICE(net);
228 trace_virtio_net_announce_notify();
229
230 net->status |= VIRTIO_NET_S_ANNOUNCE;
231 virtio_notify_config(vdev);
232 }
233
234 static void virtio_net_announce_timer(void *opaque)
235 {
236 VirtIONet *n = opaque;
237 trace_virtio_net_announce_timer(n->announce_timer.round);
238
239 n->announce_timer.round--;
240 virtio_net_announce_notify(n);
241 }
242
243 static void virtio_net_announce(NetClientState *nc)
244 {
245 VirtIONet *n = qemu_get_nic_opaque(nc);
246 VirtIODevice *vdev = VIRTIO_DEVICE(n);
247
248 /*
249 * Make sure the virtio migration announcement timer isn't running
250 * If it is, let it trigger announcement so that we do not cause
251 * confusion.
252 */
253 if (n->announce_timer.round) {
254 return;
255 }
256
257 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
258 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
259 virtio_net_announce_notify(n);
260 }
261 }
262
263 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
264 {
265 VirtIODevice *vdev = VIRTIO_DEVICE(n);
266 NetClientState *nc = qemu_get_queue(n->nic);
267 int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
268 int cvq = virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) ?
269 n->max_ncs - n->max_queue_pairs : 0;
270
271 if (!get_vhost_net(nc->peer)) {
272 return;
273 }
274
275 if ((virtio_net_started(n, status) && !nc->peer->link_down) ==
276 !!n->vhost_started) {
277 return;
278 }
279 if (!n->vhost_started) {
280 int r, i;
281
282 if (n->needs_vnet_hdr_swap) {
283 error_report("backend does not support %s vnet headers; "
284 "falling back on userspace virtio",
285 virtio_is_big_endian(vdev) ? "BE" : "LE");
286 return;
287 }
288
289 /* Any packets outstanding? Purge them to avoid touching rings
290 * when vhost is running.
291 */
292 for (i = 0; i < queue_pairs; i++) {
293 NetClientState *qnc = qemu_get_subqueue(n->nic, i);
294
295 /* Purge both directions: TX and RX. */
296 qemu_net_queue_purge(qnc->peer->incoming_queue, qnc);
297 qemu_net_queue_purge(qnc->incoming_queue, qnc->peer);
298 }
299
300 if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) {
301 r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu);
302 if (r < 0) {
303 error_report("%uBytes MTU not supported by the backend",
304 n->net_conf.mtu);
305
306 return;
307 }
308 }
309
310 n->vhost_started = 1;
311 r = vhost_net_start(vdev, n->nic->ncs, queue_pairs, cvq);
312 if (r < 0) {
313 error_report("unable to start vhost net: %d: "
314 "falling back on userspace virtio", -r);
315 n->vhost_started = 0;
316 }
317 } else {
318 vhost_net_stop(vdev, n->nic->ncs, queue_pairs, cvq);
319 n->vhost_started = 0;
320 }
321 }
322
323 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev,
324 NetClientState *peer,
325 bool enable)
326 {
327 if (virtio_is_big_endian(vdev)) {
328 return qemu_set_vnet_be(peer, enable);
329 } else {
330 return qemu_set_vnet_le(peer, enable);
331 }
332 }
333
334 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs,
335 int queue_pairs, bool enable)
336 {
337 int i;
338
339 for (i = 0; i < queue_pairs; i++) {
340 if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 &&
341 enable) {
342 while (--i >= 0) {
343 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false);
344 }
345
346 return true;
347 }
348 }
349
350 return false;
351 }
352
353 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status)
354 {
355 VirtIODevice *vdev = VIRTIO_DEVICE(n);
356 int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
357
358 if (virtio_net_started(n, status)) {
359 /* Before using the device, we tell the network backend about the
360 * endianness to use when parsing vnet headers. If the backend
361 * can't do it, we fallback onto fixing the headers in the core
362 * virtio-net code.
363 */
364 n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs,
365 queue_pairs, true);
366 } else if (virtio_net_started(n, vdev->status)) {
367 /* After using the device, we need to reset the network backend to
368 * the default (guest native endianness), otherwise the guest may
369 * lose network connectivity if it is rebooted into a different
370 * endianness.
371 */
372 virtio_net_set_vnet_endian(vdev, n->nic->ncs, queue_pairs, false);
373 }
374 }
375
376 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq)
377 {
378 unsigned int dropped = virtqueue_drop_all(vq);
379 if (dropped) {
380 virtio_notify(vdev, vq);
381 }
382 }
383
384 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
385 {
386 VirtIONet *n = VIRTIO_NET(vdev);
387 VirtIONetQueue *q;
388 int i;
389 uint8_t queue_status;
390
391 virtio_net_vnet_endian_status(n, status);
392 virtio_net_vhost_status(n, status);
393
394 for (i = 0; i < n->max_queue_pairs; i++) {
395 NetClientState *ncs = qemu_get_subqueue(n->nic, i);
396 bool queue_started;
397 q = &n->vqs[i];
398
399 if ((!n->multiqueue && i != 0) || i >= n->curr_queue_pairs) {
400 queue_status = 0;
401 } else {
402 queue_status = status;
403 }
404 queue_started =
405 virtio_net_started(n, queue_status) && !n->vhost_started;
406
407 if (queue_started) {
408 qemu_flush_queued_packets(ncs);
409 }
410
411 if (!q->tx_waiting) {
412 continue;
413 }
414
415 if (queue_started) {
416 if (q->tx_timer) {
417 timer_mod(q->tx_timer,
418 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
419 } else {
420 qemu_bh_schedule(q->tx_bh);
421 }
422 } else {
423 if (q->tx_timer) {
424 timer_del(q->tx_timer);
425 } else {
426 qemu_bh_cancel(q->tx_bh);
427 }
428 if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 &&
429 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) &&
430 vdev->vm_running) {
431 /* if tx is waiting we are likely have some packets in tx queue
432 * and disabled notification */
433 q->tx_waiting = 0;
434 virtio_queue_set_notification(q->tx_vq, 1);
435 virtio_net_drop_tx_queue_data(vdev, q->tx_vq);
436 }
437 }
438 }
439 }
440
441 static void virtio_net_set_link_status(NetClientState *nc)
442 {
443 VirtIONet *n = qemu_get_nic_opaque(nc);
444 VirtIODevice *vdev = VIRTIO_DEVICE(n);
445 uint16_t old_status = n->status;
446
447 if (nc->link_down)
448 n->status &= ~VIRTIO_NET_S_LINK_UP;
449 else
450 n->status |= VIRTIO_NET_S_LINK_UP;
451
452 if (n->status != old_status)
453 virtio_notify_config(vdev);
454
455 virtio_net_set_status(vdev, vdev->status);
456 }
457
458 static void rxfilter_notify(NetClientState *nc)
459 {
460 VirtIONet *n = qemu_get_nic_opaque(nc);
461
462 if (nc->rxfilter_notify_enabled) {
463 char *path = object_get_canonical_path(OBJECT(n->qdev));
464 qapi_event_send_nic_rx_filter_changed(n->netclient_name, path);
465 g_free(path);
466
467 /* disable event notification to avoid events flooding */
468 nc->rxfilter_notify_enabled = 0;
469 }
470 }
471
472 static intList *get_vlan_table(VirtIONet *n)
473 {
474 intList *list;
475 int i, j;
476
477 list = NULL;
478 for (i = 0; i < MAX_VLAN >> 5; i++) {
479 for (j = 0; n->vlans[i] && j <= 0x1f; j++) {
480 if (n->vlans[i] & (1U << j)) {
481 QAPI_LIST_PREPEND(list, (i << 5) + j);
482 }
483 }
484 }
485
486 return list;
487 }
488
489 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc)
490 {
491 VirtIONet *n = qemu_get_nic_opaque(nc);
492 VirtIODevice *vdev = VIRTIO_DEVICE(n);
493 RxFilterInfo *info;
494 strList *str_list;
495 int i;
496
497 info = g_malloc0(sizeof(*info));
498 info->name = g_strdup(nc->name);
499 info->promiscuous = n->promisc;
500
501 if (n->nouni) {
502 info->unicast = RX_STATE_NONE;
503 } else if (n->alluni) {
504 info->unicast = RX_STATE_ALL;
505 } else {
506 info->unicast = RX_STATE_NORMAL;
507 }
508
509 if (n->nomulti) {
510 info->multicast = RX_STATE_NONE;
511 } else if (n->allmulti) {
512 info->multicast = RX_STATE_ALL;
513 } else {
514 info->multicast = RX_STATE_NORMAL;
515 }
516
517 info->broadcast_allowed = n->nobcast;
518 info->multicast_overflow = n->mac_table.multi_overflow;
519 info->unicast_overflow = n->mac_table.uni_overflow;
520
521 info->main_mac = qemu_mac_strdup_printf(n->mac);
522
523 str_list = NULL;
524 for (i = 0; i < n->mac_table.first_multi; i++) {
525 QAPI_LIST_PREPEND(str_list,
526 qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN));
527 }
528 info->unicast_table = str_list;
529
530 str_list = NULL;
531 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
532 QAPI_LIST_PREPEND(str_list,
533 qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN));
534 }
535 info->multicast_table = str_list;
536 info->vlan_table = get_vlan_table(n);
537
538 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) {
539 info->vlan = RX_STATE_ALL;
540 } else if (!info->vlan_table) {
541 info->vlan = RX_STATE_NONE;
542 } else {
543 info->vlan = RX_STATE_NORMAL;
544 }
545
546 /* enable event notification after query */
547 nc->rxfilter_notify_enabled = 1;
548
549 return info;
550 }
551
552 static void virtio_net_queue_reset(VirtIODevice *vdev, uint32_t queue_index)
553 {
554 VirtIONet *n = VIRTIO_NET(vdev);
555 NetClientState *nc;
556
557 /* validate queue_index and skip for cvq */
558 if (queue_index >= n->max_queue_pairs * 2) {
559 return;
560 }
561
562 nc = qemu_get_subqueue(n->nic, vq2q(queue_index));
563
564 if (!nc->peer) {
565 return;
566 }
567
568 if (get_vhost_net(nc->peer) &&
569 nc->peer->info->type == NET_CLIENT_DRIVER_TAP) {
570 vhost_net_virtqueue_reset(vdev, nc, queue_index);
571 }
572
573 flush_or_purge_queued_packets(nc);
574 }
575
576 static void virtio_net_queue_enable(VirtIODevice *vdev, uint32_t queue_index)
577 {
578 VirtIONet *n = VIRTIO_NET(vdev);
579 NetClientState *nc;
580 int r;
581
582 /* validate queue_index and skip for cvq */
583 if (queue_index >= n->max_queue_pairs * 2) {
584 return;
585 }
586
587 nc = qemu_get_subqueue(n->nic, vq2q(queue_index));
588
589 if (!nc->peer || !vdev->vhost_started) {
590 return;
591 }
592
593 if (get_vhost_net(nc->peer) &&
594 nc->peer->info->type == NET_CLIENT_DRIVER_TAP) {
595 r = vhost_net_virtqueue_restart(vdev, nc, queue_index);
596 if (r < 0) {
597 error_report("unable to restart vhost net virtqueue: %d, "
598 "when resetting the queue", queue_index);
599 }
600 }
601 }
602
603 static void virtio_net_reset(VirtIODevice *vdev)
604 {
605 VirtIONet *n = VIRTIO_NET(vdev);
606 int i;
607
608 /* Reset back to compatibility mode */
609 n->promisc = 1;
610 n->allmulti = 0;
611 n->alluni = 0;
612 n->nomulti = 0;
613 n->nouni = 0;
614 n->nobcast = 0;
615 /* multiqueue is disabled by default */
616 n->curr_queue_pairs = 1;
617 timer_del(n->announce_timer.tm);
618 n->announce_timer.round = 0;
619 n->status &= ~VIRTIO_NET_S_ANNOUNCE;
620
621 /* Flush any MAC and VLAN filter table state */
622 n->mac_table.in_use = 0;
623 n->mac_table.first_multi = 0;
624 n->mac_table.multi_overflow = 0;
625 n->mac_table.uni_overflow = 0;
626 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
627 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
628 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
629 memset(n->vlans, 0, MAX_VLAN >> 3);
630
631 /* Flush any async TX */
632 for (i = 0; i < n->max_queue_pairs; i++) {
633 flush_or_purge_queued_packets(qemu_get_subqueue(n->nic, i));
634 }
635 }
636
637 static void peer_test_vnet_hdr(VirtIONet *n)
638 {
639 NetClientState *nc = qemu_get_queue(n->nic);
640 if (!nc->peer) {
641 return;
642 }
643
644 n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer);
645 }
646
647 static int peer_has_vnet_hdr(VirtIONet *n)
648 {
649 return n->has_vnet_hdr;
650 }
651
652 static int peer_has_ufo(VirtIONet *n)
653 {
654 if (!peer_has_vnet_hdr(n))
655 return 0;
656
657 n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer);
658
659 return n->has_ufo;
660 }
661
662 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs,
663 int version_1, int hash_report)
664 {
665 int i;
666 NetClientState *nc;
667
668 n->mergeable_rx_bufs = mergeable_rx_bufs;
669
670 if (version_1) {
671 n->guest_hdr_len = hash_report ?
672 sizeof(struct virtio_net_hdr_v1_hash) :
673 sizeof(struct virtio_net_hdr_mrg_rxbuf);
674 n->rss_data.populate_hash = !!hash_report;
675 } else {
676 n->guest_hdr_len = n->mergeable_rx_bufs ?
677 sizeof(struct virtio_net_hdr_mrg_rxbuf) :
678 sizeof(struct virtio_net_hdr);
679 }
680
681 for (i = 0; i < n->max_queue_pairs; i++) {
682 nc = qemu_get_subqueue(n->nic, i);
683
684 if (peer_has_vnet_hdr(n) &&
685 qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
686 qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
687 n->host_hdr_len = n->guest_hdr_len;
688 }
689 }
690 }
691
692 static int virtio_net_max_tx_queue_size(VirtIONet *n)
693 {
694 NetClientState *peer = n->nic_conf.peers.ncs[0];
695
696 /*
697 * Backends other than vhost-user or vhost-vdpa don't support max queue
698 * size.
