root / hw / virtio-net.c @ 1d41b0c1
History | View | Annotate | Download (24.5 kB)
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 "virtio.h" |
15 |
#include "net.h" |
16 |
#include "qemu-timer.h" |
17 |
#include "virtio-net.h" |
18 |
|
19 |
#define VIRTIO_NET_VM_VERSION 10 |
20 |
|
21 |
#define MAC_TABLE_ENTRIES 64 |
22 |
#define MAX_VLAN (1 << 12) /* Per 802.1Q definition */ |
23 |
|
24 |
typedef struct VirtIONet |
25 |
{ |
26 |
VirtIODevice vdev; |
27 |
uint8_t mac[ETH_ALEN]; |
28 |
uint16_t status; |
29 |
VirtQueue *rx_vq; |
30 |
VirtQueue *tx_vq; |
31 |
VirtQueue *ctrl_vq; |
32 |
VLANClientState *vc; |
33 |
QEMUTimer *tx_timer; |
34 |
int tx_timer_active;
|
35 |
uint32_t has_vnet_hdr; |
36 |
struct {
|
37 |
VirtQueueElement elem; |
38 |
ssize_t len; |
39 |
} async_tx; |
40 |
int mergeable_rx_bufs;
|
41 |
uint8_t promisc; |
42 |
uint8_t allmulti; |
43 |
uint8_t alluni; |
44 |
uint8_t nomulti; |
45 |
uint8_t nouni; |
46 |
uint8_t nobcast; |
47 |
struct {
|
48 |
int in_use;
|
49 |
int first_multi;
|
50 |
uint8_t multi_overflow; |
51 |
uint8_t uni_overflow; |
52 |
uint8_t *macs; |
53 |
} mac_table; |
54 |
uint32_t *vlans; |
55 |
} VirtIONet; |
56 |
|
57 |
/* TODO
|
58 |
* - we could suppress RX interrupt if we were so inclined.
|
59 |
*/
|
60 |
|
61 |
static VirtIONet *to_virtio_net(VirtIODevice *vdev)
|
62 |
{ |
63 |
return (VirtIONet *)vdev;
|
64 |
} |
65 |
|
66 |
static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config) |
67 |
{ |
68 |
VirtIONet *n = to_virtio_net(vdev); |
69 |
struct virtio_net_config netcfg;
|
70 |
|
71 |
netcfg.status = n->status; |
72 |
memcpy(netcfg.mac, n->mac, ETH_ALEN); |
73 |
memcpy(config, &netcfg, sizeof(netcfg));
|
74 |
} |
75 |
|
76 |
static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config) |
77 |
{ |
78 |
VirtIONet *n = to_virtio_net(vdev); |
79 |
struct virtio_net_config netcfg;
|
80 |
|
81 |
memcpy(&netcfg, config, sizeof(netcfg));
|
82 |
|
83 |
if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
|
84 |
memcpy(n->mac, netcfg.mac, ETH_ALEN); |
85 |
qemu_format_nic_info_str(n->vc, n->mac); |
86 |
} |
87 |
} |
88 |
|
89 |
static void virtio_net_set_link_status(VLANClientState *vc) |
90 |
{ |
91 |
VirtIONet *n = vc->opaque; |
92 |
uint16_t old_status = n->status; |
93 |
|
94 |
if (vc->link_down)
|
95 |
n->status &= ~VIRTIO_NET_S_LINK_UP; |
96 |
else
|
97 |
n->status |= VIRTIO_NET_S_LINK_UP; |
98 |
|
99 |
if (n->status != old_status)
|
100 |
virtio_notify_config(&n->vdev); |
101 |
} |
102 |
|
103 |
static void virtio_net_reset(VirtIODevice *vdev) |
104 |
{ |
105 |
VirtIONet *n = to_virtio_net(vdev); |
106 |
|
107 |
/* Reset back to compatibility mode */
|
108 |
n->promisc = 1;
|
109 |
n->allmulti = 0;
|
110 |
n->alluni = 0;
|
111 |
n->nomulti = 0;
|
112 |
n->nouni = 0;
|
113 |
n->nobcast = 0;
|
114 |
|
115 |
/* Flush any MAC and VLAN filter table state */
|
116 |
n->mac_table.in_use = 0;
|
117 |
n->mac_table.first_multi = 0;
|
118 |
n->mac_table.multi_overflow = 0;
|
119 |
n->mac_table.uni_overflow = 0;
|
120 |
memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
|
121 |
memset(n->vlans, 0, MAX_VLAN >> 3); |
122 |
} |
123 |
|
124 |
static int peer_has_vnet_hdr(VirtIONet *n) |
125 |
{ |
126 |
if (!n->vc->peer)
|
127 |
return 0; |
128 |
|
129 |
if (n->vc->peer->type != NET_CLIENT_TYPE_TAP)
|
130 |
return 0; |
131 |
|
132 |
n->has_vnet_hdr = tap_has_vnet_hdr(n->vc->peer); |
133 |
|
134 |
return n->has_vnet_hdr;
|
135 |
} |
136 |
|
137 |
static uint32_t virtio_net_get_features(VirtIODevice *vdev)
|
