Statistics
| Branch: | Revision:

root / hw / net / virtio-net.c @ 1773d9ee

History | View | Annotate | Download (42.1 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 "qemu/iov.h"
15
#include "hw/virtio/virtio.h"
16
#include "net/net.h"
17
#include "net/checksum.h"
18
#include "net/tap.h"
19
#include "qemu/error-report.h"
20
#include "qemu/timer.h"
21
#include "hw/virtio/virtio-net.h"
22
#include "net/vhost_net.h"
23
#include "hw/virtio/virtio-bus.h"
24

    
25
#define VIRTIO_NET_VM_VERSION    11
26

    
27
#define MAC_TABLE_ENTRIES    64
28
#define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
29

    
30
/*
31
 * Calculate the number of bytes up to and including the given 'field' of
32
 * 'container'.
33
 */
34
#define endof(container, field) \
35
    (offsetof(container, field) + sizeof(((container *)0)->field))
36

    
37
typedef struct VirtIOFeature {
38
    uint32_t flags;
39
    size_t end;
40
} VirtIOFeature;
41

    
42
static VirtIOFeature feature_sizes[] = {
43
    {.flags = 1 << VIRTIO_NET_F_MAC,
44
     .end = endof(struct virtio_net_config, mac)},
45
    {.flags = 1 << VIRTIO_NET_F_STATUS,
46
     .end = endof(struct virtio_net_config, status)},
47
    {.flags = 1 << VIRTIO_NET_F_MQ,
48
     .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
49
    {}
50
};
51

    
52
static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
53
{
54
    VirtIONet *n = qemu_get_nic_opaque(nc);
55

    
56
    return &n->vqs[nc->queue_index];
57
}
58

    
59
static int vq2q(int queue_index)
60
{
61
    return queue_index / 2;
62
}
63

    
64
/* TODO
65
 * - we could suppress RX interrupt if we were so inclined.
66
 */
67

    
68
static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
69
{
70
    VirtIONet *n = VIRTIO_NET(vdev);
71
    struct virtio_net_config netcfg;
72

    
73
    stw_p(&netcfg.status, n->status);
74
    stw_p(&netcfg.max_virtqueue_pairs, n->max_queues);
75
    memcpy(netcfg.mac, n->mac, ETH_ALEN);
76
    memcpy(config, &netcfg, n->config_size);
77
}
78

    
79
static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
80
{
81
    VirtIONet *n = VIRTIO_NET(vdev);
82
    struct virtio_net_config netcfg = {};
83

    
84
    memcpy(&netcfg, config, n->config_size);
85

    
86
    if (!(vdev->guest_features >> VIRTIO_NET_F_CTRL_MAC_ADDR & 1) &&
87
        memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
88
        memcpy(n->mac, netcfg.mac, ETH_ALEN);
89
        qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
90
    }
91
}
92

    
93
static bool virtio_net_started(VirtIONet *n, uint8_t status)
94
{
95
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
96
    return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
97
        (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
98
}
99

    
100
static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
101
{
102
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
103
    NetClientState *nc = qemu_get_queue(n->nic);
104
    int queues = n->multiqueue ? n->max_queues : 1;
105

    
106
    if (!nc->peer) {
107
        return;
108
    }
109
    if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
110
        return;
111
    }
112

    
113
    if (!tap_get_vhost_net(nc->peer)) {
114
        return;
115
    }
116

    
117
    if (!!n->vhost_started == virtio_net_started(n, status) &&
118
                              !nc->peer->link_down) {
119
        return;
120
    }
121
    if (!n->vhost_started) {
122
        int r;
123
        if (!vhost_net_query(tap_get_vhost_net(nc->peer), vdev)) {
124
            return;
125
        }
126
        n->vhost_started = 1;
127
        r = vhost_net_start(vdev, n->nic->ncs, queues);
128
        if (r < 0) {
129
            error_report("unable to start vhost net: %d: "
130
                         "falling back on userspace virtio", -r);
131
            n->vhost_started = 0;
132
        }
133
    } else {
134
        vhost_net_stop(vdev, n->nic->ncs, queues);
135
        n->vhost_started = 0;
136
    }
137
}
138

    
139
static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
140
{
141
    VirtIONet *n = VIRTIO_NET(vdev);
142
    VirtIONetQueue *q;
143
    int i;
144
    uint8_t queue_status;
145

    
146
    virtio_net_vhost_status(n, status);
147

    
148
    for (i = 0; i < n->max_queues; i++) {
149
        q = &n->vqs[i];
150

    
151
        if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
152
            queue_status = 0;
153
        } else {
154
            queue_status = status;
155
        }
156

    
157
        if (!q->tx_waiting) {
158
            continue;
159
        }
160

    
161
        if (virtio_net_started(n, queue_status) && !n->vhost_started) {
162
            if (q->tx_timer) {
163
                qemu_mod_timer(q->tx_timer,
164
                               qemu_get_clock_ns(vm_clock) + n->tx_timeout);
165
            } else {
166
                qemu_bh_schedule(q->tx_bh);
167
            }
168
        } else {
169
            if (q->tx_timer) {
170
                qemu_del_timer(q->tx_timer);
171
            } else {
172
                qemu_bh_cancel(q->tx_bh);
173
            }
174
        }
175
    }
176
}
177

    
178
static void virtio_net_set_link_status(NetClientState *nc)
179
{
180
    VirtIONet *n = qemu_get_nic_opaque(nc);
181
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
182
    uint16_t old_status = n->status;
183

    
184
    if (nc->link_down)
185
        n->status &= ~VIRTIO_NET_S_LINK_UP;
186
    else
187
        n->status |= VIRTIO_NET_S_LINK_UP;
188