699 */
700 if (!peer) {
701 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
702 }
703
704 switch(peer->info->type) {
705 case NET_CLIENT_DRIVER_VHOST_USER:
706 case NET_CLIENT_DRIVER_VHOST_VDPA:
707 return VIRTQUEUE_MAX_SIZE;
708 default:
709 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
710 };
711 }
712
713 static int peer_attach(VirtIONet *n, int index)
714 {
715 NetClientState *nc = qemu_get_subqueue(n->nic, index);
716
717 if (!nc->peer) {
718 return 0;
719 }
720
721 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
722 vhost_set_vring_enable(nc->peer, 1);
723 }
724
725 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
726 return 0;
727 }
728
729 if (n->max_queue_pairs == 1) {
730 return 0;
731 }
732
733 return tap_enable(nc->peer);
734 }
735
736 static int peer_detach(VirtIONet *n, int index)
737 {
738 NetClientState *nc = qemu_get_subqueue(n->nic, index);
739
740 if (!nc->peer) {
741 return 0;
742 }
743
744 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
745 vhost_set_vring_enable(nc->peer, 0);
746 }
747
748 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) {
749 return 0;
750 }
751
752 return tap_disable(nc->peer);
753 }
754
755 static void virtio_net_set_queue_pairs(VirtIONet *n)
756 {
757 int i;
758 int r;
759
760 if (n->nic->peer_deleted) {
761 return;
762 }
763
764 for (i = 0; i < n->max_queue_pairs; i++) {
765 if (i < n->curr_queue_pairs) {
766 r = peer_attach(n, i);
767 assert(!r);
768 } else {
769 r = peer_detach(n, i);
770 assert(!r);
771 }
772 }
773 }
774
775 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
776
777 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features,
778 Error **errp)
779 {
780 VirtIONet *n = VIRTIO_NET(vdev);
781 NetClientState *nc = qemu_get_queue(n->nic);
782
783 /* Firstly sync all virtio-net possible supported features */
784 features |= n->host_features;
785
786 virtio_add_feature(&features, VIRTIO_NET_F_MAC);
787
788 if (!peer_has_vnet_hdr(n)) {
789 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM);
790 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4);
791 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6);
792 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN);
793
794 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM);
795 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4);
796 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6);
797 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN);
798
799 virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT);
800 }
801
802 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
803 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO);
804 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO);
805 }
806
807 if (!get_vhost_net(nc->peer)) {
808 virtio_add_feature(&features, VIRTIO_F_RING_RESET);
809 return features;
810 }
811
812 if (!ebpf_rss_is_loaded(&n->ebpf_rss)) {
813 virtio_clear_feature(&features, VIRTIO_NET_F_RSS);
814 }
815 features = vhost_net_get_features(get_vhost_net(nc->peer), features);
816 vdev->backend_features = features;
817
818 if (n->mtu_bypass_backend &&
819 (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) {
820 features |= (1ULL << VIRTIO_NET_F_MTU);
821 }
822
823 return features;
824 }
825
826 static uint64_t virtio_net_bad_features(VirtIODevice *vdev)
827 {
828 uint64_t features = 0;
829
830 /* Linux kernel 2.6.25. It understood MAC (as everyone must),
831 * but also these: */
832 virtio_add_feature(&features, VIRTIO_NET_F_MAC);
833 virtio_add_feature(&features, VIRTIO_NET_F_CSUM);
834 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4);
835 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6);
836 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN);
837
838 return features;
839 }
840
841 static void virtio_net_apply_guest_offloads(VirtIONet *n)
842 {
843 qemu_set_offload(qemu_get_queue(n->nic)->peer,
844 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
845 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
846 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
847 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
848 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
849 }
850
851 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
852 {
853 static const uint64_t guest_offloads_mask =
854 (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
855 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
856 (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
857 (1ULL << VIRTIO_NET_F_GUEST_ECN) |
858 (1ULL << VIRTIO_NET_F_GUEST_UFO);
859
860 return guest_offloads_mask & features;
861 }
862
863 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
864 {
865 VirtIODevice *vdev = VIRTIO_DEVICE(n);
866 return virtio_net_guest_offloads_by_features(vdev->guest_features);
867 }
868
869 typedef struct {
870 VirtIONet *n;
871 DeviceState *dev;
872 } FailoverDevice;
873
874 /**
875 * Set the failover primary device
876 *
877 * @opaque: FailoverId to setup
878 * @opts: opts for device we are handling
879 * @errp: returns an error if this function fails
880 */
881 static int failover_set_primary(DeviceState *dev, void *opaque)
882 {
883 FailoverDevice *fdev = opaque;
884 PCIDevice *pci_dev = (PCIDevice *)
885 object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE);
886
887 if (!pci_dev) {
888 return 0;
889 }
890
891 if (!g_strcmp0(pci_dev->failover_pair_id, fdev->n->netclient_name)) {
892 fdev->dev = dev;
893 return 1;
894 }
895
896 return 0;
897 }
898
899 /**
900 * Find the primary device for this failover virtio-net
901 *
902 * @n: VirtIONet device
903 * @errp: returns an error if this function fails
904 */
905 static DeviceState *failover_find_primary_device(VirtIONet *n)
906 {
907 FailoverDevice fdev = {
908 .n = n,
909 };
910
911 qbus_walk_children(sysbus_get_default(), failover_set_primary, NULL,
912 NULL, NULL, &fdev);
913 return fdev.dev;
914 }
915
916 static void failover_add_primary(VirtIONet *n, Error **errp)
917 {
918 Error *err = NULL;
919 DeviceState *dev = failover_find_primary_device(n);
920
921 if (dev) {
922 return;
923 }
924
925 if (!n->primary_opts) {
926 error_setg(errp, "Primary device not found");
927 error_append_hint(errp, "Virtio-net failover will not work. Make "
928 "sure primary device has parameter"
929 " failover_pair_id=%s\n", n->netclient_name);
930 return;
931 }
932
933 dev = qdev_device_add_from_qdict(n->primary_opts,
934 n->primary_opts_from_json,
935 &err);
936 if (err) {
937 qobject_unref(n->primary_opts);
938 n->primary_opts = NULL;
939 } else {
940 object_unref(OBJECT(dev));
941 }
942 error_propagate(errp, err);
943 }
944
945 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features)
946 {
947 VirtIONet *n = VIRTIO_NET(vdev);
948 Error *err = NULL;
949 int i;
950
951 if (n->mtu_bypass_backend &&
952 !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) {
953 features &= ~(1ULL << VIRTIO_NET_F_MTU);
954 }
955
956 virtio_net_set_multiqueue(n,
957 virtio_has_feature(features, VIRTIO_NET_F_RSS) ||
958 virtio_has_feature(features, VIRTIO_NET_F_MQ));
959
960 virtio_net_set_mrg_rx_bufs(n,
961 virtio_has_feature(features,
962 VIRTIO_NET_F_MRG_RXBUF),
963 virtio_has_feature(features,
964 VIRTIO_F_VERSION_1),
965 virtio_has_feature(features,
966 VIRTIO_NET_F_HASH_REPORT));
967
968 n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
969 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4);
970 n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) &&
971 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6);
972 n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS);
973
974 if (n->has_vnet_hdr) {
975 n->curr_guest_offloads =
976 virtio_net_guest_offloads_by_features(features);
977 virtio_net_apply_guest_offloads(n);
978 }
979
980 for (i = 0; i < n->max_queue_pairs; i++) {
981 NetClientState *nc = qemu_get_subqueue(n->nic, i);
982
983 if (!get_vhost_net(nc->peer)) {
984 continue;
985 }
986 vhost_net_ack_features(get_vhost_net(nc->peer), features);
987
988 /*
989 * keep acked_features in NetVhostUserState up-to-date so it
990 * can't miss any features configured by guest virtio driver.
991 */
992 vhost_net_save_acked_features(nc->peer);
993 }
994
995 if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) {
996 memset(n->vlans, 0, MAX_VLAN >> 3);
997 } else {
998 memset(n->vlans, 0xff, MAX_VLAN >> 3);
999 }
1000
1001 if (virtio_has_feature(features, VIRTIO_NET_F_STANDBY)) {
1002 qapi_event_send_failover_negotiated(n->netclient_name);
1003 qatomic_set(&n->failover_primary_hidden, false);
1004 failover_add_primary(n, &err);
1005 if (err) {
1006 if (!qtest_enabled()) {
1007 warn_report_err(err);
1008 } else {
1009 error_free(err);
1010 }
1011 }
1012 }
1013 }
1014
1015 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
1016 struct iovec *iov, unsigned int iov_cnt)
1017 {
1018 uint8_t on;
1019 size_t s;
1020 NetClientState *nc = qemu_get_queue(n->nic);
1021
1022 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
1023 if (s != sizeof(on)) {
1024 return VIRTIO_NET_ERR;
1025 }
1026
1027 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
1028 n->promisc = on;
1029 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
1030 n->allmulti = on;
1031 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
1032 n->alluni = on;
1033 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
1034 n->nomulti = on;
1035 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
1036 n->nouni = on;
1037 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
1038 n->nobcast = on;
1039 } else {
1040 return VIRTIO_NET_ERR;
1041 }
1042
1043 rxfilter_notify(nc);
1044
1045 return VIRTIO_NET_OK;
1046 }
1047
1048 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
1049 struct iovec *iov, unsigned int iov_cnt)
1050 {
1051 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1052 uint64_t offloads;
1053 size_t s;
1054
1055 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
1056 return VIRTIO_NET_ERR;
1057 }
1058
1059 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
1060 if (s != sizeof(offloads)) {
1061 return VIRTIO_NET_ERR;
1062 }
1063
1064 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
1065 uint64_t supported_offloads;
1066
1067 offloads = virtio_ldq_p(vdev, &offloads);
1068
1069 if (!n->has_vnet_hdr) {
1070 return VIRTIO_NET_ERR;
1071 }
1072
1073 n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
1074 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4);
1075 n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) &&
1076 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6);
1077 virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT);
1078
1079 supported_offloads = virtio_net_supported_guest_offloads(n);
1080 if (offloads & ~supported_offloads) {
1081 return VIRTIO_NET_ERR;
1082 }
1083
1084 n->curr_guest_offloads = offloads;
1085 virtio_net_apply_guest_offloads(n);
1086
1087 return VIRTIO_NET_OK;
1088 } else {
1089 return VIRTIO_NET_ERR;
1090 }
1091 }
1092
1093 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
1094 struct iovec *iov, unsigned int iov_cnt)
1095 {
1096 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1097 struct virtio_net_ctrl_mac mac_data;
1098 size_t s;
1099 NetClientState *nc = qemu_get_queue(n->nic);
1100
1101 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
1102 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
1103 return VIRTIO_NET_ERR;
1104 }
1105 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
1106 assert(s == sizeof(n->mac));
1107 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
1108 rxfilter_notify(nc);
1109
1110 return VIRTIO_NET_OK;
1111 }
1112
1113 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
1114 return VIRTIO_NET_ERR;
1115 }
1116
1117 int in_use = 0;
1118 int first_multi = 0;
1119 uint8_t uni_overflow = 0;
1120 uint8_t multi_overflow = 0;
1121 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1122
1123 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1124 sizeof(mac_data.entries));
1125 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1126 if (s != sizeof(mac_data.entries)) {
1127 goto error;
1128 }
1129 iov_discard_front(&iov, &iov_cnt, s);
1130
1131 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
1132 goto error;
1133 }
1134
1135 if (mac_data.entries <= MAC_TABLE_ENTRIES) {
1136 s = iov_to_buf(iov, iov_cnt, 0, macs,
1137 mac_data.entries * ETH_ALEN);
1138 if (s != mac_data.entries * ETH_ALEN) {
1139 goto error;
1140 }
1141 in_use += mac_data.entries;
1142 } else {
1143 uni_overflow = 1;
1144 }
1145
1146 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
1147
1148 first_multi = in_use;
1149
1150 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
1151 sizeof(mac_data.entries));
1152 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries);
1153 if (s != sizeof(mac_data.entries)) {
1154 goto error;
1155 }
1156
1157 iov_discard_front(&iov, &iov_cnt, s);
1158
1159 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
1160 goto error;
1161 }
1162
1163 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) {
1164 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN],
1165 mac_data.entries * ETH_ALEN);
1166 if (s != mac_data.entries * ETH_ALEN) {
1167 goto error;
1168 }
1169 in_use += mac_data.entries;
1170 } else {
1171 multi_overflow = 1;
1172 }
1173
1174 n->mac_table.in_use = in_use;
1175 n->mac_table.first_multi = first_multi;
1176 n->mac_table.uni_overflow = uni_overflow;
1177 n->mac_table.multi_overflow = multi_overflow;
1178 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN);
1179 g_free(macs);
1180 rxfilter_notify(nc);
1181
1182 return VIRTIO_NET_OK;
1183
1184 error:
1185 g_free(macs);
1186 return VIRTIO_NET_ERR;
1187 }
1188
1189 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
1190 struct iovec *iov, unsigned int iov_cnt)
1191 {
1192 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1193 uint16_t vid;
1194 size_t s;
1195 NetClientState *nc = qemu_get_queue(n->nic);
1196
1197 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
1198 vid = virtio_lduw_p(vdev, &vid);
1199 if (s != sizeof(vid)) {
1200 return VIRTIO_NET_ERR;
1201 }
1202
1203 if (vid >= MAX_VLAN)
1204 return VIRTIO_NET_ERR;
1205
1206 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
1207 n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
1208 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
1209 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
1210 else
1211 return VIRTIO_NET_ERR;
1212
1213 rxfilter_notify(nc);
1214
1215 return VIRTIO_NET_OK;
1216 }
1217
1218 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd,
1219 struct iovec *iov, unsigned int iov_cnt)
1220 {
1221 trace_virtio_net_handle_announce(n->announce_timer.round);
1222 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK &&
1223 n->status & VIRTIO_NET_S_ANNOUNCE) {
1224 n->status &= ~VIRTIO_NET_S_ANNOUNCE;
1225 if (n->announce_timer.round) {
1226 qemu_announce_timer_step(&n->announce_timer);
1227 }
1228 return VIRTIO_NET_OK;
1229 } else {
1230 return VIRTIO_NET_ERR;
1231 }
1232 }
1233
1234 static void virtio_net_detach_epbf_rss(VirtIONet *n);
1235
1236 static void virtio_net_disable_rss(VirtIONet *n)
1237 {
1238 if (n->rss_data.enabled) {
1239 trace_virtio_net_rss_disable();
1240 }
1241 n->rss_data.enabled = false;
1242
1243 virtio_net_detach_epbf_rss(n);
1244 }
1245
1246 static bool virtio_net_attach_ebpf_to_backend(NICState *nic, int prog_fd)
1247 {
1248 NetClientState *nc = qemu_get_peer(qemu_get_queue(nic), 0);
1249 if (nc == NULL || nc->info->set_steering_ebpf == NULL) {
1250 return false;
1251 }
1252
1253 return nc->info->set_steering_ebpf(nc, prog_fd);
1254 }
1255
1256 static void rss_data_to_rss_config(struct VirtioNetRssData *data,
1257 struct EBPFRSSConfig *config)
1258 {
1259 config->redirect = data->redirect;
1260 config->populate_hash = data->populate_hash;
1261 config->hash_types = data->hash_types;
1262 config->indirections_len = data->indirections_len;
1263 config->default_queue = data->default_queue;
1264 }
1265
1266 static bool virtio_net_attach_epbf_rss(VirtIONet *n)
1267 {
1268 struct EBPFRSSConfig config = {};
1269
1270 if (!ebpf_rss_is_loaded(&n->ebpf_rss)) {
1271 return false;
1272 }
1273
1274 rss_data_to_rss_config(&n->rss_data, &config);
1275
1276 if (!ebpf_rss_set_all(&n->ebpf_rss, &config,
1277 n->rss_data.indirections_table, n->rss_data.key)) {
1278 return false;
1279 }
1280
1281 if (!virtio_net_attach_ebpf_to_backend(n->nic, n->ebpf_rss.program_fd)) {
1282 return false;
1283 }
1284
1285 return true;
1286 }
1287
1288 static void virtio_net_detach_epbf_rss(VirtIONet *n)
1289 {
1290 virtio_net_attach_ebpf_to_backend(n->nic, -1);
1291 }
1292
1293 static bool virtio_net_load_ebpf(VirtIONet *n)
1294 {
1295 if (!virtio_net_attach_ebpf_to_backend(n->nic, -1)) {
1296 /* backend does't support steering ebpf */
1297 return false;
1298 }
1299
1300 return ebpf_rss_load(&n->ebpf_rss);
1301 }
1302
1303 static void virtio_net_unload_ebpf(VirtIONet *n)
1304 {
1305 virtio_net_attach_ebpf_to_backend(n->nic, -1);
1306 ebpf_rss_unload(&n->ebpf_rss);
1307 }
1308
1309 static uint16_t virtio_net_handle_rss(VirtIONet *n,
1310 struct iovec *iov,
1311 unsigned int iov_cnt,
1312 bool do_rss)
1313 {
1314 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1315 struct virtio_net_rss_config cfg;
1316 size_t s, offset = 0, size_get;
1317 uint16_t queue_pairs, i;
1318 struct {
1319 uint16_t us;
1320 uint8_t b;
1321 } QEMU_PACKED temp;
1322 const char *err_msg = "";
1323 uint32_t err_value = 0;
1324
1325 if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) {
1326 err_msg = "RSS is not negotiated";
1327 goto error;
1328 }
1329 if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) {
1330 err_msg = "Hash report is not negotiated";
1331 goto error;
1332 }
1333 size_get = offsetof(struct virtio_net_rss_config, indirection_table);
1334 s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get);
1335 if (s != size_get) {
1336 err_msg = "Short command buffer";
1337 err_value = (uint32_t)s;
1338 goto error;
1339 }
1340 n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types);
1341 n->rss_data.indirections_len =
1342 virtio_lduw_p(vdev, &cfg.indirection_table_mask);
1343 n->rss_data.indirections_len++;
1344 if (!do_rss) {
1345 n->rss_data.indirections_len = 1;
1346 }
1347 if (!is_power_of_2(n->rss_data.indirections_len)) {
1348 err_msg = "Invalid size of indirection table";
1349 err_value = n->rss_data.indirections_len;
1350 goto error;
1351 }
1352 if (n->rss_data.indirections_len > VIRTIO_NET_RSS_MAX_TABLE_LEN) {
1353 err_msg = "Too large indirection table";
1354 err_value = n->rss_data.indirections_len;
1355 goto error;
1356 }
1357 n->rss_data.default_queue = do_rss ?