138 |
{ |
139 |
VirtIONet *n = to_virtio_net(vdev); |
140 |
uint32_t features = (1 << VIRTIO_NET_F_MAC) |
|
141 |
(1 << VIRTIO_NET_F_MRG_RXBUF) |
|
142 |
(1 << VIRTIO_NET_F_STATUS) |
|
143 |
(1 << VIRTIO_NET_F_CTRL_VQ) |
|
144 |
(1 << VIRTIO_NET_F_CTRL_RX) |
|
145 |
(1 << VIRTIO_NET_F_CTRL_VLAN) |
|
146 |
(1 << VIRTIO_NET_F_CTRL_RX_EXTRA);
|
147 |
|
148 |
if (peer_has_vnet_hdr(n)) {
|
149 |
tap_using_vnet_hdr(n->vc->peer, 1);
|
150 |
|
151 |
features |= (1 << VIRTIO_NET_F_CSUM);
|
152 |
features |= (1 << VIRTIO_NET_F_HOST_TSO4);
|
153 |
features |= (1 << VIRTIO_NET_F_HOST_TSO6);
|
154 |
features |= (1 << VIRTIO_NET_F_HOST_ECN);
|
155 |
|
156 |
features |= (1 << VIRTIO_NET_F_GUEST_CSUM);
|
157 |
features |= (1 << VIRTIO_NET_F_GUEST_TSO4);
|
158 |
features |= (1 << VIRTIO_NET_F_GUEST_TSO6);
|
159 |
features |= (1 << VIRTIO_NET_F_GUEST_ECN);
|
160 |
} |
161 |
|
162 |
return features;
|
163 |
} |
164 |
|
165 |
static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
|
166 |
{ |
167 |
uint32_t features = 0;
|
168 |
|
169 |
/* Linux kernel 2.6.25. It understood MAC (as everyone must),
|
170 |
* but also these: */
|
171 |
features |= (1 << VIRTIO_NET_F_MAC);
|
172 |
features |= (1 << VIRTIO_NET_F_GUEST_CSUM);
|
173 |
features |= (1 << VIRTIO_NET_F_GUEST_TSO4);
|
174 |
features |= (1 << VIRTIO_NET_F_GUEST_TSO6);
|
175 |
features |= (1 << VIRTIO_NET_F_GUEST_ECN);
|
176 |
|
177 |
return features & virtio_net_get_features(vdev);
|
178 |
} |
179 |
|
180 |
static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features) |
181 |
{ |
182 |
VirtIONet *n = to_virtio_net(vdev); |
183 |
|
184 |
n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
|
185 |
|
186 |
if (n->has_vnet_hdr) {
|
187 |
tap_set_offload(n->vc->peer, |
188 |
(features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
|
189 |
(features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
|
190 |
(features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
|
191 |
(features >> VIRTIO_NET_F_GUEST_ECN) & 1);
|
192 |
} |
193 |
} |
194 |
|
195 |
static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd, |
196 |
VirtQueueElement *elem) |
197 |
{ |
198 |
uint8_t on; |
199 |
|
200 |
if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) { |
201 |
fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
|
202 |
exit(1);
|
203 |
} |
204 |
|
205 |
on = ldub_p(elem->out_sg[1].iov_base);
|
206 |
|
207 |
if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
|
208 |
n->promisc = on; |
209 |
else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI) |
210 |
n->allmulti = on; |
211 |
else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI) |
212 |
n->alluni = on; |
213 |
else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI) |
214 |
n->nomulti = on; |
215 |
else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI) |
216 |
n->nouni = on; |
217 |
else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST) |
218 |
n->nobcast = on; |
219 |
else
|
220 |
return VIRTIO_NET_ERR;
|
221 |
|
222 |
return VIRTIO_NET_OK;
|
223 |
} |
224 |
|
225 |
static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd, |
226 |
VirtQueueElement *elem) |
227 |
{ |
228 |
struct virtio_net_ctrl_mac mac_data;
|
229 |
|
230 |
if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 || |
231 |
elem->out_sg[1].iov_len < sizeof(mac_data) || |
232 |
elem->out_sg[2].iov_len < sizeof(mac_data)) |
233 |
return VIRTIO_NET_ERR;
|
234 |
|
235 |
n->mac_table.in_use = 0;
|
236 |
n->mac_table.first_multi = 0;
|
237 |
n->mac_table.uni_overflow = 0;