    
189
    if (n->status != old_status)
190
        virtio_notify_config(vdev);
191

    
192
    virtio_net_set_status(vdev, vdev->status);
193
}
194

    
195
static void virtio_net_reset(VirtIODevice *vdev)
196
{
197
    VirtIONet *n = VIRTIO_NET(vdev);
198

    
199
    /* Reset back to compatibility mode */
200
    n->promisc = 1;
201
    n->allmulti = 0;
202
    n->alluni = 0;
203
    n->nomulti = 0;
204
    n->nouni = 0;
205
    n->nobcast = 0;
206
    /* multiqueue is disabled by default */
207
    n->curr_queues = 1;
208

    
209
    /* Flush any MAC and VLAN filter table state */
210
    n->mac_table.in_use = 0;
211
    n->mac_table.first_multi = 0;
212
    n->mac_table.multi_overflow = 0;
213
    n->mac_table.uni_overflow = 0;
214
    memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
215
    memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
216
    memset(n->vlans, 0, MAX_VLAN >> 3);
217
}
218

    
219
static void peer_test_vnet_hdr(VirtIONet *n)
220
{
221
    NetClientState *nc = qemu_get_queue(n->nic);
222
    if (!nc->peer) {
223
        return;
224
    }
225

    
226
    if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
227
        return;
228
    }
229

    
230
    n->has_vnet_hdr = tap_has_vnet_hdr(nc->peer);
231
}
232

    
233
static int peer_has_vnet_hdr(VirtIONet *n)
234
{
235
    return n->has_vnet_hdr;
236
}
237

    
238
static int peer_has_ufo(VirtIONet *n)
239
{
240
    if (!peer_has_vnet_hdr(n))
241
        return 0;
242

    
243
    n->has_ufo = tap_has_ufo(qemu_get_queue(n->nic)->peer);
244

    
245
    return n->has_ufo;
246
}
247

    
248
static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs)
249
{
250
    int i;
251
    NetClientState *nc;
252

    
253
    n->mergeable_rx_bufs = mergeable_rx_bufs;
254

    
255
    n->guest_hdr_len = n->mergeable_rx_bufs ?
256
        sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
257

    
258
    for (i = 0; i < n->max_queues; i++) {
259
        nc = qemu_get_subqueue(n->nic, i);
260

    
261
        if (peer_has_vnet_hdr(n) &&
262
            tap_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
263
            tap_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
264
            n->host_hdr_len = n->guest_hdr_len;
265
        }
266
    }
267
}
268

    
269
static int peer_attach(VirtIONet *n, int index)
270
{
271
    NetClientState *nc = qemu_get_subqueue(n->nic, index);
272

    
273
    if (!nc->peer) {
274
        return 0;
275
    }
276

    
277
    if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
278
        return 0;
279
    }
280

    
281
    return tap_enable(nc->peer);
282
}
283

    
284
static int peer_detach(VirtIONet *n, int index)
285
{
286
    NetClientState *nc = qemu_get_subqueue(n->nic, index);
287

    
288
    if (!nc->peer) {
289
        return 0;
290
    }
291

    
292
    if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
293
        return 0;
294
    }
295

    
296
    return tap_disable(nc->peer);
297
}
298

    
299
static void virtio_net_set_queues(VirtIONet *n)
300
{
301
    int i;
302

    
303
    for (i = 0; i < n->max_queues; i++) {
304
        if (i < n->curr_queues) {
305
            assert(!peer_attach(n, i));
306
        } else {
307
            assert(!peer_detach(n, i));
308
        }
309
    }
310
}
311

    
312
static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl);
313

    
314
static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
315
{
316
    VirtIONet *n = VIRTIO_NET(vdev);
317
    NetClientState *nc = qemu_get_queue(n->nic);
318

    
319
    features |= (1 << VIRTIO_NET_F_MAC);
320

    
321
    if (!peer_has_vnet_hdr(n)) {
322
        features &= ~(0x1 << VIRTIO_NET_F_CSUM);
323
        features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
324
        features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
325
        features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
326

    
327
        features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
328
        features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
329
        features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
330
        features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
331
    }
332

    
333
    if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
334
        features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
335
        features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
336
    }
337

    
338
    if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
339
        return features;
340
    }
341
    if (!tap_get_vhost_net(nc->peer)) {
342
        return features;
343
    }
344
    return vhost_net_get_features(tap_get_vhost_net(nc->peer), features);
345
}
346

    
347
static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
348
{
349
    uint32_t features = 0;
350

    
351
    /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
352
     * but also these: */
353
    features |= (1 << VIRTIO_NET_F_MAC);
354
    features |= (1 << VIRTIO_NET_F_CSUM);
355
    features |= (1 << VIRTIO_NET_F_HOST_TSO4);
356
    features |= (1 << VIRTIO_NET_F_HOST_TSO6);
357
    features |= (1 << VIRTIO_NET_F_HOST_ECN);
358

    
359
    return features;
360
}
361

    
362
static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
363
{
364
    VirtIONet *n = VIRTIO_NET(vdev);
365
    int i;
366

    
367
    virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)),
368
                              !!(features & (1 << VIRTIO_NET_F_CTRL_VQ)));
369

    
370
    virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
371

    
372
    if (n->has_vnet_hdr) {
373
        tap_set_offload(qemu_get_subqueue(n->nic, 0)->peer,
374
                        (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
375
                        (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
376
                        (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
377
                        (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
378
                        (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
379
    }
380

    
381
    for (i = 0;  i < n->max_queues; i++) {
382
        NetClientState *nc = qemu_get_subqueue(n->nic, i);
383

    
384
        if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
385
            continue;
386
        }
387
        if (!tap_get_vhost_net(nc->peer)) {
388
            continue;
389
        }
390
        vhost_net_ack_features(tap_get_vhost_net(nc->peer), features);
391
    }
392
}
393