1358 virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0;
1359 if (n->rss_data.default_queue >= n->max_queue_pairs) {
1360 err_msg = "Invalid default queue";
1361 err_value = n->rss_data.default_queue;
1362 goto error;
1363 }
1364 offset += size_get;
1365 size_get = sizeof(uint16_t) * n->rss_data.indirections_len;
1366 g_free(n->rss_data.indirections_table);
1367 n->rss_data.indirections_table = g_malloc(size_get);
1368 if (!n->rss_data.indirections_table) {
1369 err_msg = "Can't allocate indirections table";
1370 err_value = n->rss_data.indirections_len;
1371 goto error;
1372 }
1373 s = iov_to_buf(iov, iov_cnt, offset,
1374 n->rss_data.indirections_table, size_get);
1375 if (s != size_get) {
1376 err_msg = "Short indirection table buffer";
1377 err_value = (uint32_t)s;
1378 goto error;
1379 }
1380 for (i = 0; i < n->rss_data.indirections_len; ++i) {
1381 uint16_t val = n->rss_data.indirections_table[i];
1382 n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val);
1383 }
1384 offset += size_get;
1385 size_get = sizeof(temp);
1386 s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get);
1387 if (s != size_get) {
1388 err_msg = "Can't get queue_pairs";
1389 err_value = (uint32_t)s;
1390 goto error;
1391 }
1392 queue_pairs = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queue_pairs;
1393 if (queue_pairs == 0 || queue_pairs > n->max_queue_pairs) {
1394 err_msg = "Invalid number of queue_pairs";
1395 err_value = queue_pairs;
1396 goto error;
1397 }
1398 if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) {
1399 err_msg = "Invalid key size";
1400 err_value = temp.b;
1401 goto error;
1402 }
1403 if (!temp.b && n->rss_data.hash_types) {
1404 err_msg = "No key provided";
1405 err_value = 0;
1406 goto error;
1407 }
1408 if (!temp.b && !n->rss_data.hash_types) {
1409 virtio_net_disable_rss(n);
1410 return queue_pairs;
1411 }
1412 offset += size_get;
1413 size_get = temp.b;
1414 s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get);
1415 if (s != size_get) {
1416 err_msg = "Can get key buffer";
1417 err_value = (uint32_t)s;
1418 goto error;
1419 }
1420 n->rss_data.enabled = true;
1421
1422 if (!n->rss_data.populate_hash) {
1423 if (!virtio_net_attach_epbf_rss(n)) {
1424 /* EBPF must be loaded for vhost */
1425 if (get_vhost_net(qemu_get_queue(n->nic)->peer)) {
1426 warn_report("Can't load eBPF RSS for vhost");
1427 goto error;
1428 }
1429 /* fallback to software RSS */
1430 warn_report("Can't load eBPF RSS - fallback to software RSS");
1431 n->rss_data.enabled_software_rss = true;
1432 }
1433 } else {
1434 /* use software RSS for hash populating */
1435 /* and detach eBPF if was loaded before */
1436 virtio_net_detach_epbf_rss(n);
1437 n->rss_data.enabled_software_rss = true;
1438 }
1439
1440 trace_virtio_net_rss_enable(n->rss_data.hash_types,
1441 n->rss_data.indirections_len,
1442 temp.b);
1443 return queue_pairs;
1444 error:
1445 trace_virtio_net_rss_error(err_msg, err_value);
1446 virtio_net_disable_rss(n);
1447 return 0;
1448 }
1449
1450 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
1451 struct iovec *iov, unsigned int iov_cnt)
1452 {
1453 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1454 uint16_t queue_pairs;
1455 NetClientState *nc = qemu_get_queue(n->nic);
1456
1457 virtio_net_disable_rss(n);
1458 if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) {
1459 queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, false);
1460 return queue_pairs ? VIRTIO_NET_OK : VIRTIO_NET_ERR;
1461 }
1462 if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
1463 queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, true);
1464 } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
1465 struct virtio_net_ctrl_mq mq;
1466 size_t s;
1467 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) {
1468 return VIRTIO_NET_ERR;
1469 }
1470 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
1471 if (s != sizeof(mq)) {
1472 return VIRTIO_NET_ERR;
1473 }
1474 queue_pairs = virtio_lduw_p(vdev, &mq.virtqueue_pairs);
1475
1476 } else {
1477 return VIRTIO_NET_ERR;
1478 }
1479
1480 if (queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1481 queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1482 queue_pairs > n->max_queue_pairs ||
1483 !n->multiqueue) {
1484 return VIRTIO_NET_ERR;
1485 }
1486
1487 n->curr_queue_pairs = queue_pairs;
1488 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
1489 /*
1490 * Avoid updating the backend for a vdpa device: We're only interested
1491 * in updating the device model queues.
1492 */
1493 return VIRTIO_NET_OK;
1494 }
1495 /* stop the backend before changing the number of queue_pairs to avoid handling a
1496 * disabled queue */
1497 virtio_net_set_status(vdev, vdev->status);
1498 virtio_net_set_queue_pairs(n);
1499
1500 return VIRTIO_NET_OK;
1501 }
1502
1503 size_t virtio_net_handle_ctrl_iov(VirtIODevice *vdev,
1504 const struct iovec *in_sg, unsigned in_num,
1505 const struct iovec *out_sg,
1506 unsigned out_num)
1507 {
1508 VirtIONet *n = VIRTIO_NET(vdev);
1509 struct virtio_net_ctrl_hdr ctrl;
1510 virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
1511 size_t s;
1512 struct iovec *iov, *iov2;
1513
1514 if (iov_size(in_sg, in_num) < sizeof(status) ||
1515 iov_size(out_sg, out_num) < sizeof(ctrl)) {
1516 virtio_error(vdev, "virtio-net ctrl missing headers");
1517 return 0;
1518 }
1519
1520 iov2 = iov = g_memdup2(out_sg, sizeof(struct iovec) * out_num);
1521 s = iov_to_buf(iov, out_num, 0, &ctrl, sizeof(ctrl));
1522 iov_discard_front(&iov, &out_num, sizeof(ctrl));
1523 if (s != sizeof(ctrl)) {
1524 status = VIRTIO_NET_ERR;
1525 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
1526 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, out_num);
1527 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
1528 status = virtio_net_handle_mac(n, ctrl.cmd, iov, out_num);
1529 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
1530 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, out_num);
1531 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) {
1532 status = virtio_net_handle_announce(n, ctrl.cmd, iov, out_num);
1533 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
1534 status = virtio_net_handle_mq(n, ctrl.cmd, iov, out_num);
1535 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
1536 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, out_num);
1537 }
1538
1539 s = iov_from_buf(in_sg, in_num, 0, &status, sizeof(status));
1540 assert(s == sizeof(status));
1541
1542 g_free(iov2);
1543 return sizeof(status);
1544 }
1545
1546 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
1547 {
1548 VirtQueueElement *elem;
1549
1550 for (;;) {
1551 size_t written;
1552 elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
1553 if (!elem) {
1554 break;
1555 }
1556
1557 written = virtio_net_handle_ctrl_iov(vdev, elem->in_sg, elem->in_num,
1558 elem->out_sg, elem->out_num);
1559 if (written > 0) {
1560 virtqueue_push(vq, elem, written);
1561 virtio_notify(vdev, vq);
1562 g_free(elem);
1563 } else {
1564 virtqueue_detach_element(vq, elem, 0);
1565 g_free(elem);
1566 break;
1567 }
1568 }
1569 }
1570
1571 /* RX */
1572
1573 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
1574 {
1575 VirtIONet *n = VIRTIO_NET(vdev);
1576 int queue_index = vq2q(virtio_get_queue_index(vq));
1577
1578 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
1579 }
1580
1581 static bool virtio_net_can_receive(NetClientState *nc)
1582 {
1583 VirtIONet *n = qemu_get_nic_opaque(nc);
1584 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1585 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1586
1587 if (!vdev->vm_running) {
1588 return false;
1589 }
1590
1591 if (nc->queue_index >= n->curr_queue_pairs) {
1592 return false;
1593 }
1594
1595 if (!virtio_queue_ready(q->rx_vq) ||
1596 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1597 return false;
1598 }
1599
1600 return true;
1601 }
1602
1603 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
1604 {
1605 VirtIONet *n = q->n;
1606 if (virtio_queue_empty(q->rx_vq) ||
1607 (n->mergeable_rx_bufs &&
1608 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1609 virtio_queue_set_notification(q->rx_vq, 1);
1610
1611 /* To avoid a race condition where the guest has made some buffers
1612 * available after the above check but before notification was
1613 * enabled, check for available buffers again.
1614 */
1615 if (virtio_queue_empty(q->rx_vq) ||
1616 (n->mergeable_rx_bufs &&
1617 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
1618 return 0;
1619 }
1620 }
1621
1622 virtio_queue_set_notification(q->rx_vq, 0);
1623 return 1;
1624 }
1625
1626 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr)
1627 {
1628 virtio_tswap16s(vdev, &hdr->hdr_len);
1629 virtio_tswap16s(vdev, &hdr->gso_size);
1630 virtio_tswap16s(vdev, &hdr->csum_start);
1631 virtio_tswap16s(vdev, &hdr->csum_offset);
1632 }
1633
1634 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
1635 * it never finds out that the packets don't have valid checksums. This
1636 * causes dhclient to get upset. Fedora's carried a patch for ages to
1637 * fix this with Xen but it hasn't appeared in an upstream release of
1638 * dhclient yet.
1639 *
1640 * To avoid breaking existing guests, we catch udp packets and add
1641 * checksums. This is terrible but it's better than hacking the guest
1642 * kernels.
1643 *
1644 * N.B. if we introduce a zero-copy API, this operation is no longer free so
1645 * we should provide a mechanism to disable it to avoid polluting the host
1646 * cache.
1647 */
1648 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
1649 uint8_t *buf, size_t size)
1650 {
1651 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
1652 (size > 27 && size < 1500) && /* normal sized MTU */
1653 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
1654 (buf[23] == 17) && /* ip.protocol == UDP */
1655 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
1656 net_checksum_calculate(buf, size, CSUM_UDP);
1657 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
1658 }
1659 }
1660
1661 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
1662 const void *buf, size_t size)
1663 {
1664 if (n->has_vnet_hdr) {
1665 /* FIXME this cast is evil */
1666 void *wbuf = (void *)buf;
1667 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
1668 size - n->host_hdr_len);
1669
1670 if (n->needs_vnet_hdr_swap) {
1671 virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf);
1672 }
1673 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
1674 } else {
1675 struct virtio_net_hdr hdr = {
1676 .flags = 0,
1677 .gso_type = VIRTIO_NET_HDR_GSO_NONE
1678 };
1679 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
1680 }
1681 }
1682
1683 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
1684 {
1685 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
1686 static const uint8_t vlan[] = {0x81, 0x00};
1687 uint8_t *ptr = (uint8_t *)buf;
1688 int i;
1689
1690 if (n->promisc)
1691 return 1;
1692
1693 ptr += n->host_hdr_len;
1694
1695 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
1696 int vid = lduw_be_p(ptr + 14) & 0xfff;
1697 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
1698 return 0;
1699 }
1700
1701 if (ptr[0] & 1) { // multicast
1702 if (!memcmp(ptr, bcast, sizeof(bcast))) {
1703 return !n->nobcast;
1704 } else if (n->nomulti) {
1705 return 0;
1706 } else if (n->allmulti || n->mac_table.multi_overflow) {
1707 return 1;
1708 }
1709
1710 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
1711 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1712 return 1;
1713 }
1714 }
1715 } else { // unicast
1716 if (n->nouni) {
1717 return 0;
1718 } else if (n->alluni || n->mac_table.uni_overflow) {
1719 return 1;
1720 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
1721 return 1;
1722 }
1723
1724 for (i = 0; i < n->mac_table.first_multi; i++) {
1725 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
1726 return 1;
1727 }
1728 }
1729 }
1730
1731 return 0;
1732 }
1733
1734 static uint8_t virtio_net_get_hash_type(bool isip4,
1735 bool isip6,
1736 bool isudp,
1737 bool istcp,
1738 uint32_t types)
1739 {
1740 if (isip4) {
1741 if (istcp && (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4)) {
1742 return NetPktRssIpV4Tcp;
1743 }
1744 if (isudp && (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4)) {
1745 return NetPktRssIpV4Udp;
1746 }
1747 if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
1748 return NetPktRssIpV4;
1749 }
1750 } else if (isip6) {
1751 uint32_t mask = VIRTIO_NET_RSS_HASH_TYPE_TCP_EX |
1752 VIRTIO_NET_RSS_HASH_TYPE_TCPv6;
1753
1754 if (istcp && (types & mask)) {
1755 return (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) ?