|
238 |
n->mac_table.multi_overflow = 0;
|
239 |
memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
|
240 |
|
241 |
mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
|
242 |
|
243 |
if (sizeof(mac_data.entries) + |
244 |
(mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
|
245 |
return VIRTIO_NET_ERR;
|
246 |
|
247 |
if (mac_data.entries <= MAC_TABLE_ENTRIES) {
|
248 |
memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data), |
249 |
mac_data.entries * ETH_ALEN); |
250 |
n->mac_table.in_use += mac_data.entries; |
251 |
} else {
|
252 |
n->mac_table.uni_overflow = 1;
|
253 |
} |
254 |
|
255 |
n->mac_table.first_multi = n->mac_table.in_use; |
256 |
|
257 |
mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
|
258 |
|
259 |
if (sizeof(mac_data.entries) + |
260 |
(mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
|
261 |
return VIRTIO_NET_ERR;
|
262 |
|
263 |
if (mac_data.entries) {
|
264 |
if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
|
265 |
memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN), |
266 |
elem->out_sg[2].iov_base + sizeof(mac_data), |
267 |
mac_data.entries * ETH_ALEN); |
268 |
n->mac_table.in_use += mac_data.entries; |
269 |
} else {
|
270 |
n->mac_table.multi_overflow = 1;
|
271 |
} |
272 |
} |
273 |
|
274 |
return VIRTIO_NET_OK;
|
275 |
} |
276 |
|
277 |
static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd, |
278 |
VirtQueueElement *elem) |
279 |
{ |
280 |
uint16_t vid; |
281 |
|
282 |
if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) { |
283 |
fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
|
284 |
return VIRTIO_NET_ERR;
|
285 |
} |
286 |
|
287 |
vid = lduw_le_p(elem->out_sg[1].iov_base);
|
288 |
|
289 |
if (vid >= MAX_VLAN)
|
290 |
return VIRTIO_NET_ERR;
|
291 |
|
292 |
if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
|
293 |
n->vlans[vid >> 5] |= (1U << (vid & 0x1f)); |
294 |
else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL) |
295 |
n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f)); |
296 |
else
|
297 |
return VIRTIO_NET_ERR;
|
298 |
|
299 |
return VIRTIO_NET_OK;
|
300 |
} |
301 |
|
302 |
static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq) |
303 |
{ |
304 |
VirtIONet *n = to_virtio_net(vdev); |
305 |
struct virtio_net_ctrl_hdr ctrl;
|
306 |
virtio_net_ctrl_ack status = VIRTIO_NET_ERR; |
307 |
VirtQueueElement elem; |
308 |
|
309 |
while (virtqueue_pop(vq, &elem)) {
|
310 |
if ((elem.in_num < 1) || (elem.out_num < 1)) { |
311 |
fprintf(stderr, "virtio-net ctrl missing headers\n");
|
312 |
exit(1);
|
313 |
} |
314 |
|
315 |
if (elem.out_sg[0].iov_len < sizeof(ctrl) || |
316 |
elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) { |
317 |
fprintf(stderr, "virtio-net ctrl header not in correct element\n");
|
318 |
exit(1);
|
319 |
} |
320 |
|
321 |
ctrl.class = ldub_p(elem.out_sg[0].iov_base);
|
322 |
ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class)); |
323 |
|
324 |
if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
|
325 |
status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem); |
326 |
else if (ctrl.class == VIRTIO_NET_CTRL_MAC) |
327 |
status = virtio_net_handle_mac(n, ctrl.cmd, &elem); |
328 |
else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) |
329 |
status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem); |
330 |
|
331 |
stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
|
332 |
|
333 |
virtqueue_push(vq, &elem, sizeof(status));
|
334 |
virtio_notify(vdev, vq); |
335 |
} |
336 |
} |
337 |