    
394
static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
395
                                     struct iovec *iov, unsigned int iov_cnt)
396
{
397
    uint8_t on;
398
    size_t s;
399

    
400
    s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
401
    if (s != sizeof(on)) {
402
        return VIRTIO_NET_ERR;
403
    }
404

    
405
    if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
406
        n->promisc = on;
407
    } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
408
        n->allmulti = on;
409
    } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
410
        n->alluni = on;
411
    } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
412
        n->nomulti = on;
413
    } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
414
        n->nouni = on;
415
    } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
416
        n->nobcast = on;
417
    } else {
418
        return VIRTIO_NET_ERR;
419
    }
420

    
421
    return VIRTIO_NET_OK;
422
}
423

    
424
static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
425
                                 struct iovec *iov, unsigned int iov_cnt)
426
{
427
    struct virtio_net_ctrl_mac mac_data;
428
    size_t s;
429

    
430
    if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
431
        if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
432
            return VIRTIO_NET_ERR;
433
        }
434
        s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
435
        assert(s == sizeof(n->mac));
436
        qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
437
        return VIRTIO_NET_OK;
438
    }
439

    
440
    if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
441
        return VIRTIO_NET_ERR;
442
    }
443

    
444
    n->mac_table.in_use = 0;
445
    n->mac_table.first_multi = 0;
446
    n->mac_table.uni_overflow = 0;
447
    n->mac_table.multi_overflow = 0;
448
    memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
449

    
450
    s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
451
                   sizeof(mac_data.entries));
452
    mac_data.entries = ldl_p(&mac_data.entries);
453
    if (s != sizeof(mac_data.entries)) {
454
        return VIRTIO_NET_ERR;
455
    }
456
    iov_discard_front(&iov, &iov_cnt, s);
457

    
458
    if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
459
        return VIRTIO_NET_ERR;
460
    }
461

    
462
    if (mac_data.entries <= MAC_TABLE_ENTRIES) {
463
        s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
464
                       mac_data.entries * ETH_ALEN);
465
        if (s != mac_data.entries * ETH_ALEN) {
466
            return VIRTIO_NET_ERR;
467
        }
468
        n->mac_table.in_use += mac_data.entries;
469
    } else {
470
        n->mac_table.uni_overflow = 1;
471
    }
472

    
473
    iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
474

    
475
    n->mac_table.first_multi = n->mac_table.in_use;
476

    
477
    s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
478
                   sizeof(mac_data.entries));
479
    mac_data.entries = ldl_p(&mac_data.entries);
480
    if (s != sizeof(mac_data.entries)) {
481
        return VIRTIO_NET_ERR;
482
    }
483

    
484
    iov_discard_front(&iov, &iov_cnt, s);
485

    
486
    if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
487
        return VIRTIO_NET_ERR;
488
    }
489

    
490
    if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
491
        s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
492
                       mac_data.entries * ETH_ALEN);
493
        if (s != mac_data.entries * ETH_ALEN) {
494
            return VIRTIO_NET_ERR;
495
        }
496
        n->mac_table.in_use += mac_data.entries;
497
    } else {
498
        n->mac_table.multi_overflow = 1;
499
    }
500

    
501
    return VIRTIO_NET_OK;
502
}
503

    
504
static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
505
                                        struct iovec *iov, unsigned int iov_cnt)
506
{
507
    uint16_t vid;
508
    size_t s;
509

    
510
    s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
511
    vid = lduw_p(&vid);
512
    if (s != sizeof(vid)) {
513
        return VIRTIO_NET_ERR;
514
    }
515

    
516
    if (vid >= MAX_VLAN)
517
        return VIRTIO_NET_ERR;
518

    
519
    if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
520
        n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
521
    else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
522
        n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
523
    else
524
        return VIRTIO_NET_ERR;
525

    
526
    return VIRTIO_NET_OK;
527
}
528

    
529
static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
530
                                struct iovec *iov, unsigned int iov_cnt)
531
{
532
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
533
    struct virtio_net_ctrl_mq mq;
534
    size_t s;
535
    uint16_t queues;
536

    
537
    s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
538
    if (s != sizeof(mq)) {
539
        return VIRTIO_NET_ERR;
540
    }
541

    
542
    if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
543
        return VIRTIO_NET_ERR;
544
    }
545

    
546
    queues = lduw_p(&mq.virtqueue_pairs);
547

    
548
    if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
549
        queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
550
        queues > n->max_queues ||
551
        !n->multiqueue) {
552
        return VIRTIO_NET_ERR;
553
    }
554

    
555
    n->curr_queues = queues;
556
    /* stop the backend before changing the number of queues to avoid handling a
557
     * disabled queue */
558
    virtio_net_set_status(vdev, vdev->status);
559
    virtio_net_set_queues(n);
560

    
561
    return VIRTIO_NET_OK;
562
}
563
static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
564
{
565
    VirtIONet *n = VIRTIO_NET(vdev);
566
    struct virtio_net_ctrl_hdr ctrl;
567
    virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
568
    VirtQueueElement elem;
569
    size_t s;
570
    struct iovec *iov;
571
    unsigned int iov_cnt;
572

    
573
    while (virtqueue_pop(vq, &elem)) {
574
        if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
575
            iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
576
            error_report("virtio-net ctrl missing headers");
577
            exit(1);
578
        }
579

    
580
        iov = elem.out_sg;
581
        iov_cnt = elem.out_num;
582
        s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
583
        iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
584
        if (s != sizeof(ctrl)) {
585
            status = VIRTIO_NET_ERR;
586
        } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
587
            status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
588
        } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
589
            status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
590
        } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
591
            status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
592
        } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
593
            status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
594
        }
595

    
596
        s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
597
        assert(s == sizeof(status));
598

    
599
        virtqueue_push(vq, &elem, sizeof(status));
600
        virtio_notify(vdev, vq);
601
    }
602
}
603