1756 NetPktRssIpV6TcpEx : NetPktRssIpV6Tcp;
1757 }
1758 mask = VIRTIO_NET_RSS_HASH_TYPE_UDP_EX | VIRTIO_NET_RSS_HASH_TYPE_UDPv6;
1759 if (isudp && (types & mask)) {
1760 return (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) ?
1761 NetPktRssIpV6UdpEx : NetPktRssIpV6Udp;
1762 }
1763 mask = VIRTIO_NET_RSS_HASH_TYPE_IP_EX | VIRTIO_NET_RSS_HASH_TYPE_IPv6;
1764 if (types & mask) {
1765 return (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) ?
1766 NetPktRssIpV6Ex : NetPktRssIpV6;
1767 }
1768 }
1769 return 0xff;
1770 }
1771
1772 static void virtio_set_packet_hash(const uint8_t *buf, uint8_t report,
1773 uint32_t hash)
1774 {
1775 struct virtio_net_hdr_v1_hash *hdr = (void *)buf;
1776 hdr->hash_value = hash;
1777 hdr->hash_report = report;
1778 }
1779
1780 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf,
1781 size_t size)
1782 {
1783 VirtIONet *n = qemu_get_nic_opaque(nc);
1784 unsigned int index = nc->queue_index, new_index = index;
1785 struct NetRxPkt *pkt = n->rx_pkt;
1786 uint8_t net_hash_type;
1787 uint32_t hash;
1788 bool isip4, isip6, isudp, istcp;
1789 static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = {
1790 VIRTIO_NET_HASH_REPORT_IPv4,
1791 VIRTIO_NET_HASH_REPORT_TCPv4,
1792 VIRTIO_NET_HASH_REPORT_TCPv6,
1793 VIRTIO_NET_HASH_REPORT_IPv6,
1794 VIRTIO_NET_HASH_REPORT_IPv6_EX,
1795 VIRTIO_NET_HASH_REPORT_TCPv6_EX,
1796 VIRTIO_NET_HASH_REPORT_UDPv4,
1797 VIRTIO_NET_HASH_REPORT_UDPv6,
1798 VIRTIO_NET_HASH_REPORT_UDPv6_EX
1799 };
1800
1801 net_rx_pkt_set_protocols(pkt, buf + n->host_hdr_len,
1802 size - n->host_hdr_len);
1803 net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
1804 if (isip4 && (net_rx_pkt_get_ip4_info(pkt)->fragment)) {
1805 istcp = isudp = false;
1806 }
1807 if (isip6 && (net_rx_pkt_get_ip6_info(pkt)->fragment)) {
1808 istcp = isudp = false;
1809 }
1810 net_hash_type = virtio_net_get_hash_type(isip4, isip6, isudp, istcp,
1811 n->rss_data.hash_types);
1812 if (net_hash_type > NetPktRssIpV6UdpEx) {
1813 if (n->rss_data.populate_hash) {
1814 virtio_set_packet_hash(buf, VIRTIO_NET_HASH_REPORT_NONE, 0);
1815 }
1816 return n->rss_data.redirect ? n->rss_data.default_queue : -1;
1817 }
1818
1819 hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key);
1820
1821 if (n->rss_data.populate_hash) {
1822 virtio_set_packet_hash(buf, reports[net_hash_type], hash);
1823 }
1824
1825 if (n->rss_data.redirect) {
1826 new_index = hash & (n->rss_data.indirections_len - 1);
1827 new_index = n->rss_data.indirections_table[new_index];
1828 }
1829
1830 return (index == new_index) ? -1 : new_index;
1831 }
1832
1833 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
1834 size_t size, bool no_rss)
1835 {
1836 VirtIONet *n = qemu_get_nic_opaque(nc);
1837 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
1838 VirtIODevice *vdev = VIRTIO_DEVICE(n);
1839 VirtQueueElement *elems[VIRTQUEUE_MAX_SIZE];
1840 size_t lens[VIRTQUEUE_MAX_SIZE];
1841 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
1842 struct virtio_net_hdr_mrg_rxbuf mhdr;
1843 unsigned mhdr_cnt = 0;
1844 size_t offset, i, guest_offset, j;
1845 ssize_t err;
1846
1847 if (!virtio_net_can_receive(nc)) {
1848 return -1;
1849 }
1850
1851 if (!no_rss && n->rss_data.enabled && n->rss_data.enabled_software_rss) {
1852 int index = virtio_net_process_rss(nc, buf, size);
1853 if (index >= 0) {
1854 NetClientState *nc2 = qemu_get_subqueue(n->nic, index);
1855 return virtio_net_receive_rcu(nc2, buf, size, true);
1856 }
1857 }
1858
1859 /* hdr_len refers to the header we supply to the guest */
1860 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
1861 return 0;
1862 }
1863
1864 if (!receive_filter(n, buf, size))
1865 return size;
1866
1867 offset = i = 0;
1868
1869 while (offset < size) {
1870 VirtQueueElement *elem;
1871 int len, total;
1872 const struct iovec *sg;
1873
1874 total = 0;
1875
1876 if (i == VIRTQUEUE_MAX_SIZE) {
1877 virtio_error(vdev, "virtio-net unexpected long buffer chain");
1878 err = size;
1879 goto err;
1880 }
1881
1882 elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement));
1883 if (!elem) {
1884 if (i) {
1885 virtio_error(vdev, "virtio-net unexpected empty queue: "
1886 "i %zd mergeable %d offset %zd, size %zd, "
1887 "guest hdr len %zd, host hdr len %zd "
1888 "guest features 0x%" PRIx64,
1889 i, n->mergeable_rx_bufs, offset, size,
1890 n->guest_hdr_len, n->host_hdr_len,
1891 vdev->guest_features);
1892 }
1893 err = -1;
1894 goto err;
1895 }
1896
1897 if (elem->in_num < 1) {
1898 virtio_error(vdev,
1899 "virtio-net receive queue contains no in buffers");
1900 virtqueue_detach_element(q->rx_vq, elem, 0);
1901 g_free(elem);
1902 err = -1;
1903 goto err;
1904 }
1905
1906 sg = elem->in_sg;
1907 if (i == 0) {
1908 assert(offset == 0);
1909 if (n->mergeable_rx_bufs) {
1910 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
1911 sg, elem->in_num,
1912 offsetof(typeof(mhdr), num_buffers),
1913 sizeof(mhdr.num_buffers));
1914 }
1915
1916 receive_header(n, sg, elem->in_num, buf, size);
1917 if (n->rss_data.populate_hash) {
1918 offset = sizeof(mhdr);
1919 iov_from_buf(sg, elem->in_num, offset,
1920 buf + offset, n->host_hdr_len - sizeof(mhdr));
1921 }
1922 offset = n->host_hdr_len;
1923 total += n->guest_hdr_len;
1924 guest_offset = n->guest_hdr_len;
1925 } else {
1926 guest_offset = 0;
1927 }
1928
1929 /* copy in packet. ugh */
1930 len = iov_from_buf(sg, elem->in_num, guest_offset,
1931 buf + offset, size - offset);
1932 total += len;
1933 offset += len;
1934 /* If buffers can't be merged, at this point we
1935 * must have consumed the complete packet.
1936 * Otherwise, drop it. */
1937 if (!n->mergeable_rx_bufs && offset < size) {
1938 virtqueue_unpop(q->rx_vq, elem, total);
1939 g_free(elem);
1940 err = size;
1941 goto err;
1942 }
1943
1944 elems[i] = elem;
1945 lens[i] = total;
1946 i++;
1947 }
1948
1949 if (mhdr_cnt) {
1950 virtio_stw_p(vdev, &mhdr.num_buffers, i);
1951 iov_from_buf(mhdr_sg, mhdr_cnt,
1952 0,
1953 &mhdr.num_buffers, sizeof mhdr.num_buffers);
1954 }
1955
1956 for (j = 0; j < i; j++) {
1957 /* signal other side */
1958 virtqueue_fill(q->rx_vq, elems[j], lens[j], j);
1959 g_free(elems[j]);
1960 }
1961
1962 virtqueue_flush(q->rx_vq, i);
1963 virtio_notify(vdev, q->rx_vq);
1964
1965 return size;
1966
1967 err:
1968 for (j = 0; j < i; j++) {
1969 virtqueue_detach_element(q->rx_vq, elems[j], lens[j]);
1970 g_free(elems[j]);
1971 }
1972
1973 return err;
1974 }
1975
1976 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf,
1977 size_t size)
1978 {
1979 RCU_READ_LOCK_GUARD();
1980
1981 return virtio_net_receive_rcu(nc, buf, size, false);
1982 }
1983
1984 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain,
1985 const uint8_t *buf,
1986 VirtioNetRscUnit *unit)
1987 {
1988 uint16_t ip_hdrlen;
1989 struct ip_header *ip;
1990
1991 ip = (struct ip_header *)(buf + chain->n->guest_hdr_len
1992 + sizeof(struct eth_header));
1993 unit->ip = (void *)ip;
1994 ip_hdrlen = (ip->ip_ver_len & 0xF) << 2;
1995 unit->ip_plen = &ip->ip_len;
1996 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen);
1997 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
1998 unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen;
1999 }
2000
2001 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain,
2002 const uint8_t *buf,
2003 VirtioNetRscUnit *unit)
2004 {
2005 struct ip6_header *ip6;
2006
2007 ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len
2008 + sizeof(struct eth_header));
2009 unit->ip = ip6;
2010 unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
2011 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip)
2012 + sizeof(struct ip6_header));
2013 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10;
2014
2015 /* There is a difference between payload lenght in ipv4 and v6,
2016 ip header is excluded in ipv6 */
2017 unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen;
2018 }
2019
2020 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain,
2021 VirtioNetRscSeg *seg)
2022 {
2023 int ret;
2024 struct virtio_net_hdr_v1 *h;
2025
2026 h = (struct virtio_net_hdr_v1 *)seg->buf;
2027 h->flags = 0;
2028 h->gso_type = VIRTIO_NET_HDR_GSO_NONE;
2029
2030 if (seg->is_coalesced) {
2031 h->rsc.segments = seg->packets;
2032 h->rsc.dup_acks = seg->dup_ack;
2033 h->flags = VIRTIO_NET_HDR_F_RSC_INFO;
2034 if (chain->proto == ETH_P_IP) {
2035 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2036 } else {
2037 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2038 }
2039 }
2040
2041 ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size);
2042 QTAILQ_REMOVE(&chain->buffers, seg, next);
2043 g_free(seg->buf);
2044 g_free(seg);
2045
2046 return ret;
2047 }
2048
2049 static void virtio_net_rsc_purge(void *opq)
2050 {
2051 VirtioNetRscSeg *seg, *rn;
2052 VirtioNetRscChain *chain = (VirtioNetRscChain *)opq;
2053
2054 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) {
2055 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2056 chain->stat.purge_failed++;
2057 continue;
2058 }
2059 }
2060
2061 chain->stat.timer++;
2062 if (!QTAILQ_EMPTY(&chain->buffers)) {
2063 timer_mod(chain->drain_timer,
2064 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
2065 }
2066 }
2067
2068 static void virtio_net_rsc_cleanup(VirtIONet *n)
2069 {
2070 VirtioNetRscChain *chain, *rn_chain;
2071 VirtioNetRscSeg *seg, *rn_seg;
2072
2073 QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) {
2074 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) {
2075 QTAILQ_REMOVE(&chain->buffers, seg, next);
2076 g_free(seg->buf);
2077 g_free(seg);
2078 }
2079
2080 timer_free(chain->drain_timer);
2081 QTAILQ_REMOVE(&n->rsc_chains, chain, next);
2082 g_free(chain);
2083 }
2084 }
2085
2086 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain,
2087 NetClientState *nc,
2088 const uint8_t *buf, size_t size)
2089 {
2090 uint16_t hdr_len;
2091 VirtioNetRscSeg *seg;
2092
2093 hdr_len = chain->n->guest_hdr_len;
2094 seg = g_new(VirtioNetRscSeg, 1);
2095 seg->buf = g_malloc(hdr_len + sizeof(struct eth_header)
2096 + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD);
2097 memcpy(seg->buf, buf, size);
2098 seg->size = size;
2099 seg->packets = 1;
2100 seg->dup_ack = 0;
2101 seg->is_coalesced = 0;
2102 seg->nc = nc;
2103
2104 QTAILQ_INSERT_TAIL(&chain->buffers, seg, next);
2105 chain->stat.cache++;
2106
2107 switch (chain->proto) {
2108 case ETH_P_IP:
2109 virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit);
2110 break;
2111 case ETH_P_IPV6:
2112 virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit);
2113 break;
2114 default:
2115 g_assert_not_reached();
2116 }
2117 }
2118
2119 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain,
2120 VirtioNetRscSeg *seg,
2121 const uint8_t *buf,
2122 struct tcp_header *n_tcp,
2123 struct tcp_header *o_tcp)
2124 {
2125 uint32_t nack, oack;
2126 uint16_t nwin, owin;
2127
2128 nack = htonl(n_tcp->th_ack);
2129 nwin = htons(n_tcp->th_win);
2130 oack = htonl(o_tcp->th_ack);
2131 owin = htons(o_tcp->th_win);
2132
2133 if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) {
2134 chain->stat.ack_out_of_win++;
2135 return RSC_FINAL;
2136 } else if (nack == oack) {
2137 /* duplicated ack or window probe */
2138 if (nwin == owin) {
2139 /* duplicated ack, add dup ack count due to whql test up to 1 */
2140 chain->stat.dup_ack++;
2141 return RSC_FINAL;
2142 } else {
2143 /* Coalesce window update */
2144 o_tcp->th_win = n_tcp->th_win;
2145 chain->stat.win_update++;
2146 return RSC_COALESCE;
2147 }
2148 } else {
2149 /* pure ack, go to 'C', finalize*/
2150 chain->stat.pure_ack++;
2151 return RSC_FINAL;
2152 }
2153 }
2154
2155 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain,
2156 VirtioNetRscSeg *seg,
2157 const uint8_t *buf,
2158 VirtioNetRscUnit *n_unit)
2159 {
2160 void *data;
2161 uint16_t o_ip_len;
2162 uint32_t nseq, oseq;
2163 VirtioNetRscUnit *o_unit;
2164
2165 o_unit = &seg->unit;
2166 o_ip_len = htons(*o_unit->ip_plen);
2167 nseq = htonl(n_unit->tcp->th_seq);
2168 oseq = htonl(o_unit->tcp->th_seq);
2169
2170 /* out of order or retransmitted. */
2171 if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) {
2172 chain->stat.data_out_of_win++;
2173 return RSC_FINAL;
2174 }
2175
2176 data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen;
2177 if (nseq == oseq) {
2178 if ((o_unit->payload == 0) && n_unit->payload) {
2179 /* From no payload to payload, normal case, not a dup ack or etc */
2180 chain->stat.data_after_pure_ack++;
2181 goto coalesce;
2182 } else {
2183 return virtio_net_rsc_handle_ack(chain, seg, buf,
2184 n_unit->tcp, o_unit->tcp);
2185 }
2186 } else if ((nseq - oseq) != o_unit->payload) {
2187 /* Not a consistent packet, out of order */
2188 chain->stat.data_out_of_order++;
2189 return RSC_FINAL;
2190 } else {
2191 coalesce:
2192 if ((o_ip_len + n_unit->payload) > chain->max_payload) {
2193 chain->stat.over_size++;
2194 return RSC_FINAL;
2195 }
2196
2197 /* Here comes the right data, the payload length in v4/v6 is different,
2198 so use the field value to update and record the new data len */
2199 o_unit->payload += n_unit->payload; /* update new data len */
2200
2201 /* update field in ip header */
2202 *o_unit->ip_plen = htons(o_ip_len + n_unit->payload);
2203
2204 /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced
2205 for windows guest, while this may change the behavior for linux
2206 guest (only if it uses RSC feature). */
2207 o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags;
2208
2209 o_unit->tcp->th_ack = n_unit->tcp->th_ack;
2210 o_unit->tcp->th_win = n_unit->tcp->th_win;
2211
2212 memmove(seg->buf + seg->size, data, n_unit->payload);
2213 seg->size += n_unit->payload;
2214 seg->packets++;
2215 chain->stat.coalesced++;
2216 return RSC_COALESCE;
2217 }
2218 }
2219