|
338 |
/* RX */
|
339 |
|
340 |
static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq) |
341 |
{ |
342 |
VirtIONet *n = to_virtio_net(vdev); |
343 |
|
344 |
qemu_flush_queued_packets(n->vc); |
345 |
|
346 |
/* We now have RX buffers, signal to the IO thread to break out of the
|
347 |
* select to re-poll the tap file descriptor */
|
348 |
qemu_notify_event(); |
349 |
} |
350 |
|
351 |
static int do_virtio_net_can_receive(VirtIONet *n, int bufsize) |
352 |
{ |
353 |
if (!virtio_queue_ready(n->rx_vq) ||
|
354 |
!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) |
355 |
return 0; |
356 |
|
357 |
if (virtio_queue_empty(n->rx_vq) ||
|
358 |
(n->mergeable_rx_bufs && |
359 |
!virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
|
360 |
virtio_queue_set_notification(n->rx_vq, 1);
|
361 |
return 0; |
362 |
} |
363 |
|
364 |
virtio_queue_set_notification(n->rx_vq, 0);
|
365 |
return 1; |
366 |
} |
367 |
|
368 |
static int virtio_net_can_receive(VLANClientState *vc) |
369 |
{ |
370 |
VirtIONet *n = vc->opaque; |
371 |
|
372 |
return do_virtio_net_can_receive(n, VIRTIO_NET_MAX_BUFSIZE);
|
373 |
} |
374 |
|
375 |
/* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
|
376 |
* it never finds out that the packets don't have valid checksums. This
|
377 |
* causes dhclient to get upset. Fedora's carried a patch for ages to
|
378 |
* fix this with Xen but it hasn't appeared in an upstream release of
|
379 |
* dhclient yet.
|
380 |
*
|
381 |
* To avoid breaking existing guests, we catch udp packets and add
|
382 |
* checksums. This is terrible but it's better than hacking the guest
|
383 |
* kernels.
|
384 |
*
|
385 |
* N.B. if we introduce a zero-copy API, this operation is no longer free so
|
386 |
* we should provide a mechanism to disable it to avoid polluting the host
|
387 |
* cache.
|
388 |
*/
|
389 |
static void work_around_broken_dhclient(struct virtio_net_hdr *hdr, |
390 |
const uint8_t *buf, size_t size)
|
391 |
{ |
392 |
if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */ |
393 |
(size > 27 && size < 1500) && /* normal sized MTU */ |
394 |
(buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */ |
395 |
(buf[23] == 17) && /* ip.protocol == UDP */ |
396 |
(buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */ |
397 |
/* FIXME this cast is evil */
|
398 |
net_checksum_calculate((uint8_t *)buf, size); |
399 |
hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM; |
400 |
} |
401 |
} |
402 |
|
403 |
static int iov_fill(struct iovec *iov, int iovcnt, const void *buf, int count) |
404 |
{ |
405 |
int offset, i;
|
406 |
|
407 |
offset = i = 0;
|
408 |
while (offset < count && i < iovcnt) {
|
409 |
int len = MIN(iov[i].iov_len, count - offset);
|
410 |
memcpy(iov[i].iov_base, buf + offset, len); |
411 |
offset += len; |
412 |
i++; |
413 |
} |
414 |
|
415 |
return offset;
|
416 |
} |
417 |
|
418 |
static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt, |
419 |
const void *buf, size_t size, size_t hdr_len) |
420 |
{ |
421 |
struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base; |
422 |
int offset = 0; |
423 |
|
424 |
hdr->flags = 0;
|
425 |
hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; |
426 |
|
427 |
if (n->has_vnet_hdr) {
|
428 |
memcpy(hdr, buf, sizeof(*hdr));
|
429 |
offset = sizeof(*hdr);
|
430 |
work_around_broken_dhclient(hdr, buf + offset, size - offset); |
431 |
} |
432 |
|
433 |
/* We only ever receive a struct virtio_net_hdr from the tapfd,
|
434 |
* but we may be passing along a larger header to the guest.