    
604
/* RX */
605

    
606
static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
607
{
608
    VirtIONet *n = VIRTIO_NET(vdev);
609
    int queue_index = vq2q(virtio_get_queue_index(vq));
610

    
611
    qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
612
}
613

    
614
static int virtio_net_can_receive(NetClientState *nc)
615
{
616
    VirtIONet *n = qemu_get_nic_opaque(nc);
617
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
618
    VirtIONetQueue *q = virtio_net_get_subqueue(nc);
619

    
620
    if (!vdev->vm_running) {
621
        return 0;
622
    }
623

    
624
    if (nc->queue_index >= n->curr_queues) {
625
        return 0;
626
    }
627

    
628
    if (!virtio_queue_ready(q->rx_vq) ||
629
        !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
630
        return 0;
631
    }
632

    
633
    return 1;
634
}
635

    
636
static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
637
{
638
    VirtIONet *n = q->n;
639
    if (virtio_queue_empty(q->rx_vq) ||
640
        (n->mergeable_rx_bufs &&
641
         !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
642
        virtio_queue_set_notification(q->rx_vq, 1);
643

    
644
        /* To avoid a race condition where the guest has made some buffers
645
         * available after the above check but before notification was
646
         * enabled, check for available buffers again.
647
         */
648
        if (virtio_queue_empty(q->rx_vq) ||
649
            (n->mergeable_rx_bufs &&
650
             !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
651
            return 0;
652
        }
653
    }
654

    
655
    virtio_queue_set_notification(q->rx_vq, 0);
656
    return 1;
657
}
658

    
659
/* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
660
 * it never finds out that the packets don't have valid checksums.  This
661
 * causes dhclient to get upset.  Fedora's carried a patch for ages to
662
 * fix this with Xen but it hasn't appeared in an upstream release of
663
 * dhclient yet.
664
 *
665
 * To avoid breaking existing guests, we catch udp packets and add
666
 * checksums.  This is terrible but it's better than hacking the guest
667
 * kernels.
668
 *
669
 * N.B. if we introduce a zero-copy API, this operation is no longer free so
670
 * we should provide a mechanism to disable it to avoid polluting the host
671
 * cache.
672
 */
673
static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
674
                                        uint8_t *buf, size_t size)
675
{
676
    if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
677
        (size > 27 && size < 1500) && /* normal sized MTU */
678
        (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
679
        (buf[23] == 17) && /* ip.protocol == UDP */
680
        (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
681
        net_checksum_calculate(buf, size);
682
        hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
683
    }
684
}
685

    
686
static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
687
                           const void *buf, size_t size)
688
{
689
    if (n->has_vnet_hdr) {
690
        /* FIXME this cast is evil */
691
        void *wbuf = (void *)buf;
692
        work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
693
                                    size - n->host_hdr_len);
694
        iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
695
    } else {
696
        struct virtio_net_hdr hdr = {
697
            .flags = 0,
698
            .gso_type = VIRTIO_NET_HDR_GSO_NONE
699
        };
700
        iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
701
    }
702
}
703

    
704
static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
705
{
706
    static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
707
    static const uint8_t vlan[] = {0x81, 0x00};
708
    uint8_t *ptr = (uint8_t *)buf;
709
    int i;
710

    
711
    if (n->promisc)
712
        return 1;
713

    
714
    ptr += n->host_hdr_len;
715

    
716
    if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
717
        int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
718
        if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
719
            return 0;
720
    }
721

    
722
    if (ptr[0] & 1) { // multicast
723
        if (!memcmp(ptr, bcast, sizeof(bcast))) {
724
            return !n->nobcast;
725
        } else if (n->nomulti) {
726
            return 0;
727
        } else if (n->allmulti || n->mac_table.multi_overflow) {
728
            return 1;
729
        }
730

    
731
        for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
732
            if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
733
                return 1;
734
            }
735
        }
736
    } else { // unicast
737
        if (n->nouni) {
738
            return 0;
739
        } else if (n->alluni || n->mac_table.uni_overflow) {
740
            return 1;
741
        } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
742
            return 1;
743
        }
744

    
745
        for (i = 0; i < n->mac_table.first_multi; i++) {
746
            if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
747
                return 1;
748
            }
749
        }
750
    }
751

    
752
    return 0;
753
}
754

    
755
static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
756
{
757
    VirtIONet *n = qemu_get_nic_opaque(nc);
758
    VirtIONetQueue *q = virtio_net_get_subqueue(nc);
759
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
760
    struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
761
    struct virtio_net_hdr_mrg_rxbuf mhdr;
762
    unsigned mhdr_cnt = 0;
763
    size_t offset, i, guest_offset;
764

    
765
    if (!virtio_net_can_receive(nc)) {
766
        return -1;
767
    }
768

    
769
    /* hdr_len refers to the header we supply to the guest */
770
    if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
771
        return 0;
772
    }
773

    
774
    if (!receive_filter(n, buf, size))
775
        return size;
776

    
777
    offset = i = 0;
778

    
779
    while (offset < size) {
780
        VirtQueueElement elem;
781
        int len, total;
782
        const struct iovec *sg = elem.in_sg;
783

    
784
        total = 0;
785

    
786
        if (virtqueue_pop(q->rx_vq, &elem) == 0) {
787
            if (i == 0)
788
                return -1;
789
            error_report("virtio-net unexpected empty queue: "
790
                    "i %zd mergeable %d offset %zd, size %zd, "
791
                    "guest hdr len %zd, host hdr len %zd guest features 0x%x",
792
                    i, n->mergeable_rx_bufs, offset, size,
793
                    n->guest_hdr_len, n->host_hdr_len, vdev->guest_features);
794
            exit(1);
795
        }
796

    
797
        if (elem.in_num < 1) {
798
            error_report("virtio-net receive queue contains no in buffers");
799
            exit(1);
800
        }
801