2220 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain,
2221 VirtioNetRscSeg *seg,
2222 const uint8_t *buf, size_t size,
2223 VirtioNetRscUnit *unit)
2224 {
2225 struct ip_header *ip1, *ip2;
2226
2227 ip1 = (struct ip_header *)(unit->ip);
2228 ip2 = (struct ip_header *)(seg->unit.ip);
2229 if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst)
2230 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2231 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2232 chain->stat.no_match++;
2233 return RSC_NO_MATCH;
2234 }
2235
2236 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2237 }
2238
2239 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain,
2240 VirtioNetRscSeg *seg,
2241 const uint8_t *buf, size_t size,
2242 VirtioNetRscUnit *unit)
2243 {
2244 struct ip6_header *ip1, *ip2;
2245
2246 ip1 = (struct ip6_header *)(unit->ip);
2247 ip2 = (struct ip6_header *)(seg->unit.ip);
2248 if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address))
2249 || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address))
2250 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport)
2251 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) {
2252 chain->stat.no_match++;
2253 return RSC_NO_MATCH;
2254 }
2255
2256 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit);
2257 }
2258
2259 /* Packets with 'SYN' should bypass, other flag should be sent after drain
2260 * to prevent out of order */
2261 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain,
2262 struct tcp_header *tcp)
2263 {
2264 uint16_t tcp_hdr;
2265 uint16_t tcp_flag;
2266
2267 tcp_flag = htons(tcp->th_offset_flags);
2268 tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10;
2269 tcp_flag &= VIRTIO_NET_TCP_FLAG;
2270 if (tcp_flag & TH_SYN) {
2271 chain->stat.tcp_syn++;
2272 return RSC_BYPASS;
2273 }
2274
2275 if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) {
2276 chain->stat.tcp_ctrl_drain++;
2277 return RSC_FINAL;
2278 }
2279
2280 if (tcp_hdr > sizeof(struct tcp_header)) {
2281 chain->stat.tcp_all_opt++;
2282 return RSC_FINAL;
2283 }
2284
2285 return RSC_CANDIDATE;
2286 }
2287
2288 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain,
2289 NetClientState *nc,
2290 const uint8_t *buf, size_t size,
2291 VirtioNetRscUnit *unit)
2292 {
2293 int ret;
2294 VirtioNetRscSeg *seg, *nseg;
2295
2296 if (QTAILQ_EMPTY(&chain->buffers)) {
2297 chain->stat.empty_cache++;
2298 virtio_net_rsc_cache_buf(chain, nc, buf, size);
2299 timer_mod(chain->drain_timer,
2300 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout);
2301 return size;
2302 }
2303
2304 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2305 if (chain->proto == ETH_P_IP) {
2306 ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit);
2307 } else {
2308 ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit);
2309 }
2310
2311 if (ret == RSC_FINAL) {
2312 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2313 /* Send failed */
2314 chain->stat.final_failed++;
2315 return 0;
2316 }
2317
2318 /* Send current packet */
2319 return virtio_net_do_receive(nc, buf, size);
2320 } else if (ret == RSC_NO_MATCH) {
2321 continue;
2322 } else {
2323 /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */
2324 seg->is_coalesced = 1;
2325 return size;
2326 }
2327 }
2328
2329 chain->stat.no_match_cache++;
2330 virtio_net_rsc_cache_buf(chain, nc, buf, size);
2331 return size;
2332 }
2333
2334 /* Drain a connection data, this is to avoid out of order segments */
2335 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain,
2336 NetClientState *nc,
2337 const uint8_t *buf, size_t size,
2338 uint16_t ip_start, uint16_t ip_size,
2339 uint16_t tcp_port)
2340 {
2341 VirtioNetRscSeg *seg, *nseg;
2342 uint32_t ppair1, ppair2;
2343
2344 ppair1 = *(uint32_t *)(buf + tcp_port);
2345 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) {
2346 ppair2 = *(uint32_t *)(seg->buf + tcp_port);
2347 if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size)
2348 || (ppair1 != ppair2)) {
2349 continue;
2350 }
2351 if (virtio_net_rsc_drain_seg(chain, seg) == 0) {
2352 chain->stat.drain_failed++;
2353 }
2354
2355 break;
2356 }
2357
2358 return virtio_net_do_receive(nc, buf, size);
2359 }
2360
2361 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain,
2362 struct ip_header *ip,
2363 const uint8_t *buf, size_t size)
2364 {
2365 uint16_t ip_len;
2366
2367 /* Not an ipv4 packet */
2368 if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) {
2369 chain->stat.ip_option++;
2370 return RSC_BYPASS;
2371 }
2372
2373 /* Don't handle packets with ip option */
2374 if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) {
2375 chain->stat.ip_option++;
2376 return RSC_BYPASS;
2377 }
2378
2379 if (ip->ip_p != IPPROTO_TCP) {
2380 chain->stat.bypass_not_tcp++;
2381 return RSC_BYPASS;
2382 }
2383
2384 /* Don't handle packets with ip fragment */
2385 if (!(htons(ip->ip_off) & IP_DF)) {
2386 chain->stat.ip_frag++;
2387 return RSC_BYPASS;
2388 }
2389
2390 /* Don't handle packets with ecn flag */
2391 if (IPTOS_ECN(ip->ip_tos)) {
2392 chain->stat.ip_ecn++;
2393 return RSC_BYPASS;
2394 }
2395
2396 ip_len = htons(ip->ip_len);
2397 if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header))
2398 || ip_len > (size - chain->n->guest_hdr_len -
2399 sizeof(struct eth_header))) {
2400 chain->stat.ip_hacked++;
2401 return RSC_BYPASS;
2402 }
2403
2404 return RSC_CANDIDATE;
2405 }
2406
2407 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain,
2408 NetClientState *nc,
2409 const uint8_t *buf, size_t size)
2410 {
2411 int32_t ret;
2412 uint16_t hdr_len;
2413 VirtioNetRscUnit unit;
2414
2415 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2416
2417 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)
2418 + sizeof(struct tcp_header))) {
2419 chain->stat.bypass_not_tcp++;
2420 return virtio_net_do_receive(nc, buf, size);
2421 }
2422
2423 virtio_net_rsc_extract_unit4(chain, buf, &unit);
2424 if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size)
2425 != RSC_CANDIDATE) {
2426 return virtio_net_do_receive(nc, buf, size);
2427 }
2428
2429 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2430 if (ret == RSC_BYPASS) {
2431 return virtio_net_do_receive(nc, buf, size);
2432 } else if (ret == RSC_FINAL) {
2433 return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2434 ((hdr_len + sizeof(struct eth_header)) + 12),
2435 VIRTIO_NET_IP4_ADDR_SIZE,
2436 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header));
2437 }
2438
2439 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2440 }
2441
2442 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain,
2443 struct ip6_header *ip6,
2444 const uint8_t *buf, size_t size)
2445 {
2446 uint16_t ip_len;
2447
2448 if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4)
2449 != IP_HEADER_VERSION_6) {
2450 return RSC_BYPASS;
2451 }
2452
2453 /* Both option and protocol is checked in this */
2454 if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) {
2455 chain->stat.bypass_not_tcp++;
2456 return RSC_BYPASS;
2457 }
2458
2459 ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen);
2460 if (ip_len < sizeof(struct tcp_header) ||
2461 ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header)
2462 - sizeof(struct ip6_header))) {
2463 chain->stat.ip_hacked++;
2464 return RSC_BYPASS;
2465 }
2466
2467 /* Don't handle packets with ecn flag */
2468 if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) {
2469 chain->stat.ip_ecn++;
2470 return RSC_BYPASS;
2471 }
2472
2473 return RSC_CANDIDATE;
2474 }
2475
2476 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc,
2477 const uint8_t *buf, size_t size)
2478 {
2479 int32_t ret;
2480 uint16_t hdr_len;
2481 VirtioNetRscChain *chain;
2482 VirtioNetRscUnit unit;
2483
2484 chain = opq;
2485 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len;
2486
2487 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header)
2488 + sizeof(tcp_header))) {
2489 return virtio_net_do_receive(nc, buf, size);
2490 }
2491
2492 virtio_net_rsc_extract_unit6(chain, buf, &unit);
2493 if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain,
2494 unit.ip, buf, size)) {
2495 return virtio_net_do_receive(nc, buf, size);
2496 }
2497
2498 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp);
2499 if (ret == RSC_BYPASS) {
2500 return virtio_net_do_receive(nc, buf, size);
2501 } else if (ret == RSC_FINAL) {
2502 return virtio_net_rsc_drain_flow(chain, nc, buf, size,
2503 ((hdr_len + sizeof(struct eth_header)) + 8),
2504 VIRTIO_NET_IP6_ADDR_SIZE,
2505 hdr_len + sizeof(struct eth_header)
2506 + sizeof(struct ip6_header));
2507 }
2508
2509 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit);
2510 }
2511
2512 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n,
2513 NetClientState *nc,
2514 uint16_t proto)
2515 {
2516 VirtioNetRscChain *chain;
2517
2518 if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) {
2519 return NULL;
2520 }
2521
2522 QTAILQ_FOREACH(chain, &n->rsc_chains, next) {
2523 if (chain->proto == proto) {
2524 return chain;
2525 }
2526 }
2527
2528 chain = g_malloc(sizeof(*chain));
2529 chain->n = n;
2530 chain->proto = proto;
2531 if (proto == (uint16_t)ETH_P_IP) {
2532 chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD;
2533 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2534 } else {
2535 chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD;
2536 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2537 }
2538 chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST,
2539 virtio_net_rsc_purge, chain);
2540 memset(&chain->stat, 0, sizeof(chain->stat));
2541
2542 QTAILQ_INIT(&chain->buffers);
2543 QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next);
2544
2545 return chain;
2546 }
2547
2548 static ssize_t virtio_net_rsc_receive(NetClientState *nc,
2549 const uint8_t *buf,
2550 size_t size)
2551 {
2552 uint16_t proto;
2553 VirtioNetRscChain *chain;
2554 struct eth_header *eth;
2555 VirtIONet *n;
2556
2557 n = qemu_get_nic_opaque(nc);
2558 if (size < (n->host_hdr_len + sizeof(struct eth_header))) {
2559 return virtio_net_do_receive(nc, buf, size);
2560 }
2561
2562 eth = (struct eth_header *)(buf + n->guest_hdr_len);
2563 proto = htons(eth->h_proto);
2564
2565 chain = virtio_net_rsc_lookup_chain(n, nc, proto);
2566 if (chain) {
2567 chain->stat.received++;
2568 if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) {
2569 return virtio_net_rsc_receive4(chain, nc, buf, size);
2570 } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) {
2571 return virtio_net_rsc_receive6(chain, nc, buf, size);
2572 }
2573 }
2574 return virtio_net_do_receive(nc, buf, size);
2575 }
2576
2577 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf,
2578 size_t size)
2579 {
2580 VirtIONet *n = qemu_get_nic_opaque(nc);
2581 if ((n->rsc4_enabled || n->rsc6_enabled)) {
2582 return virtio_net_rsc_receive(nc, buf, size);
2583 } else {
2584 return virtio_net_do_receive(nc, buf, size);
2585 }
2586 }
2587
2588 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
2589
2590 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
2591 {
2592 VirtIONet *n = qemu_get_nic_opaque(nc);
2593 VirtIONetQueue *q = virtio_net_get_subqueue(nc);
2594 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2595 int ret;
2596
2597 virtqueue_push(q->tx_vq, q->async_tx.elem, 0);
2598 virtio_notify(vdev, q->tx_vq);
2599
2600 g_free(q->async_tx.elem);
2601 q->async_tx.elem = NULL;
2602
2603 virtio_queue_set_notification(q->tx_vq, 1);
2604 ret = virtio_net_flush_tx(q);
2605 if (ret >= n->tx_burst) {
2606 /*
2607 * the flush has been stopped by tx_burst
2608 * we will not receive notification for the
2609 * remainining part, so re-schedule
2610 */
2611 virtio_queue_set_notification(q->tx_vq, 0);
2612 if (q->tx_bh) {
2613 qemu_bh_schedule(q->tx_bh);
2614 } else {
2615 timer_mod(q->tx_timer,
2616 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2617 }
2618 q->tx_waiting = 1;
2619 }
2620 }
2621
2622 /* TX */
2623 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
2624 {
2625 VirtIONet *n = q->n;
2626 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2627 VirtQueueElement *elem;
2628 int32_t num_packets = 0;
2629 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
2630 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2631 return num_packets;
2632 }
2633
2634 if (q->async_tx.elem) {
2635 virtio_queue_set_notification(q->tx_vq, 0);
2636 return num_packets;
2637 }
2638
2639 for (;;) {
2640 ssize_t ret;
2641 unsigned int out_num;
2642 struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg;
2643 struct virtio_net_hdr_mrg_rxbuf mhdr;
2644
2645 elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement));
2646 if (!elem) {
2647 break;
2648 }
2649
2650 out_num = elem->out_num;
2651 out_sg = elem->out_sg;
2652 if (out_num < 1) {
2653 virtio_error(vdev, "virtio-net header not in first element");
2654 virtqueue_detach_element(q->tx_vq, elem, 0);
2655 g_free(elem);
2656 return -EINVAL;
2657 }
2658
2659 if (n->has_vnet_hdr) {
2660 if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) <
2661 n->guest_hdr_len) {
2662 virtio_error(vdev, "virtio-net header incorrect");
2663 virtqueue_detach_element(q->tx_vq, elem, 0);
2664 g_free(elem);
2665 return -EINVAL;
2666 }
2667 if (n->needs_vnet_hdr_swap) {
2668 virtio_net_hdr_swap(vdev, (void *) &mhdr);
2669 sg2[0].iov_base = &mhdr;
2670 sg2[0].iov_len = n->guest_hdr_len;
2671 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
2672 out_sg, out_num,
2673 n->guest_hdr_len, -1);
2674 if (out_num == VIRTQUEUE_MAX_SIZE) {
2675 goto drop;
2676 }
2677 out_num += 1;
2678 out_sg = sg2;
2679 }
2680 }
2681 /*
2682 * If host wants to see the guest header as is, we can
2683 * pass it on unchanged. Otherwise, copy just the parts
2684 * that host is interested in.