|
435 |
*/
|
436 |
iov[0].iov_base += hdr_len;
|
437 |
iov[0].iov_len -= hdr_len;
|
438 |
|
439 |
return offset;
|
440 |
} |
441 |
|
442 |
static int receive_filter(VirtIONet *n, const uint8_t *buf, int size) |
443 |
{ |
444 |
static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; |
445 |
static const uint8_t vlan[] = {0x81, 0x00}; |
446 |
uint8_t *ptr = (uint8_t *)buf; |
447 |
int i;
|
448 |
|
449 |
if (n->promisc)
|
450 |
return 1; |
451 |
|
452 |
if (n->has_vnet_hdr) {
|
453 |
ptr += sizeof(struct virtio_net_hdr); |
454 |
} |
455 |
|
456 |
if (!memcmp(&ptr[12], vlan, sizeof(vlan))) { |
457 |
int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff; |
458 |
if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f)))) |
459 |
return 0; |
460 |
} |
461 |
|
462 |
if (ptr[0] & 1) { // multicast |
463 |
if (!memcmp(ptr, bcast, sizeof(bcast))) { |
464 |
return !n->nobcast;
|
465 |
} else if (n->nomulti) { |
466 |
return 0; |
467 |
} else if (n->allmulti || n->mac_table.multi_overflow) { |
468 |
return 1; |
469 |
} |
470 |
|
471 |
for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
|
472 |
if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
|
473 |
return 1; |
474 |
} |
475 |
} |
476 |
} else { // unicast |
477 |
if (n->nouni) {
|
478 |
return 0; |
479 |
} else if (n->alluni || n->mac_table.uni_overflow) { |
480 |
return 1; |
481 |
} else if (!memcmp(ptr, n->mac, ETH_ALEN)) { |
482 |
return 1; |
483 |
} |
484 |
|
485 |
for (i = 0; i < n->mac_table.first_multi; i++) { |
486 |
if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
|
487 |
return 1; |
488 |
} |
489 |
} |
490 |
} |
491 |
|
492 |
return 0; |
493 |
} |
494 |
|
495 |
static ssize_t virtio_net_receive(VLANClientState *vc, const uint8_t *buf, size_t size) |
496 |
{ |
497 |
VirtIONet *n = vc->opaque; |
498 |
struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL; |
499 |
size_t hdr_len, offset, i; |
500 |
|
501 |
if (!do_virtio_net_can_receive(n, size))
|
502 |
return 0; |
503 |
|
504 |
if (!receive_filter(n, buf, size))
|
505 |
return size;
|
506 |
|
507 |
/* hdr_len refers to the header we supply to the guest */
|
508 |
hdr_len = n->mergeable_rx_bufs ? |
509 |
sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr); |
510 |
|
511 |
offset = i = 0;
|
512 |
|
513 |
while (offset < size) {
|
514 |
VirtQueueElement elem; |
515 |
int len, total;
|
516 |
struct iovec sg[VIRTQUEUE_MAX_SIZE];
|
517 |
|
518 |
len = total = 0;
|
519 |
|
520 |
if ((i != 0 && !n->mergeable_rx_bufs) || |
521 |
virtqueue_pop(n->rx_vq, &elem) == 0) {
|
522 |
if (i == 0) |
523 |
return -1; |
524 |
fprintf(stderr, "virtio-net truncating packet\n");
|
525 |
exit(1);
|
526 |
} |
527 |
|
528 |
if (elem.in_num < 1) { |
529 |
fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
|
530 |
exit(1);
|
531 |
} |
532 |
|
533 |
if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != hdr_len) { |
534 |
fprintf(stderr, "virtio-net header not in first element\n");
|
535 |
exit(1);
|
536 |
} |
537 |
|
538 |
memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num); |
539 |
|
540 |
if (i == 0) { |
541 |
if (n->mergeable_rx_bufs)
|
542 |
mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base; |
543 |
|
544 |
offset += receive_header(n, sg, elem.in_num, |
545 |