    
802
        if (i == 0) {
803
            assert(offset == 0);
804
            if (n->mergeable_rx_bufs) {
805
                mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
806
                                    sg, elem.in_num,
807
                                    offsetof(typeof(mhdr), num_buffers),
808
                                    sizeof(mhdr.num_buffers));
809
            }
810

    
811
            receive_header(n, sg, elem.in_num, buf, size);
812
            offset = n->host_hdr_len;
813
            total += n->guest_hdr_len;
814
            guest_offset = n->guest_hdr_len;
815
        } else {
816
            guest_offset = 0;
817
        }
818

    
819
        /* copy in packet.  ugh */
820
        len = iov_from_buf(sg, elem.in_num, guest_offset,
821
                           buf + offset, size - offset);
822
        total += len;
823
        offset += len;
824
        /* If buffers can't be merged, at this point we
825
         * must have consumed the complete packet.
826
         * Otherwise, drop it. */
827
        if (!n->mergeable_rx_bufs && offset < size) {
828
#if 0
829
            error_report("virtio-net truncated non-mergeable packet: "
830
                         "i %zd mergeable %d offset %zd, size %zd, "
831
                         "guest hdr len %zd, host hdr len %zd",
832
                         i, n->mergeable_rx_bufs,
833
                         offset, size, n->guest_hdr_len, n->host_hdr_len);
834
#endif
835
            return size;
836
        }
837

    
838
        /* signal other side */
839
        virtqueue_fill(q->rx_vq, &elem, total, i++);
840
    }
841

    
842
    if (mhdr_cnt) {
843
        stw_p(&mhdr.num_buffers, i);
844
        iov_from_buf(mhdr_sg, mhdr_cnt,
845
                     0,
846
                     &mhdr.num_buffers, sizeof mhdr.num_buffers);
847
    }
848

    
849
    virtqueue_flush(q->rx_vq, i);
850
    virtio_notify(vdev, q->rx_vq);
851

    
852
    return size;
853
}
854

    
855
static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
856

    
857
static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
858
{
859
    VirtIONet *n = qemu_get_nic_opaque(nc);
860
    VirtIONetQueue *q = virtio_net_get_subqueue(nc);
861
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
862

    
863
    virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
864
    virtio_notify(vdev, q->tx_vq);
865

    
866
    q->async_tx.elem.out_num = q->async_tx.len = 0;
867

    
868
    virtio_queue_set_notification(q->tx_vq, 1);
869
    virtio_net_flush_tx(q);
870
}
871

    
872
/* TX */
873
static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
874
{
875
    VirtIONet *n = q->n;
876
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
877
    VirtQueueElement elem;
878
    int32_t num_packets = 0;
879
    int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
880
    if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
881
        return num_packets;
882
    }
883

    
884
    assert(vdev->vm_running);
885

    
886
    if (q->async_tx.elem.out_num) {
887
        virtio_queue_set_notification(q->tx_vq, 0);
888
        return num_packets;
889
    }
890

    
891
    while (virtqueue_pop(q->tx_vq, &elem)) {
892
        ssize_t ret, len;
893
        unsigned int out_num = elem.out_num;
894
        struct iovec *out_sg = &elem.out_sg[0];
895
        struct iovec sg[VIRTQUEUE_MAX_SIZE];
896

    
897
        if (out_num < 1) {
898
            error_report("virtio-net header not in first element");
899
            exit(1);
900
        }
901

    
902
        /*
903
         * If host wants to see the guest header as is, we can
904
         * pass it on unchanged. Otherwise, copy just the parts
905
         * that host is interested in.
906
         */
907
        assert(n->host_hdr_len <= n->guest_hdr_len);
908
        if (n->host_hdr_len != n->guest_hdr_len) {
909
            unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
910
                                       out_sg, out_num,
911
                                       0, n->host_hdr_len);
912
            sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
913
                             out_sg, out_num,
914
                             n->guest_hdr_len, -1);
915
            out_num = sg_num;
916
            out_sg = sg;
917
        }
918

    
919
        len = n->guest_hdr_len;
920

    
921
        ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
922
                                      out_sg, out_num, virtio_net_tx_complete);
923
        if (ret == 0) {
924
            virtio_queue_set_notification(q->tx_vq, 0);
925
            q->async_tx.elem = elem;
926
            q->async_tx.len  = len;
927
            return -EBUSY;
928
        }
929

    
930
        len += ret;
931

    
932
        virtqueue_push(q->tx_vq, &elem, 0);
933
        virtio_notify(vdev, q->tx_vq);
934

    
935
        if (++num_packets >= n->tx_burst) {
936
            break;
937
        }
938
    }
939
    return num_packets;
940
}
941

    
942
static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
943
{
944
    VirtIONet *n = VIRTIO_NET(vdev);
945
    VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
946

    
947
    /* This happens when device was stopped but VCPU wasn't. */
948
    if (!vdev->vm_running) {
949
        q->tx_waiting = 1;
950
        return;
951
    }
952

    
953
    if (q->tx_waiting) {
954
        virtio_queue_set_notification(vq, 1);
955
        qemu_del_timer(q->tx_timer);
956
        q->tx_waiting = 0;
957
        virtio_net_flush_tx(q);
958
    } else {
959
        qemu_mod_timer(q->tx_timer,
960
                       qemu_get_clock_ns(vm_clock) + n->tx_timeout);
961
        q->tx_waiting = 1;
962
        virtio_queue_set_notification(vq, 0);
963
    }
964
}
965

    
966
static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
967
{
968
    VirtIONet *n = VIRTIO_NET(vdev);
969
    VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
970