2685 */
2686 assert(n->host_hdr_len <= n->guest_hdr_len);
2687 if (n->host_hdr_len != n->guest_hdr_len) {
2688 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
2689 out_sg, out_num,
2690 0, n->host_hdr_len);
2691 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
2692 out_sg, out_num,
2693 n->guest_hdr_len, -1);
2694 out_num = sg_num;
2695 out_sg = sg;
2696 }
2697
2698 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
2699 out_sg, out_num, virtio_net_tx_complete);
2700 if (ret == 0) {
2701 virtio_queue_set_notification(q->tx_vq, 0);
2702 q->async_tx.elem = elem;
2703 return -EBUSY;
2704 }
2705
2706 drop:
2707 virtqueue_push(q->tx_vq, elem, 0);
2708 virtio_notify(vdev, q->tx_vq);
2709 g_free(elem);
2710
2711 if (++num_packets >= n->tx_burst) {
2712 break;
2713 }
2714 }
2715 return num_packets;
2716 }
2717
2718 static void virtio_net_tx_timer(void *opaque);
2719
2720 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
2721 {
2722 VirtIONet *n = VIRTIO_NET(vdev);
2723 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2724
2725 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2726 virtio_net_drop_tx_queue_data(vdev, vq);
2727 return;
2728 }
2729
2730 /* This happens when device was stopped but VCPU wasn't. */
2731 if (!vdev->vm_running) {
2732 q->tx_waiting = 1;
2733 return;
2734 }
2735
2736 if (q->tx_waiting) {
2737 /* We already have queued packets, immediately flush */
2738 timer_del(q->tx_timer);
2739 virtio_net_tx_timer(q);
2740 } else {
2741 /* re-arm timer to flush it (and more) on next tick */
2742 timer_mod(q->tx_timer,
2743 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2744 q->tx_waiting = 1;
2745 virtio_queue_set_notification(vq, 0);
2746 }
2747 }
2748
2749 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
2750 {
2751 VirtIONet *n = VIRTIO_NET(vdev);
2752 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
2753
2754 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
2755 virtio_net_drop_tx_queue_data(vdev, vq);
2756 return;
2757 }
2758
2759 if (unlikely(q->tx_waiting)) {
2760 return;
2761 }
2762 q->tx_waiting = 1;
2763 /* This happens when device was stopped but VCPU wasn't. */
2764 if (!vdev->vm_running) {
2765 return;
2766 }
2767 virtio_queue_set_notification(vq, 0);
2768 qemu_bh_schedule(q->tx_bh);
2769 }
2770
2771 static void virtio_net_tx_timer(void *opaque)
2772 {
2773 VirtIONetQueue *q = opaque;
2774 VirtIONet *n = q->n;
2775 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2776 int ret;
2777
2778 /* This happens when device was stopped but BH wasn't. */
2779 if (!vdev->vm_running) {
2780 /* Make sure tx waiting is set, so we'll run when restarted. */
2781 assert(q->tx_waiting);
2782 return;
2783 }
2784
2785 q->tx_waiting = 0;
2786
2787 /* Just in case the driver is not ready on more */
2788 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
2789 return;
2790 }
2791
2792 ret = virtio_net_flush_tx(q);
2793 if (ret == -EBUSY || ret == -EINVAL) {
2794 return;
2795 }
2796 /*
2797 * If we flush a full burst of packets, assume there are
2798 * more coming and immediately rearm
2799 */
2800 if (ret >= n->tx_burst) {
2801 q->tx_waiting = 1;
2802 timer_mod(q->tx_timer,
2803 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2804 return;
2805 }
2806 /*
2807 * If less than a full burst, re-enable notification and flush
2808 * anything that may have come in while we weren't looking. If
2809 * we find something, assume the guest is still active and rearm
2810 */
2811 virtio_queue_set_notification(q->tx_vq, 1);
2812 ret = virtio_net_flush_tx(q);
2813 if (ret > 0) {
2814 virtio_queue_set_notification(q->tx_vq, 0);
2815 q->tx_waiting = 1;
2816 timer_mod(q->tx_timer,
2817 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout);
2818 }
2819 }
2820
2821 static void virtio_net_tx_bh(void *opaque)
2822 {
2823 VirtIONetQueue *q = opaque;
2824 VirtIONet *n = q->n;
2825 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2826 int32_t ret;
2827
2828 /* This happens when device was stopped but BH wasn't. */
2829 if (!vdev->vm_running) {
2830 /* Make sure tx waiting is set, so we'll run when restarted. */
2831 assert(q->tx_waiting);
2832 return;
2833 }
2834
2835 q->tx_waiting = 0;
2836
2837 /* Just in case the driver is not ready on more */
2838 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
2839 return;
2840 }
2841
2842 ret = virtio_net_flush_tx(q);
2843 if (ret == -EBUSY || ret == -EINVAL) {
2844 return; /* Notification re-enable handled by tx_complete or device
2845 * broken */
2846 }
2847
2848 /* If we flush a full burst of packets, assume there are
2849 * more coming and immediately reschedule */
2850 if (ret >= n->tx_burst) {
2851 qemu_bh_schedule(q->tx_bh);
2852 q->tx_waiting = 1;
2853 return;
2854 }
2855
2856 /* If less than a full burst, re-enable notification and flush
2857 * anything that may have come in while we weren't looking. If
2858 * we find something, assume the guest is still active and reschedule */
2859 virtio_queue_set_notification(q->tx_vq, 1);
2860 ret = virtio_net_flush_tx(q);
2861 if (ret == -EINVAL) {
2862 return;
2863 } else if (ret > 0) {
2864 virtio_queue_set_notification(q->tx_vq, 0);
2865 qemu_bh_schedule(q->tx_bh);
2866 q->tx_waiting = 1;
2867 }
2868 }
2869
2870 static void virtio_net_add_queue(VirtIONet *n, int index)
2871 {
2872 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2873
2874 n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
2875 virtio_net_handle_rx);
2876
2877 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
2878 n->vqs[index].tx_vq =
2879 virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2880 virtio_net_handle_tx_timer);
2881 n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
2882 virtio_net_tx_timer,
2883 &n->vqs[index]);
2884 } else {
2885 n->vqs[index].tx_vq =
2886 virtio_add_queue(vdev, n->net_conf.tx_queue_size,
2887 virtio_net_handle_tx_bh);
2888 n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
2889 }
2890
2891 n->vqs[index].tx_waiting = 0;
2892 n->vqs[index].n = n;
2893 }
2894
2895 static void virtio_net_del_queue(VirtIONet *n, int index)
2896 {
2897 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2898 VirtIONetQueue *q = &n->vqs[index];
2899 NetClientState *nc = qemu_get_subqueue(n->nic, index);
2900
2901 qemu_purge_queued_packets(nc);
2902
2903 virtio_del_queue(vdev, index * 2);
2904 if (q->tx_timer) {
2905 timer_free(q->tx_timer);
2906 q->tx_timer = NULL;
2907 } else {
2908 qemu_bh_delete(q->tx_bh);
2909 q->tx_bh = NULL;
2910 }
2911 q->tx_waiting = 0;
2912 virtio_del_queue(vdev, index * 2 + 1);
2913 }
2914
2915 static void virtio_net_change_num_queue_pairs(VirtIONet *n, int new_max_queue_pairs)
2916 {
2917 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2918 int old_num_queues = virtio_get_num_queues(vdev);
2919 int new_num_queues = new_max_queue_pairs * 2 + 1;
2920 int i;
2921
2922 assert(old_num_queues >= 3);
2923 assert(old_num_queues % 2 == 1);
2924
2925 if (old_num_queues == new_num_queues) {
2926 return;
2927 }
2928
2929 /*
2930 * We always need to remove and add ctrl vq if
2931 * old_num_queues != new_num_queues. Remove ctrl_vq first,
2932 * and then we only enter one of the following two loops.
2933 */
2934 virtio_del_queue(vdev, old_num_queues - 1);
2935
2936 for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) {
2937 /* new_num_queues < old_num_queues */
2938 virtio_net_del_queue(n, i / 2);
2939 }
2940
2941 for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) {
2942 /* new_num_queues > old_num_queues */
2943 virtio_net_add_queue(n, i / 2);
2944 }
2945
2946 /* add ctrl_vq last */
2947 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
2948 }
2949
2950 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
2951 {
2952 int max = multiqueue ? n->max_queue_pairs : 1;
2953
2954 n->multiqueue = multiqueue;
2955 virtio_net_change_num_queue_pairs(n, max);
2956
2957 virtio_net_set_queue_pairs(n);
2958 }
2959
2960 static int virtio_net_post_load_device(void *opaque, int version_id)
2961 {
2962 VirtIONet *n = opaque;
2963 VirtIODevice *vdev = VIRTIO_DEVICE(n);
2964 int i, link_down;
2965
2966 trace_virtio_net_post_load_device();
2967 virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs,
2968 virtio_vdev_has_feature(vdev,
2969 VIRTIO_F_VERSION_1),
2970 virtio_vdev_has_feature(vdev,
2971 VIRTIO_NET_F_HASH_REPORT));
2972
2973 /* MAC_TABLE_ENTRIES may be different from the saved image */
2974 if (n->mac_table.in_use > MAC_TABLE_ENTRIES) {
2975 n->mac_table.in_use = 0;
2976 }
2977
2978 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
2979 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
2980 }
2981
2982 /*
2983 * curr_guest_offloads will be later overwritten by the
2984 * virtio_set_features_nocheck call done from the virtio_load.
2985 * Here we make sure it is preserved and restored accordingly
2986 * in the virtio_net_post_load_virtio callback.
2987 */
2988 n->saved_guest_offloads = n->curr_guest_offloads;
2989
2990 virtio_net_set_queue_pairs(n);
2991
2992 /* Find the first multicast entry in the saved MAC filter */
2993 for (i = 0; i < n->mac_table.in_use; i++) {
2994 if (n->mac_table.macs[i * ETH_ALEN] & 1) {
2995 break;
2996 }
2997 }
2998 n->mac_table.first_multi = i;
2999
3000 /* nc.link_down can't be migrated, so infer link_down according
3001 * to link status bit in n->status */
3002 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
3003 for (i = 0; i < n->max_queue_pairs; i++) {
3004 qemu_get_subqueue(n->nic, i)->link_down = link_down;
3005 }
3006
3007 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) &&
3008 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
3009 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
3010 QEMU_CLOCK_VIRTUAL,
3011 virtio_net_announce_timer, n);
3012 if (n->announce_timer.round) {
3013 timer_mod(n->announce_timer.tm,
3014 qemu_clock_get_ms(n->announce_timer.type));
3015 } else {
3016 qemu_announce_timer_del(&n->announce_timer, false);
3017 }
3018 }
3019
3020 if (n->rss_data.enabled) {
3021 n->rss_data.enabled_software_rss = n->rss_data.populate_hash;
3022 if (!n->rss_data.populate_hash) {
3023 if (!virtio_net_attach_epbf_rss(n)) {
3024 if (get_vhost_net(qemu_get_queue(n->nic)->peer)) {
3025 warn_report("Can't post-load eBPF RSS for vhost");
3026 } else {
3027 warn_report("Can't post-load eBPF RSS - "
3028 "fallback to software RSS");
3029 n->rss_data.enabled_software_rss = true;
3030 }
3031 }
3032 }
3033
3034 trace_virtio_net_rss_enable(n->rss_data.hash_types,
3035 n->rss_data.indirections_len,
3036 sizeof(n->rss_data.key));
3037 } else {
3038 trace_virtio_net_rss_disable();
3039 }
3040 return 0;
3041 }
3042
3043 static int virtio_net_post_load_virtio(VirtIODevice *vdev)
3044 {
3045 VirtIONet *n = VIRTIO_NET(vdev);
3046 /*
3047 * The actual needed state is now in saved_guest_offloads,
3048 * see virtio_net_post_load_device for detail.
3049 * Restore it back and apply the desired offloads.
3050 */
3051 n->curr_guest_offloads = n->saved_guest_offloads;
3052 if (peer_has_vnet_hdr(n)) {
3053 virtio_net_apply_guest_offloads(n);
3054 }
3055
3056 return 0;
3057 }
3058
3059 /* tx_waiting field of a VirtIONetQueue */
3060 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = {
3061 .name = "virtio-net-queue-tx_waiting",
3062 .fields = (VMStateField[]) {
3063 VMSTATE_UINT32(tx_waiting, VirtIONetQueue),
3064 VMSTATE_END_OF_LIST()
3065 },
3066 };
3067
3068 static bool max_queue_pairs_gt_1(void *opaque, int version_id)
3069 {
3070 return VIRTIO_NET(opaque)->max_queue_pairs > 1;
3071 }
3072
3073 static bool has_ctrl_guest_offloads(void *opaque, int version_id)
3074 {
3075 return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque),
3076 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS);
3077 }
3078
3079 static bool mac_table_fits(void *opaque, int version_id)
3080 {
3081 return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES;
3082 }
3083
3084 static bool mac_table_doesnt_fit(void *opaque, int version_id)
3085 {
3086 return !mac_table_fits(opaque, version_id);
3087 }
3088
3089 /* This temporary type is shared by all the WITH_TMP methods
3090 * although only some fields are used by each.
3091 */
3092 struct VirtIONetMigTmp {
3093 VirtIONet *parent;
3094 VirtIONetQueue *vqs_1;
3095 uint16_t curr_queue_pairs_1;
3096 uint8_t has_ufo;
3097 uint32_t has_vnet_hdr;
3098 };
3099
3100 /* The 2nd and subsequent tx_waiting flags are loaded later than
3101 * the 1st entry in the queue_pairs and only if there's more than one
3102 * entry. We use the tmp mechanism to calculate a temporary
3103 * pointer and count and also validate the count.