buf + offset, size - offset, hdr_len); |
546 |
total += hdr_len; |
547 |
} |
548 |
|
549 |
/* copy in packet. ugh */
|
550 |
len = iov_fill(sg, elem.in_num, |
551 |
buf + offset, size - offset); |
552 |
total += len; |
553 |
|
554 |
/* signal other side */
|
555 |
virtqueue_fill(n->rx_vq, &elem, total, i++); |
556 |
|
557 |
offset += len; |
558 |
} |
559 |
|
560 |
if (mhdr)
|
561 |
mhdr->num_buffers = i; |
562 |
|
563 |
virtqueue_flush(n->rx_vq, i); |
564 |
virtio_notify(&n->vdev, n->rx_vq); |
565 |
|
566 |
return size;
|
567 |
} |
568 |
|
569 |
static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq); |
570 |
|
571 |
static void virtio_net_tx_complete(VLANClientState *vc, ssize_t len) |
572 |
{ |
573 |
VirtIONet *n = vc->opaque; |
574 |
|
575 |
virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len); |
576 |
virtio_notify(&n->vdev, n->tx_vq); |
577 |
|
578 |
n->async_tx.elem.out_num = n->async_tx.len = 0;
|
579 |
|
580 |
virtio_queue_set_notification(n->tx_vq, 1);
|
581 |
virtio_net_flush_tx(n, n->tx_vq); |
582 |
} |
583 |
|
584 |
/* TX */
|
585 |
static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq) |
586 |
{ |
587 |
VirtQueueElement elem; |
588 |
|
589 |
if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
|
590 |
return;
|
591 |
|
592 |
if (n->async_tx.elem.out_num) {
|
593 |
virtio_queue_set_notification(n->tx_vq, 0);
|
594 |
return;
|
595 |
} |
596 |
|
597 |
while (virtqueue_pop(vq, &elem)) {
|
598 |
ssize_t ret, len = 0;
|
599 |
unsigned int out_num = elem.out_num; |
600 |
struct iovec *out_sg = &elem.out_sg[0]; |
601 |
unsigned hdr_len;
|
602 |
|
603 |
/* hdr_len refers to the header received from the guest */
|
604 |
hdr_len = n->mergeable_rx_bufs ? |
605 |
sizeof(struct virtio_net_hdr_mrg_rxbuf) : |
606 |
sizeof(struct virtio_net_hdr); |
607 |
|
608 |
if (out_num < 1 || out_sg->iov_len != hdr_len) { |
609 |
fprintf(stderr, "virtio-net header not in first element\n");
|
610 |
exit(1);
|
611 |
} |
612 |
|
613 |
/* ignore the header if GSO is not supported */
|
614 |
if (!n->has_vnet_hdr) {
|
615 |
out_num--; |
616 |
out_sg++; |
617 |
len += hdr_len; |
618 |
} else if (n->mergeable_rx_bufs) { |
619 |
/* tapfd expects a struct virtio_net_hdr */
|
620 |
hdr_len -= sizeof(struct virtio_net_hdr); |
621 |
out_sg->iov_len -= hdr_len; |
622 |
len += hdr_len; |
623 |
} |
624 |
|
625 |
ret = qemu_sendv_packet_async(n->vc, out_sg, out_num, |
626 |
virtio_net_tx_complete); |
627 |
if (ret == 0) { |
628 |
virtio_queue_set_notification(n->tx_vq, 0);
|
629 |
n->async_tx.elem = elem; |
630 |
n->async_tx.len = len; |
631 |
return;
|
632 |
} |
633 |
|
634 |
len += ret; |
635 |
|
636 |
virtqueue_push(vq, &elem, len); |
637 |
virtio_notify(&n->vdev, vq); |
638 |
} |
639 |
} |
640 |
|
641 |
static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq) |
642 |
{ |
643 |
VirtIONet *n = to_virtio_net(vdev); |
644 |
|
645 |
if (n->tx_timer_active) {
|
646 |
virtio_queue_set_notification(vq, 1);
|
647 |
qemu_del_timer(n->tx_timer); |
648 |
n->tx_timer_active = 0;
|
649 |
virtio_net_flush_tx(n, vq); |
650 |
} else {
|
651 |
qemu_mod_timer(n->tx_timer, |
652 |
qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL); |
653 |
n->tx_timer_active = 1;
|
654 |
virtio_queue_set_notification(vq, 0);
|
655 |
} |
656 |