    
971
    if (unlikely(q->tx_waiting)) {
972
        return;
973
    }
974
    q->tx_waiting = 1;
975
    /* This happens when device was stopped but VCPU wasn't. */
976
    if (!vdev->vm_running) {
977
        return;
978
    }
979
    virtio_queue_set_notification(vq, 0);
980
    qemu_bh_schedule(q->tx_bh);
981
}
982

    
983
static void virtio_net_tx_timer(void *opaque)
984
{
985
    VirtIONetQueue *q = opaque;
986
    VirtIONet *n = q->n;
987
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
988
    assert(vdev->vm_running);
989

    
990
    q->tx_waiting = 0;
991

    
992
    /* Just in case the driver is not ready on more */
993
    if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
994
        return;
995
    }
996

    
997
    virtio_queue_set_notification(q->tx_vq, 1);
998
    virtio_net_flush_tx(q);
999
}
1000

    
1001
static void virtio_net_tx_bh(void *opaque)
1002
{
1003
    VirtIONetQueue *q = opaque;
1004
    VirtIONet *n = q->n;
1005
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
1006
    int32_t ret;
1007

    
1008
    assert(vdev->vm_running);
1009

    
1010
    q->tx_waiting = 0;
1011

    
1012
    /* Just in case the driver is not ready on more */
1013
    if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1014
        return;
1015
    }
1016

    
1017
    ret = virtio_net_flush_tx(q);
1018
    if (ret == -EBUSY) {
1019
        return; /* Notification re-enable handled by tx_complete */
1020
    }
1021

    
1022
    /* If we flush a full burst of packets, assume there are
1023
     * more coming and immediately reschedule */
1024
    if (ret >= n->tx_burst) {
1025
        qemu_bh_schedule(q->tx_bh);
1026
        q->tx_waiting = 1;
1027
        return;
1028
    }
1029

    
1030
    /* If less than a full burst, re-enable notification and flush
1031
     * anything that may have come in while we weren't looking.  If
1032
     * we find something, assume the guest is still active and reschedule */
1033
    virtio_queue_set_notification(q->tx_vq, 1);
1034
    if (virtio_net_flush_tx(q) > 0) {
1035
        virtio_queue_set_notification(q->tx_vq, 0);
1036
        qemu_bh_schedule(q->tx_bh);
1037
        q->tx_waiting = 1;
1038
    }
1039
}
1040

    
1041
static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl)
1042
{
1043
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
1044
    int i, max = multiqueue ? n->max_queues : 1;
1045

    
1046
    n->multiqueue = multiqueue;
1047

    
1048
    for (i = 2; i <= n->max_queues * 2 + 1; i++) {
1049
        virtio_del_queue(vdev, i);
1050
    }
1051

    
1052
    for (i = 1; i < max; i++) {
1053
        n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1054
        if (n->vqs[i].tx_timer) {
1055
            n->vqs[i].tx_vq =
1056
                virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1057
            n->vqs[i].tx_timer = qemu_new_timer_ns(vm_clock,
1058
                                                   virtio_net_tx_timer,
1059
                                                   &n->vqs[i]);
1060
        } else {
1061
            n->vqs[i].tx_vq =
1062
                virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1063
            n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1064
        }
1065

    
1066
        n->vqs[i].tx_waiting = 0;
1067
        n->vqs[i].n = n;
1068
    }
1069

    
1070
    if (ctrl) {
1071
        n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1072
    }
1073

    
1074
    virtio_net_set_queues(n);
1075
}
1076

    
1077
static void virtio_net_save(QEMUFile *f, void *opaque)
1078
{
1079
    int i;
1080
    VirtIONet *n = opaque;
1081
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
1082

    
1083
    /* At this point, backend must be stopped, otherwise
1084
     * it might keep writing to memory. */
1085
    assert(!n->vhost_started);
1086
    virtio_save(vdev, f);
1087

    
1088
    qemu_put_buffer(f, n->mac, ETH_ALEN);
1089
    qemu_put_be32(f, n->vqs[0].tx_waiting);
1090
    qemu_put_be32(f, n->mergeable_rx_bufs);
1091
    qemu_put_be16(f, n->status);
1092
    qemu_put_byte(f, n->promisc);
1093
    qemu_put_byte(f, n->allmulti);
1094
    qemu_put_be32(f, n->mac_table.in_use);
1095
    qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1096
    qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1097
    qemu_put_be32(f, n->has_vnet_hdr);
1098
    qemu_put_byte(f, n->mac_table.multi_overflow);
1099
    qemu_put_byte(f, n->mac_table.uni_overflow);
1100
    qemu_put_byte(f, n->alluni);
1101
    qemu_put_byte(f, n->nomulti);
1102
    qemu_put_byte(f, n->nouni);
1103
    qemu_put_byte(f, n->nobcast);
1104
    qemu_put_byte(f, n->has_ufo);
1105
    if (n->max_queues > 1) {
1106
        qemu_put_be16(f, n->max_queues);
1107
        qemu_put_be16(f, n->curr_queues);
1108
        for (i = 1; i < n->curr_queues; i++) {
1109
            qemu_put_be32(f, n->vqs[i].tx_waiting);
1110
        }
1111
    }
1112
}
1113

    
1114
static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1115
{
1116
    VirtIONet *n = opaque;
1117
    VirtIODevice *vdev = VIRTIO_DEVICE(n);
1118
    int ret, i, link_down;
1119

    
1120
    if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1121
        return -EINVAL;
1122

    
1123
    ret = virtio_load(vdev, f);
1124
    if (ret) {
1125
        return ret;
1126
    }
1127

    
1128
    qemu_get_buffer(f, n->mac, ETH_ALEN);
1129
    n->vqs[0].tx_waiting = qemu_get_be32(f);
1130