3104 */
3105
3106 static int virtio_net_tx_waiting_pre_save(void *opaque)
3107 {
3108 struct VirtIONetMigTmp *tmp = opaque;
3109
3110 tmp->vqs_1 = tmp->parent->vqs + 1;
3111 tmp->curr_queue_pairs_1 = tmp->parent->curr_queue_pairs - 1;
3112 if (tmp->parent->curr_queue_pairs == 0) {
3113 tmp->curr_queue_pairs_1 = 0;
3114 }
3115
3116 return 0;
3117 }
3118
3119 static int virtio_net_tx_waiting_pre_load(void *opaque)
3120 {
3121 struct VirtIONetMigTmp *tmp = opaque;
3122
3123 /* Reuse the pointer setup from save */
3124 virtio_net_tx_waiting_pre_save(opaque);
3125
3126 if (tmp->parent->curr_queue_pairs > tmp->parent->max_queue_pairs) {
3127 error_report("virtio-net: curr_queue_pairs %x > max_queue_pairs %x",
3128 tmp->parent->curr_queue_pairs, tmp->parent->max_queue_pairs);
3129
3130 return -EINVAL;
3131 }
3132
3133 return 0; /* all good */
3134 }
3135
3136 static const VMStateDescription vmstate_virtio_net_tx_waiting = {
3137 .name = "virtio-net-tx_waiting",
3138 .pre_load = virtio_net_tx_waiting_pre_load,
3139 .pre_save = virtio_net_tx_waiting_pre_save,
3140 .fields = (VMStateField[]) {
3141 VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp,
3142 curr_queue_pairs_1,
3143 vmstate_virtio_net_queue_tx_waiting,
3144 struct VirtIONetQueue),
3145 VMSTATE_END_OF_LIST()
3146 },
3147 };
3148
3149 /* the 'has_ufo' flag is just tested; if the incoming stream has the
3150 * flag set we need to check that we have it
3151 */
3152 static int virtio_net_ufo_post_load(void *opaque, int version_id)
3153 {
3154 struct VirtIONetMigTmp *tmp = opaque;
3155
3156 if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) {
3157 error_report("virtio-net: saved image requires TUN_F_UFO support");
3158 return -EINVAL;
3159 }
3160
3161 return 0;
3162 }
3163
3164 static int virtio_net_ufo_pre_save(void *opaque)
3165 {
3166 struct VirtIONetMigTmp *tmp = opaque;
3167
3168 tmp->has_ufo = tmp->parent->has_ufo;
3169
3170 return 0;
3171 }
3172
3173 static const VMStateDescription vmstate_virtio_net_has_ufo = {
3174 .name = "virtio-net-ufo",
3175 .post_load = virtio_net_ufo_post_load,
3176 .pre_save = virtio_net_ufo_pre_save,
3177 .fields = (VMStateField[]) {
3178 VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp),
3179 VMSTATE_END_OF_LIST()
3180 },
3181 };
3182
3183 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the
3184 * flag set we need to check that we have it
3185 */
3186 static int virtio_net_vnet_post_load(void *opaque, int version_id)
3187 {
3188 struct VirtIONetMigTmp *tmp = opaque;
3189
3190 if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) {
3191 error_report("virtio-net: saved image requires vnet_hdr=on");
3192 return -EINVAL;
3193 }
3194
3195 return 0;
3196 }
3197
3198 static int virtio_net_vnet_pre_save(void *opaque)
3199 {
3200 struct VirtIONetMigTmp *tmp = opaque;
3201
3202 tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr;
3203
3204 return 0;
3205 }
3206
3207 static const VMStateDescription vmstate_virtio_net_has_vnet = {
3208 .name = "virtio-net-vnet",
3209 .post_load = virtio_net_vnet_post_load,
3210 .pre_save = virtio_net_vnet_pre_save,
3211 .fields = (VMStateField[]) {
3212 VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp),
3213 VMSTATE_END_OF_LIST()
3214 },
3215 };
3216
3217 static bool virtio_net_rss_needed(void *opaque)
3218 {
3219 return VIRTIO_NET(opaque)->rss_data.enabled;
3220 }
3221
3222 static const VMStateDescription vmstate_virtio_net_rss = {
3223 .name = "virtio-net-device/rss",
3224 .version_id = 1,
3225 .minimum_version_id = 1,
3226 .needed = virtio_net_rss_needed,
3227 .fields = (VMStateField[]) {
3228 VMSTATE_BOOL(rss_data.enabled, VirtIONet),
3229 VMSTATE_BOOL(rss_data.redirect, VirtIONet),
3230 VMSTATE_BOOL(rss_data.populate_hash, VirtIONet),
3231 VMSTATE_UINT32(rss_data.hash_types, VirtIONet),
3232 VMSTATE_UINT16(rss_data.indirections_len, VirtIONet),
3233 VMSTATE_UINT16(rss_data.default_queue, VirtIONet),
3234 VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet,
3235 VIRTIO_NET_RSS_MAX_KEY_SIZE),
3236 VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet,
3237 rss_data.indirections_len, 0,
3238 vmstate_info_uint16, uint16_t),
3239 VMSTATE_END_OF_LIST()
3240 },
3241 };
3242
3243 static const VMStateDescription vmstate_virtio_net_device = {
3244 .name = "virtio-net-device",
3245 .version_id = VIRTIO_NET_VM_VERSION,
3246 .minimum_version_id = VIRTIO_NET_VM_VERSION,
3247 .post_load = virtio_net_post_load_device,
3248 .fields = (VMStateField[]) {
3249 VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN),
3250 VMSTATE_STRUCT_POINTER(vqs, VirtIONet,
3251 vmstate_virtio_net_queue_tx_waiting,
3252 VirtIONetQueue),
3253 VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet),
3254 VMSTATE_UINT16(status, VirtIONet),
3255 VMSTATE_UINT8(promisc, VirtIONet),
3256 VMSTATE_UINT8(allmulti, VirtIONet),
3257 VMSTATE_UINT32(mac_table.in_use, VirtIONet),
3258
3259 /* Guarded pair: If it fits we load it, else we throw it away
3260 * - can happen if source has a larger MAC table.; post-load
3261 * sets flags in this case.
3262 */
3263 VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet,
3264 0, mac_table_fits, mac_table.in_use,
3265 ETH_ALEN),
3266 VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0,
3267 mac_table.in_use, ETH_ALEN),
3268
3269 /* Note: This is an array of uint32's that's always been saved as a
3270 * buffer; hold onto your endiannesses; it's actually used as a bitmap
3271 * but based on the uint.
3272 */
3273 VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3),
3274 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3275 vmstate_virtio_net_has_vnet),
3276 VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet),
3277 VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet),
3278 VMSTATE_UINT8(alluni, VirtIONet),
3279 VMSTATE_UINT8(nomulti, VirtIONet),
3280 VMSTATE_UINT8(nouni, VirtIONet),
3281 VMSTATE_UINT8(nobcast, VirtIONet),
3282 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3283 vmstate_virtio_net_has_ufo),
3284 VMSTATE_SINGLE_TEST(max_queue_pairs, VirtIONet, max_queue_pairs_gt_1, 0,
3285 vmstate_info_uint16_equal, uint16_t),
3286 VMSTATE_UINT16_TEST(curr_queue_pairs, VirtIONet, max_queue_pairs_gt_1),
3287 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp,
3288 vmstate_virtio_net_tx_waiting),
3289 VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet,
3290 has_ctrl_guest_offloads),
3291 VMSTATE_END_OF_LIST()
3292 },
3293 .subsections = (const VMStateDescription * []) {
3294 &vmstate_virtio_net_rss,
3295 NULL
3296 }
3297 };
3298
3299 static NetClientInfo net_virtio_info = {
3300 .type = NET_CLIENT_DRIVER_NIC,
3301 .size = sizeof(NICState),
3302 .can_receive = virtio_net_can_receive,
3303 .receive = virtio_net_receive,
3304 .link_status_changed = virtio_net_set_link_status,
3305 .query_rx_filter = virtio_net_query_rxfilter,
3306 .announce = virtio_net_announce,
3307 };
3308
3309 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
3310 {
3311 VirtIONet *n = VIRTIO_NET(vdev);
3312 NetClientState *nc;
3313 assert(n->vhost_started);
3314 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ) && idx == 2) {
3315 /* Must guard against invalid features and bogus queue index
3316 * from being set by malicious guest, or penetrated through
3317 * buggy migration stream.
3318 */
3319 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
3320 qemu_log_mask(LOG_GUEST_ERROR,
3321 "%s: bogus vq index ignored\n", __func__);
3322 return false;
3323 }
3324 nc = qemu_get_subqueue(n->nic, n->max_queue_pairs);
3325 } else {
3326 nc = qemu_get_subqueue(n->nic, vq2q(idx));
3327 }
3328 /*
3329 * Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1
3330 * as the Marco of configure interrupt's IDX, If this driver does not
3331 * support, the function will return false
3332 */
3333
3334 if (idx == VIRTIO_CONFIG_IRQ_IDX) {
3335 return false;
3336 }
3337 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx);
3338 }
3339
3340 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
3341 bool mask)
3342 {
3343 VirtIONet *n = VIRTIO_NET(vdev);
3344 NetClientState *nc;
3345 assert(n->vhost_started);
3346 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ) && idx == 2) {
3347 /* Must guard against invalid features and bogus queue index
3348 * from being set by malicious guest, or penetrated through
3349 * buggy migration stream.
3350 */
3351 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) {
3352 qemu_log_mask(LOG_GUEST_ERROR,
3353 "%s: bogus vq index ignored\n", __func__);
3354 return;
3355 }
3356 nc = qemu_get_subqueue(n->nic, n->max_queue_pairs);
3357 } else {
3358 nc = qemu_get_subqueue(n->nic, vq2q(idx));
3359 }
3360 /*
3361 *Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1
3362 * as the Marco of configure interrupt's IDX, If this driver does not
3363 * support, the function will return
3364 */
3365
3366 if (idx == VIRTIO_CONFIG_IRQ_IDX) {
3367 return;
3368 }
3369
3370 vhost_net_virtqueue_mask(get_vhost_net(nc->peer), vdev, idx, mask);
3371 }
3372
3373 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features)
3374 {
3375 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC);
3376
3377 n->config_size = virtio_get_config_size(&cfg_size_params, host_features);
3378 }
3379
3380 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
3381 const char *type)
3382 {
3383 /*
3384 * The name can be NULL, the netclient name will be type.x.
3385 */
3386 assert(type != NULL);
3387
3388 g_free(n->netclient_name);
3389 g_free(n->netclient_type);
3390 n->netclient_name = g_strdup(name);
3391 n->netclient_type = g_strdup(type);
3392 }
3393
3394 static bool failover_unplug_primary(VirtIONet *n, DeviceState *dev)
3395 {
3396 HotplugHandler *hotplug_ctrl;
3397 PCIDevice *pci_dev;
3398 Error *err = NULL;
3399
3400 hotplug_ctrl = qdev_get_hotplug_handler(dev);
3401 if (hotplug_ctrl) {
3402 pci_dev = PCI_DEVICE(dev);
3403 pci_dev->partially_hotplugged = true;
3404 hotplug_handler_unplug_request(hotplug_ctrl, dev, &err);
3405 if (err) {
3406 error_report_err(err);
3407 return false;
3408 }
3409 } else {
3410 return false;
3411 }
3412 return true;
3413 }
3414
3415 static bool failover_replug_primary(VirtIONet *n, DeviceState *dev,
3416 Error **errp)
3417 {
3418 Error *err = NULL;
3419 HotplugHandler *hotplug_ctrl;
3420 PCIDevice *pdev = PCI_DEVICE(dev);
3421 BusState *primary_bus;
3422
3423 if (!pdev->partially_hotplugged) {
3424 return true;
3425 }
3426 primary_bus = dev->parent_bus;
3427 if (!primary_bus) {
3428 error_setg(errp, "virtio_net: couldn't find primary bus");
3429 return false;
3430 }
3431 qdev_set_parent_bus(dev, primary_bus, &error_abort);
3432 qatomic_set(&n->failover_primary_hidden, false);
3433 hotplug_ctrl = qdev_get_hotplug_handler(dev);
3434 if (hotplug_ctrl) {
3435 hotplug_handler_pre_plug(hotplug_ctrl, dev, &err);
3436 if (err) {
3437 goto out;
3438 }
3439 hotplug_handler_plug(hotplug_ctrl, dev, &err);
3440 }
3441 pdev->partially_hotplugged = false;
3442
3443 out:
3444 error_propagate(errp, err);
3445 return !err;
3446 }
3447
3448 static void virtio_net_handle_migration_primary(VirtIONet *n, MigrationState *s)
3449 {
3450 bool should_be_hidden;
3451 Error *err = NULL;
3452 DeviceState *dev = failover_find_primary_device(n);
3453
3454 if (!dev) {
3455 return;
3456 }
3457
3458 should_be_hidden = qatomic_read(&n->failover_primary_hidden);
3459
3460 if (migration_in_setup(s) && !should_be_hidden) {
3461 if (failover_unplug_primary(n, dev)) {
3462 vmstate_unregister(VMSTATE_IF(dev), qdev_get_vmsd(dev), dev);
3463 qapi_event_send_unplug_primary(dev->id);
3464 qatomic_set(&n->failover_primary_hidden, true);
3465 } else {
3466 warn_report("couldn't unplug primary device");
3467 }
3468 } else if (migration_has_failed(s)) {
3469 /* We already unplugged the device let's plug it back */
3470 if (!failover_replug_primary(n, dev, &err)) {
3471 if (err) {
3472 error_report_err(err);
3473 }
3474 }
3475 }
3476 }
3477
3478 static void virtio_net_migration_state_notifier(Notifier *notifier, void *data)
3479 {
3480 MigrationState *s = data;
3481 VirtIONet *n = container_of(notifier, VirtIONet, migration_state);
3482 virtio_net_handle_migration_primary(n, s);
3483 }
3484
3485 static bool failover_hide_primary_device(DeviceListener *listener,
3486 const QDict *device_opts,
3487 bool from_json,
3488 Error **errp)
3489 {
3490 VirtIONet *n = container_of(listener, VirtIONet, primary_listener);
3491 const char *standby_id;
3492
3493 if (!device_opts) {
3494 return false;
3495 }
3496
3497 if (!qdict_haskey(device_opts, "failover_pair_id")) {
3498 return false;
3499 }
3500
3501 if (!qdict_haskey(device_opts, "id")) {
3502 error_setg(errp, "Device with failover_pair_id needs to have id");
3503 return false;
3504 }
3505
3506 standby_id = qdict_get_str(device_opts, "failover_pair_id");
3507 if (g_strcmp0(standby_id, n->netclient_name) != 0) {
3508 return false;
3509 }
3510
3511 /*
3512 * The hide helper can be called several times for a given device.
3513 * Check there is only one primary for a virtio-net device but
3514 * don't duplicate the qdict several times if it's called for the same
3515 * device.