} |
657 |
|
658 |
static void virtio_net_tx_timer(void *opaque) |
659 |
{ |
660 |
VirtIONet *n = opaque; |
661 |
|
662 |
n->tx_timer_active = 0;
|
663 |
|
664 |
/* Just in case the driver is not ready on more */
|
665 |
if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
|
666 |
return;
|
667 |
|
668 |
virtio_queue_set_notification(n->tx_vq, 1);
|
669 |
virtio_net_flush_tx(n, n->tx_vq); |
670 |
} |
671 |
|
672 |
static void virtio_net_save(QEMUFile *f, void *opaque) |
673 |
{ |
674 |
VirtIONet *n = opaque; |
675 |
|
676 |
virtio_save(&n->vdev, f); |
677 |
|
678 |
qemu_put_buffer(f, n->mac, ETH_ALEN); |
679 |
qemu_put_be32(f, n->tx_timer_active); |
680 |
qemu_put_be32(f, n->mergeable_rx_bufs); |
681 |
qemu_put_be16(f, n->status); |
682 |
qemu_put_byte(f, n->promisc); |
683 |
qemu_put_byte(f, n->allmulti); |
684 |
qemu_put_be32(f, n->mac_table.in_use); |
685 |
qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN); |
686 |
qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
|
687 |
qemu_put_be32(f, n->has_vnet_hdr); |
688 |
qemu_put_byte(f, n->mac_table.multi_overflow); |
689 |
qemu_put_byte(f, n->mac_table.uni_overflow); |
690 |
qemu_put_byte(f, n->alluni); |
691 |
qemu_put_byte(f, n->nomulti); |
692 |
qemu_put_byte(f, n->nouni); |
693 |
qemu_put_byte(f, n->nobcast); |
694 |
} |
695 |
|
696 |
static int virtio_net_load(QEMUFile *f, void *opaque, int version_id) |
697 |
{ |
698 |
VirtIONet *n = opaque; |
699 |
int i;
|
700 |
|
701 |
if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION) |
702 |
return -EINVAL;
|
703 |
|
704 |
virtio_load(&n->vdev, f); |
705 |
|
706 |
qemu_get_buffer(f, n->mac, ETH_ALEN); |
707 |
n->tx_timer_active = qemu_get_be32(f); |
708 |
n->mergeable_rx_bufs = qemu_get_be32(f); |
709 |
|
710 |
if (version_id >= 3) |
711 |
n->status = qemu_get_be16(f); |
712 |
|
713 |
if (version_id >= 4) { |
714 |
if (version_id < 8) { |
715 |
n->promisc = qemu_get_be32(f); |
716 |
n->allmulti = qemu_get_be32(f); |
717 |
} else {
|
718 |
n->promisc = qemu_get_byte(f); |
719 |
n->allmulti = qemu_get_byte(f); |
720 |
} |
721 |
} |
722 |
|
723 |
if (version_id >= 5) { |
724 |
n->mac_table.in_use = qemu_get_be32(f); |
725 |
/* MAC_TABLE_ENTRIES may be different from the saved image */
|
726 |
if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
|
727 |
qemu_get_buffer(f, n->mac_table.macs, |
728 |
n->mac_table.in_use * ETH_ALEN); |
729 |
} else if (n->mac_table.in_use) { |
730 |
qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR); |
731 |
n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
|
732 |
n->mac_table.in_use = 0;
|
733 |
} |
734 |
} |
735 |
|
736 |
if (version_id >= 6) |
737 |
qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
|
738 |
|
739 |
if (version_id >= 7) { |
740 |
if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
|
741 |
qemu_error("virtio-net: saved image requires vnet_hdr=on\n");
|
742 |
return -1; |
743 |
} |
744 |
|
745 |
if (n->has_vnet_hdr) {
|
746 |
tap_using_vnet_hdr(n->vc->peer, 1);
|
747 |
tap_set_offload(n->vc->peer, |
748 |
(n->vdev.features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
|
749 |
(n->vdev.features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
|
750 |
(n->vdev.features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