    
1131
    virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1132

    
1133
    if (version_id >= 3)
1134
        n->status = qemu_get_be16(f);
1135

    
1136
    if (version_id >= 4) {
1137
        if (version_id < 8) {
1138
            n->promisc = qemu_get_be32(f);
1139
            n->allmulti = qemu_get_be32(f);
1140
        } else {
1141
            n->promisc = qemu_get_byte(f);
1142
            n->allmulti = qemu_get_byte(f);
1143
        }
1144
    }
1145

    
1146
    if (version_id >= 5) {
1147
        n->mac_table.in_use = qemu_get_be32(f);
1148
        /* MAC_TABLE_ENTRIES may be different from the saved image */
1149
        if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1150
            qemu_get_buffer(f, n->mac_table.macs,
1151
                            n->mac_table.in_use * ETH_ALEN);
1152
        } else if (n->mac_table.in_use) {
1153
            uint8_t *buf = g_malloc0(n->mac_table.in_use);
1154
            qemu_get_buffer(f, buf, n->mac_table.in_use * ETH_ALEN);
1155
            g_free(buf);
1156
            n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1157
            n->mac_table.in_use = 0;
1158
        }
1159
    }
1160
 
1161
    if (version_id >= 6)
1162
        qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1163

    
1164
    if (version_id >= 7) {
1165
        if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1166
            error_report("virtio-net: saved image requires vnet_hdr=on");
1167
            return -1;
1168
        }
1169

    
1170
        if (n->has_vnet_hdr) {
1171
            tap_set_offload(qemu_get_queue(n->nic)->peer,
1172
                    (vdev->guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
1173
                    (vdev->guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
1174
                    (vdev->guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
1175
                    (vdev->guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
1176
                    (vdev->guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
1177
        }
1178
    }
1179

    
1180
    if (version_id >= 9) {
1181
        n->mac_table.multi_overflow = qemu_get_byte(f);
1182
        n->mac_table.uni_overflow = qemu_get_byte(f);
1183
    }
1184

    
1185
    if (version_id >= 10) {
1186
        n->alluni = qemu_get_byte(f);
1187
        n->nomulti = qemu_get_byte(f);
1188
        n->nouni = qemu_get_byte(f);
1189
        n->nobcast = qemu_get_byte(f);
1190
    }
1191

    
1192
    if (version_id >= 11) {
1193
        if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1194
            error_report("virtio-net: saved image requires TUN_F_UFO support");
1195
            return -1;
1196
        }
1197
    }
1198

    
1199
    if (n->max_queues > 1) {
1200
        if (n->max_queues != qemu_get_be16(f)) {
1201
            error_report("virtio-net: different max_queues ");
1202
            return -1;
1203
        }
1204

    
1205
        n->curr_queues = qemu_get_be16(f);
1206
        for (i = 1; i < n->curr_queues; i++) {
1207
            n->vqs[i].tx_waiting = qemu_get_be32(f);
1208
        }
1209
    }
1210

    
1211
    virtio_net_set_queues(n);
1212

    
1213
    /* Find the first multicast entry in the saved MAC filter */
1214
    for (i = 0; i < n->mac_table.in_use; i++) {
1215
        if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1216
            break;
1217
        }
1218
    }
1219
    n->mac_table.first_multi = i;
1220

    
1221
    /* nc.link_down can't be migrated, so infer link_down according
1222
     * to link status bit in n->status */
1223
    link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1224
    for (i = 0; i < n->max_queues; i++) {
1225
        qemu_get_subqueue(n->nic, i)->link_down = link_down;
1226
    }
1227

    
1228
    return 0;
1229
}
1230

    
1231
static void virtio_net_cleanup(NetClientState *nc)
1232
{
1233
    VirtIONet *n = qemu_get_nic_opaque(nc);
1234

    
1235
    n->nic = NULL;
1236
}
1237

    
1238
static NetClientInfo net_virtio_info = {
1239
    .type = NET_CLIENT_OPTIONS_KIND_NIC,
1240
    .size = sizeof(NICState),
1241
    .can_receive = virtio_net_can_receive,
1242
    .receive = virtio_net_receive,
1243
        .cleanup = virtio_net_cleanup,
1244
    .link_status_changed = virtio_net_set_link_status,
1245
};
1246

    
1247
static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1248
{
1249
    VirtIONet *n = VIRTIO_NET(vdev);
1250
    NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1251
    assert(n->vhost_started);
1252
    return vhost_net_virtqueue_pending(tap_get_vhost_net(nc->peer), idx);
1253
}
1254

    
1255
static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1256
                                           bool mask)
1257
{
1258
    VirtIONet *n = VIRTIO_NET(vdev);
1259
    NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1260
    assert(n->vhost_started);
1261
    vhost_net_virtqueue_mask(tap_get_vhost_net(nc->peer),
1262
                             vdev, idx, mask);
1263
}
1264

    
1265
void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features)
1266
{
1267
    int i, config_size = 0;
1268
    for (i = 0; feature_sizes[i].flags != 0; i++) {
1269
        if (host_features & feature_sizes[i].flags) {
1270
            config_size = MAX(feature_sizes[i].end, config_size);
1271
        }
1272
    }
1273
    n->config_size = config_size;
1274
}
1275

    
1276
static int virtio_net_device_init(VirtIODevice *vdev)
1277
{
1278
    int i;
1279

    
1280
    DeviceState *qdev = DEVICE(vdev);
1281
    VirtIONet *n = VIRTIO_NET(vdev);
1282

    
1283
    virtio_init(VIRTIO_DEVICE(n), "virtio-net", VIRTIO_ID_NET,
1284
                                  n->config_size);
1285