3516 */
3517 if (n->primary_opts) {
3518 const char *old, *new;
3519 /* devices with failover_pair_id always have an id */
3520 old = qdict_get_str(n->primary_opts, "id");
3521 new = qdict_get_str(device_opts, "id");
3522 if (strcmp(old, new) != 0) {
3523 error_setg(errp, "Cannot attach more than one primary device to "
3524 "'%s': '%s' and '%s'", n->netclient_name, old, new);
3525 return false;
3526 }
3527 } else {
3528 n->primary_opts = qdict_clone_shallow(device_opts);
3529 n->primary_opts_from_json = from_json;
3530 }
3531
3532 /* failover_primary_hidden is set during feature negotiation */
3533 return qatomic_read(&n->failover_primary_hidden);
3534 }
3535
3536 static void virtio_net_device_realize(DeviceState *dev, Error **errp)
3537 {
3538 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3539 VirtIONet *n = VIRTIO_NET(dev);
3540 NetClientState *nc;
3541 int i;
3542
3543 if (n->net_conf.mtu) {
3544 n->host_features |= (1ULL << VIRTIO_NET_F_MTU);
3545 }
3546
3547 if (n->net_conf.duplex_str) {
3548 if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) {
3549 n->net_conf.duplex = DUPLEX_HALF;
3550 } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) {
3551 n->net_conf.duplex = DUPLEX_FULL;
3552 } else {
3553 error_setg(errp, "'duplex' must be 'half' or 'full'");
3554 return;
3555 }
3556 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3557 } else {
3558 n->net_conf.duplex = DUPLEX_UNKNOWN;
3559 }
3560
3561 if (n->net_conf.speed < SPEED_UNKNOWN) {
3562 error_setg(errp, "'speed' must be between 0 and INT_MAX");
3563 return;
3564 }
3565 if (n->net_conf.speed >= 0) {
3566 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX);
3567 }
3568
3569 if (n->failover) {
3570 n->primary_listener.hide_device = failover_hide_primary_device;
3571 qatomic_set(&n->failover_primary_hidden, true);
3572 device_listener_register(&n->primary_listener);
3573 n->migration_state.notify = virtio_net_migration_state_notifier;
3574 add_migration_state_change_notifier(&n->migration_state);
3575 n->host_features |= (1ULL << VIRTIO_NET_F_STANDBY);
3576 }
3577
3578 virtio_net_set_config_size(n, n->host_features);
3579 virtio_init(vdev, VIRTIO_ID_NET, n->config_size);
3580
3581 /*
3582 * We set a lower limit on RX queue size to what it always was.
3583 * Guests that want a smaller ring can always resize it without
3584 * help from us (using virtio 1 and up).
3585 */
3586 if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE ||
3587 n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
3588 !is_power_of_2(n->net_conf.rx_queue_size)) {
3589 error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), "
3590 "must be a power of 2 between %d and %d.",
3591 n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE,
3592 VIRTQUEUE_MAX_SIZE);
3593 virtio_cleanup(vdev);
3594 return;
3595 }
3596
3597 if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE ||
3598 n->net_conf.tx_queue_size > VIRTQUEUE_MAX_SIZE ||
3599 !is_power_of_2(n->net_conf.tx_queue_size)) {
3600 error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), "
3601 "must be a power of 2 between %d and %d",
3602 n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE,
3603 VIRTQUEUE_MAX_SIZE);
3604 virtio_cleanup(vdev);
3605 return;
3606 }
3607
3608 n->max_ncs = MAX(n->nic_conf.peers.queues, 1);
3609
3610 /*
3611 * Figure out the datapath queue pairs since the backend could
3612 * provide control queue via peers as well.
3613 */
3614 if (n->nic_conf.peers.queues) {
3615 for (i = 0; i < n->max_ncs; i++) {
3616 if (n->nic_conf.peers.ncs[i]->is_datapath) {
3617 ++n->max_queue_pairs;
3618 }
3619 }
3620 }
3621 n->max_queue_pairs = MAX(n->max_queue_pairs, 1);
3622
3623 if (n->max_queue_pairs * 2 + 1 > VIRTIO_QUEUE_MAX) {
3624 error_setg(errp, "Invalid number of queue pairs (= %" PRIu32 "), "
3625 "must be a positive integer less than %d.",
3626 n->max_queue_pairs, (VIRTIO_QUEUE_MAX - 1) / 2);
3627 virtio_cleanup(vdev);
3628 return;
3629 }
3630 n->vqs = g_new0(VirtIONetQueue, n->max_queue_pairs);
3631 n->curr_queue_pairs = 1;
3632 n->tx_timeout = n->net_conf.txtimer;
3633
3634 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
3635 && strcmp(n->net_conf.tx, "bh")) {
3636 warn_report("virtio-net: "
3637 "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
3638 n->net_conf.tx);
3639 error_printf("Defaulting to \"bh\"");
3640 }
3641
3642 n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n),
3643 n->net_conf.tx_queue_size);
3644
3645 for (i = 0; i < n->max_queue_pairs; i++) {
3646 virtio_net_add_queue(n, i);
3647 }
3648
3649 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
3650 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
3651 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
3652 n->status = VIRTIO_NET_S_LINK_UP;
3653 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(),
3654 QEMU_CLOCK_VIRTUAL,
3655 virtio_net_announce_timer, n);
3656 n->announce_timer.round = 0;
3657
3658 if (n->netclient_type) {
3659 /*
3660 * Happen when virtio_net_set_netclient_name has been called.
3661 */
3662 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3663 n->netclient_type, n->netclient_name, n);
3664 } else {
3665 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
3666 object_get_typename(OBJECT(dev)), dev->id, n);
3667 }
3668
3669 for (i = 0; i < n->max_queue_pairs; i++) {
3670 n->nic->ncs[i].do_not_pad = true;
3671 }
3672
3673 peer_test_vnet_hdr(n);
3674 if (peer_has_vnet_hdr(n)) {
3675 for (i = 0; i < n->max_queue_pairs; i++) {
3676 qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
3677 }
3678 n->host_hdr_len = sizeof(struct virtio_net_hdr);
3679 } else {
3680 n->host_hdr_len = 0;
3681 }
3682
3683 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
3684
3685 n->vqs[0].tx_waiting = 0;
3686 n->tx_burst = n->net_conf.txburst;
3687 virtio_net_set_mrg_rx_bufs(n, 0, 0, 0);
3688 n->promisc = 1; /* for compatibility */
3689
3690 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
3691
3692 n->vlans = g_malloc0(MAX_VLAN >> 3);
3693
3694 nc = qemu_get_queue(n->nic);
3695 nc->rxfilter_notify_enabled = 1;
3696
3697 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
3698 struct virtio_net_config netcfg = {};
3699 memcpy(&netcfg.mac, &n->nic_conf.macaddr, ETH_ALEN);
3700 vhost_net_set_config(get_vhost_net(nc->peer),
3701 (uint8_t *)&netcfg, 0, ETH_ALEN, VHOST_SET_CONFIG_TYPE_MASTER);
3702 }
3703 QTAILQ_INIT(&n->rsc_chains);
3704 n->qdev = dev;
3705
3706 net_rx_pkt_init(&n->rx_pkt, false);
3707
3708 if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) {
3709 virtio_net_load_ebpf(n);
3710 }
3711 }
3712
3713 static void virtio_net_device_unrealize(DeviceState *dev)
3714 {
3715 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3716 VirtIONet *n = VIRTIO_NET(dev);
3717 int i, max_queue_pairs;
3718
3719 if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) {
3720 virtio_net_unload_ebpf(n);
3721 }
3722
3723 /* This will stop vhost backend if appropriate. */
3724 virtio_net_set_status(vdev, 0);
3725
3726 g_free(n->netclient_name);
3727 n->netclient_name = NULL;
3728 g_free(n->netclient_type);
3729 n->netclient_type = NULL;
3730
3731 g_free(n->mac_table.macs);
3732 g_free(n->vlans);
3733
3734 if (n->failover) {
3735 qobject_unref(n->primary_opts);
3736 device_listener_unregister(&n->primary_listener);
3737 remove_migration_state_change_notifier(&n->migration_state);
3738 } else {
3739 assert(n->primary_opts == NULL);
3740 }
3741
3742 max_queue_pairs = n->multiqueue ? n->max_queue_pairs : 1;
3743 for (i = 0; i < max_queue_pairs; i++) {
3744 virtio_net_del_queue(n, i);
3745 }
3746 /* delete also control vq */
3747 virtio_del_queue(vdev, max_queue_pairs * 2);
3748 qemu_announce_timer_del(&n->announce_timer, false);
3749 g_free(n->vqs);
3750 qemu_del_nic(n->nic);
3751 virtio_net_rsc_cleanup(n);
3752 g_free(n->rss_data.indirections_table);
3753 net_rx_pkt_uninit(n->rx_pkt);
3754 virtio_cleanup(vdev);
3755 }
3756
3757 static void virtio_net_instance_init(Object *obj)
3758 {
3759 VirtIONet *n = VIRTIO_NET(obj);
3760
3761 /*
3762 * The default config_size is sizeof(struct virtio_net_config).
3763 * Can be overriden with virtio_net_set_config_size.
3764 */
3765 n->config_size = sizeof(struct virtio_net_config);
3766 device_add_bootindex_property(obj, &n->nic_conf.bootindex,
3767 "bootindex", "/ethernet-phy@0",
3768 DEVICE(n));
3769
3770 ebpf_rss_init(&n->ebpf_rss);
3771 }
3772
3773 static int virtio_net_pre_save(void *opaque)
3774 {
3775 VirtIONet *n = opaque;
3776
3777 /* At this point, backend must be stopped, otherwise
3778 * it might keep writing to memory. */
3779 assert(!n->vhost_started);
3780
3781 return 0;
3782 }
3783
3784 static bool primary_unplug_pending(void *opaque)
3785 {
3786 DeviceState *dev = opaque;
3787 DeviceState *primary;
3788 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
3789 VirtIONet *n = VIRTIO_NET(vdev);
3790
3791 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
3792 return false;
3793 }
3794 primary = failover_find_primary_device(n);
3795 return primary ? primary->pending_deleted_event : false;
3796 }
3797
3798 static bool dev_unplug_pending(void *opaque)
3799 {
3800 DeviceState *dev = opaque;
3801 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
3802
3803 return vdc->primary_unplug_pending(dev);
3804 }
3805
3806 static struct vhost_dev *virtio_net_get_vhost(VirtIODevice *vdev)
3807 {
3808 VirtIONet *n = VIRTIO_NET(vdev);
3809 NetClientState *nc = qemu_get_queue(n->nic);
3810 struct vhost_net *net = get_vhost_net(nc->peer);
3811 return &net->dev;
3812 }
3813
3814 static const VMStateDescription vmstate_virtio_net = {
3815 .name = "virtio-net",
3816 .minimum_version_id = VIRTIO_NET_VM_VERSION,
3817 .version_id = VIRTIO_NET_VM_VERSION,
3818 .fields = (VMStateField[]) {
3819 VMSTATE_VIRTIO_DEVICE,
3820 VMSTATE_END_OF_LIST()
3821 },
3822 .pre_save = virtio_net_pre_save,
3823 .dev_unplug_pending = dev_unplug_pending,
3824 };
3825
3826 static Property virtio_net_properties[] = {
3827 DEFINE_PROP_BIT64("csum", VirtIONet, host_features,
3828 VIRTIO_NET_F_CSUM, true),
3829 DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features,
3830 VIRTIO_NET_F_GUEST_CSUM, true),
3831 DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true),
3832 DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features,
3833 VIRTIO_NET_F_GUEST_TSO4, true),
3834 DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features,
3835 VIRTIO_NET_F_GUEST_TSO6, true),
3836 DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features,
3837 VIRTIO_NET_F_GUEST_ECN, true),
3838 DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features,
3839 VIRTIO_NET_F_GUEST_UFO, true),
3840 DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features,
3841 VIRTIO_NET_F_GUEST_ANNOUNCE, true),
3842 DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features,
3843 VIRTIO_NET_F_HOST_TSO4, true),
3844 DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features,
3845 VIRTIO_NET_F_HOST_TSO6, true),
3846 DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features,
3847 VIRTIO_NET_F_HOST_ECN, true),
3848 DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features,
3849 VIRTIO_NET_F_HOST_UFO, true),
3850 DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features,
3851 VIRTIO_NET_F_MRG_RXBUF, true),
3852 DEFINE_PROP_BIT64("status", VirtIONet, host_features,
3853 VIRTIO_NET_F_STATUS, true),
3854 DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features,
3855 VIRTIO_NET_F_CTRL_VQ, true),
3856 DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features,
3857 VIRTIO_NET_F_CTRL_RX, true),
3858 DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features,
3859 VIRTIO_NET_F_CTRL_VLAN, true),
3860 DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features,
3861 VIRTIO_NET_F_CTRL_RX_EXTRA, true),
3862 DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features,
3863 VIRTIO_NET_F_CTRL_MAC_ADDR, true),
3864 DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features,
3865 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true),
3866 DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false),
3867 DEFINE_PROP_BIT64("rss", VirtIONet, host_features,
3868 VIRTIO_NET_F_RSS, false),
3869 DEFINE_PROP_BIT64("hash", VirtIONet, host_features,
3870 VIRTIO_NET_F_HASH_REPORT, false),
3871 DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features,
3872 VIRTIO_NET_F_RSC_EXT, false),
3873 DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout,
3874 VIRTIO_NET_RSC_DEFAULT_INTERVAL),
3875 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
3876 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
3877 TX_TIMER_INTERVAL),
3878 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
3879 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
3880 DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size,
3881 VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
3882 DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size,
3883 VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE),
3884 DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
3885 DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend,
3886 true),
3887 DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN),
3888 DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str),
3889 DEFINE_PROP_BOOL("failover", VirtIONet, failover, false),
3890 DEFINE_PROP_END_OF_LIST(),
3891 };
3892
3893 static void virtio_net_class_init(ObjectClass *klass, void *data)
3894 {
3895 DeviceClass *dc = DEVICE_CLASS(klass);
3896 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
3897
3898 device_class_set_props(dc, virtio_net_properties);
3899 dc->vmsd = &vmstate_virtio_net;
3900 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
3901 vdc->realize = virtio_net_device_realize;
3902 vdc->unrealize = virtio_net_device_unrealize;
3903 vdc->get_config = virtio_net_get_config;
3904 vdc->set_config = virtio_net_set_config;
3905 vdc->get_features = virtio_net_get_features;
3906 vdc->set_features = virtio_net_set_features;
3907 vdc->bad_features = virtio_net_bad_features;
3908 vdc->reset = virtio_net_reset;
3909 vdc->queue_reset = virtio_net_queue_reset;
3910 vdc->queue_enable = virtio_net_queue_enable;
3911 vdc->set_status = virtio_net_set_status;
3912 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
3913 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
3914 vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO);
3915 vdc->post_load = virtio_net_post_load_virtio;
3916 vdc->vmsd = &vmstate_virtio_net_device;
3917 vdc->primary_unplug_pending = primary_unplug_pending;
3918 vdc->get_vhost = virtio_net_get_vhost;
3919 }
3920
3921 static const TypeInfo virtio_net_info = {
3922 .name = TYPE_VIRTIO_NET,
3923 .parent = TYPE_VIRTIO_DEVICE,
3924 .instance_size = sizeof(VirtIONet),
3925 .instance_init = virtio_net_instance_init,
3926 .class_init = virtio_net_class_init,
3927 };
3928
3929 static void virtio_register_types(void)
3930 {
3931 type_register_static(&virtio_net_info);
3932 }
3933
3934 type_init(virtio_register_types)
QEmu