|
751 |
(n->vdev.features >> VIRTIO_NET_F_GUEST_ECN) & 1);
|
752 |
} |
753 |
} |
754 |
|
755 |
if (version_id >= 9) { |
756 |
n->mac_table.multi_overflow = qemu_get_byte(f); |
757 |
n->mac_table.uni_overflow = qemu_get_byte(f); |
758 |
} |
759 |
|
760 |
if (version_id >= 10) { |
761 |
n->alluni = qemu_get_byte(f); |
762 |
n->nomulti = qemu_get_byte(f); |
763 |
n->nouni = qemu_get_byte(f); |
764 |
n->nobcast = qemu_get_byte(f); |
765 |
} |
766 |
|
767 |
/* Find the first multicast entry in the saved MAC filter */
|
768 |
for (i = 0; i < n->mac_table.in_use; i++) { |
769 |
if (n->mac_table.macs[i * ETH_ALEN] & 1) { |
770 |
break;
|
771 |
} |
772 |
} |
773 |
n->mac_table.first_multi = i; |
774 |
|
775 |
if (n->tx_timer_active) {
|
776 |
qemu_mod_timer(n->tx_timer, |
777 |
qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL); |
778 |
} |
779 |
|
780 |
return 0; |
781 |
} |
782 |
|
783 |
static void virtio_net_cleanup(VLANClientState *vc) |
784 |
{ |
785 |
VirtIONet *n = vc->opaque; |
786 |
|
787 |
n->vc = NULL;
|
788 |
} |
789 |
|
790 |
VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf) |
791 |
{ |
792 |
VirtIONet *n; |
793 |
static int virtio_net_id; |
794 |
|
795 |
n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
|
796 |
sizeof(struct virtio_net_config), |
797 |
sizeof(VirtIONet));
|
798 |
|
799 |
n->vdev.get_config = virtio_net_get_config; |
800 |
n->vdev.set_config = virtio_net_set_config; |
801 |
n->vdev.get_features = virtio_net_get_features; |
802 |
n->vdev.set_features = virtio_net_set_features; |
803 |
n->vdev.bad_features = virtio_net_bad_features; |
804 |
n->vdev.reset = virtio_net_reset; |
805 |
n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
|
806 |
n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
|
807 |
n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
|
808 |
qemu_macaddr_default_if_unset(&conf->macaddr); |
809 |
n->status = VIRTIO_NET_S_LINK_UP; |
810 |
n->vc = qemu_new_vlan_client(NET_CLIENT_TYPE_NIC, conf->vlan, conf->peer, |
811 |
dev->info->name, dev->id, |
812 |
virtio_net_can_receive, |
813 |
virtio_net_receive, NULL, NULL, |
814 |
virtio_net_cleanup, n); |
815 |
n->vc->link_status_changed = virtio_net_set_link_status; |
816 |
|
817 |
qemu_format_nic_info_str(n->vc, conf->macaddr.a); |
818 |
|
819 |
n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n); |
820 |
n->tx_timer_active = 0;
|
821 |
n->mergeable_rx_bufs = 0;
|
822 |
n->promisc = 1; /* for compatibility */ |
823 |
|
824 |
n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN); |
825 |
|
826 |
n->vlans = qemu_mallocz(MAX_VLAN >> 3);
|
827 |
|
828 |
register_savevm("virtio-net", virtio_net_id++, VIRTIO_NET_VM_VERSION,
|
829 |
virtio_net_save, virtio_net_load, n); |
830 |
|
831 |
return &n->vdev;
|
832 |
} |
833 |
|
834 |
void virtio_net_exit(VirtIODevice *vdev)
|
835 |
{ |
836 |
VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev); |
837 |
|
838 |
qemu_purge_queued_packets(n->vc); |
839 |
|
840 |
unregister_savevm("virtio-net", n);
|
841 |
|
842 |
qemu_free(n->mac_table.macs); |
843 |
qemu_free(n->vlans); |
844 |
|
845 |
qemu_del_timer(n->tx_timer); |
846 |
qemu_free_timer(n->tx_timer); |
847 |
|
848 |
virtio_cleanup(&n->vdev); |
849 |
qemu_del_vlan_client(n->vc); |
850 |
} |