    
1286
    vdev->get_config = virtio_net_get_config;
1287
    vdev->set_config = virtio_net_set_config;
1288
    vdev->get_features = virtio_net_get_features;
1289
    vdev->set_features = virtio_net_set_features;
1290
    vdev->bad_features = virtio_net_bad_features;
1291
    vdev->reset = virtio_net_reset;
1292
    vdev->set_status = virtio_net_set_status;
1293
    vdev->guest_notifier_mask = virtio_net_guest_notifier_mask;
1294
    vdev->guest_notifier_pending = virtio_net_guest_notifier_pending;
1295
    n->max_queues = MAX(n->nic_conf.queues, 1);
1296
    n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1297
    n->vqs[0].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1298
    n->curr_queues = 1;
1299
    n->vqs[0].n = n;
1300
    n->tx_timeout = n->net_conf.txtimer;
1301

    
1302
    if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1303
                       && strcmp(n->net_conf.tx, "bh")) {
1304
        error_report("virtio-net: "
1305
                     "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1306
                     n->net_conf.tx);
1307
        error_report("Defaulting to \"bh\"");
1308
    }
1309

    
1310
    if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1311
        n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1312
                                           virtio_net_handle_tx_timer);
1313
        n->vqs[0].tx_timer = qemu_new_timer_ns(vm_clock, virtio_net_tx_timer,
1314
                                               &n->vqs[0]);
1315
    } else {
1316
        n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1317
                                           virtio_net_handle_tx_bh);
1318
        n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]);
1319
    }
1320
    n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1321
    qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1322
    memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1323
    n->status = VIRTIO_NET_S_LINK_UP;
1324

    
1325
    n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1326
                          object_get_typename(OBJECT(qdev)), qdev->id, n);
1327
    peer_test_vnet_hdr(n);
1328
    if (peer_has_vnet_hdr(n)) {
1329
        for (i = 0; i < n->max_queues; i++) {
1330
            tap_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1331
        }
1332
        n->host_hdr_len = sizeof(struct virtio_net_hdr);
1333
    } else {
1334
        n->host_hdr_len = 0;
1335
    }
1336

    
1337
    qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1338

    
1339
    n->vqs[0].tx_waiting = 0;
1340
    n->tx_burst = n->net_conf.txburst;
1341
    virtio_net_set_mrg_rx_bufs(n, 0);
1342
    n->promisc = 1; /* for compatibility */
1343

    
1344
    n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1345

    
1346
    n->vlans = g_malloc0(MAX_VLAN >> 3);
1347

    
1348
    n->qdev = qdev;
1349
    register_savevm(qdev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1350
                    virtio_net_save, virtio_net_load, n);
1351

    
1352
    add_boot_device_path(n->nic_conf.bootindex, qdev, "/ethernet-phy@0");
1353
    return 0;
1354
}
1355

    
1356
static int virtio_net_device_exit(DeviceState *qdev)
1357
{
1358
    VirtIONet *n = VIRTIO_NET(qdev);
1359
    VirtIODevice *vdev = VIRTIO_DEVICE(qdev);
1360
    int i;
1361

    
1362
    /* This will stop vhost backend if appropriate. */
1363
    virtio_net_set_status(vdev, 0);
1364

    
1365
    unregister_savevm(qdev, "virtio-net", n);
1366

    
1367
    g_free(n->mac_table.macs);
1368
    g_free(n->vlans);
1369

    
1370
    for (i = 0; i < n->max_queues; i++) {
1371
        VirtIONetQueue *q = &n->vqs[i];
1372
        NetClientState *nc = qemu_get_subqueue(n->nic, i);
1373

    
1374
        qemu_purge_queued_packets(nc);
1375

    
1376
        if (q->tx_timer) {
1377
            qemu_del_timer(q->tx_timer);
1378
            qemu_free_timer(q->tx_timer);
1379
        } else {
1380
            qemu_bh_delete(q->tx_bh);
1381
        }
1382
    }
1383

    
1384
    g_free(n->vqs);
1385
    qemu_del_nic(n->nic);
1386
    virtio_common_cleanup(vdev);
1387

    
1388
    return 0;
1389
}
1390

    
1391
static void virtio_net_instance_init(Object *obj)
1392
{
1393
    VirtIONet *n = VIRTIO_NET(obj);
1394

    
1395
    /*
1396
     * The default config_size is sizeof(struct virtio_net_config).
1397
     * Can be overriden with virtio_net_set_config_size.
1398
     */
1399
    n->config_size = sizeof(struct virtio_net_config);
1400
}
1401

    
1402
static Property virtio_net_properties[] = {
1403
    DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
1404
    DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
1405
                                               TX_TIMER_INTERVAL),
1406
    DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
1407
    DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
1408
    DEFINE_PROP_END_OF_LIST(),
1409
};
1410

    
1411
static void virtio_net_class_init(ObjectClass *klass, void *data)
1412
{
1413
    DeviceClass *dc = DEVICE_CLASS(klass);
1414
    VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1415
    dc->exit = virtio_net_device_exit;
1416
    dc->props = virtio_net_properties;
1417
    vdc->init = virtio_net_device_init;
1418
    vdc->get_config = virtio_net_get_config;
1419
    vdc->set_config = virtio_net_set_config;
1420
    vdc->get_features = virtio_net_get_features;
1421
    vdc->set_features = virtio_net_set_features;
1422
    vdc->bad_features = virtio_net_bad_features;
1423
    vdc->reset = virtio_net_reset;
1424
    vdc->set_status = virtio_net_set_status;
1425
    vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
1426
    vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
1427
}
1428

    
1429
static const TypeInfo virtio_net_info = {
1430
    .name = TYPE_VIRTIO_NET,
1431
    .parent = TYPE_VIRTIO_DEVICE,
1432
    .instance_size = sizeof(VirtIONet),
1433
    .instance_init = virtio_net_instance_init,
1434
    .class_init = virtio_net_class_init,
1435
};
1436

    
1437
static void virtio_register_types(void)
1438
{
1439
    type_register_static(&virtio_net_info);
1440
}
1441

    
1442
type_init(virtio_register_types)