Statistics
| Branch: | Revision:

root / hw / virtio-net.c @ 01657c86

History | View | Annotate | Download (27.8 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 "iov.h"
15
#include "virtio.h"
16
#include "net.h"
17
#include "net/checksum.h"
18
#include "net/tap.h"
19
#include "qemu-error.h"
20
#include "qemu-timer.h"
21
#include "virtio-net.h"
22
#include "vhost_net.h"
23

    
24
#define VIRTIO_NET_VM_VERSION    11
25

    
26
#define MAC_TABLE_ENTRIES    64
27
#define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
28

    
29
typedef struct VirtIONet
30
{
31
    VirtIODevice vdev;
32
    uint8_t mac[ETH_ALEN];
33
    uint16_t status;
34
    VirtQueue *rx_vq;
35
    VirtQueue *tx_vq;
36
    VirtQueue *ctrl_vq;
37
    NICState *nic;
38
    QEMUTimer *tx_timer;
39
    int tx_timer_active;
40
    uint32_t has_vnet_hdr;
41
    uint8_t has_ufo;
42
    struct {
43
        VirtQueueElement elem;
44
        ssize_t len;
45
    } async_tx;
46
    int mergeable_rx_bufs;
47
    uint8_t promisc;
48
    uint8_t allmulti;
49
    uint8_t alluni;
50
    uint8_t nomulti;
51
    uint8_t nouni;
52
    uint8_t nobcast;
53
    uint8_t vhost_started;
54
    VMChangeStateEntry *vmstate;
55
    struct {
56
        int in_use;
57
        int first_multi;
58
        uint8_t multi_overflow;
59
        uint8_t uni_overflow;
60
        uint8_t *macs;
61
    } mac_table;
62
    uint32_t *vlans;
63
    DeviceState *qdev;
64
} VirtIONet;
65

    
66
/* TODO
67
 * - we could suppress RX interrupt if we were so inclined.
68
 */
69

    
70
static VirtIONet *to_virtio_net(VirtIODevice *vdev)
71
{
72
    return (VirtIONet *)vdev;
73
}
74

    
75
static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
76
{
77
    VirtIONet *n = to_virtio_net(vdev);
78
    struct virtio_net_config netcfg;
79

    
80
    netcfg.status = n->status;
81
    memcpy(netcfg.mac, n->mac, ETH_ALEN);
82
    memcpy(config, &netcfg, sizeof(netcfg));
83
}
84

    
85
static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
86
{
87
    VirtIONet *n = to_virtio_net(vdev);
88
    struct virtio_net_config netcfg;
89

    
90
    memcpy(&netcfg, config, sizeof(netcfg));
91

    
92
    if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
93
        memcpy(n->mac, netcfg.mac, ETH_ALEN);
94
        qemu_format_nic_info_str(&n->nic->nc, n->mac);
95
    }
96
}
97

    
98
static void virtio_net_set_link_status(VLANClientState *nc)
99
{
100
    VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
101
    uint16_t old_status = n->status;
102

    
103
    if (nc->link_down)
104
        n->status &= ~VIRTIO_NET_S_LINK_UP;
105
    else
106
        n->status |= VIRTIO_NET_S_LINK_UP;
107

    
108
    if (n->status != old_status)
109
        virtio_notify_config(&n->vdev);
110
}
111

    
112
static void virtio_net_reset(VirtIODevice *vdev)
113
{
114
    VirtIONet *n = to_virtio_net(vdev);
115

    
116
    /* Reset back to compatibility mode */
117
    n->promisc = 1;
118
    n->allmulti = 0;
119
    n->alluni = 0;
120
    n->nomulti = 0;
121
    n->nouni = 0;
122
    n->nobcast = 0;
123
    if (n->vhost_started) {
124
        vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
125
        n->vhost_started = 0;
126
    }
127

    
128
    /* Flush any MAC and VLAN filter table state */
129
    n->mac_table.in_use = 0;
130
    n->mac_table.first_multi = 0;
131
    n->mac_table.multi_overflow = 0;
132
    n->mac_table.uni_overflow = 0;
133
    memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
134
    memset(n->vlans, 0, MAX_VLAN >> 3);
135
}
136

    
137
static int peer_has_vnet_hdr(VirtIONet *n)
138
{
139
    if (!n->nic->nc.peer)
140
        return 0;
141

    
142
    if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP)
143
        return 0;
144

    
145
    n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
146

    
147
    return n->has_vnet_hdr;
148
}
149

    
150
static int peer_has_ufo(VirtIONet *n)
151
{
152
    if (!peer_has_vnet_hdr(n))
153
        return 0;
154

    
155
    n->has_ufo = tap_has_ufo(n->nic->nc.peer);
156

    
157
    return n->has_ufo;
158
}
159

    
160
static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
161
{
162
    VirtIONet *n = to_virtio_net(vdev);
163

    
164
    features |= (1 << VIRTIO_NET_F_MAC);
165

    
166
    if (peer_has_vnet_hdr(n)) {
167
        tap_using_vnet_hdr(n->nic->nc.peer, 1);
168
    } else {
169
        features &= ~(0x1 << VIRTIO_NET_F_CSUM);
170
        features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
171
        features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
172
        features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
173

    
174
        features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
175
        features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
176
        features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
177
        features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
178
    }
179

    
180
    if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
181
        features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
182
        features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
183
    }
184

    
185
    if (!n->nic->nc.peer ||
186
        n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
187
        return features;
188
    }
189
    if (!tap_get_vhost_net(n->nic->nc.peer)) {
190
        return features;
191
    }
192
    return vhost_net_get_features(tap_get_vhost_net(n->nic->nc.peer), features);
193
}
194

    
195
static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
196
{
197
    uint32_t features = 0;
198

    
199
    /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
200
     * but also these: */
201
    features |= (1 << VIRTIO_NET_F_MAC);
202
    features |= (1 << VIRTIO_NET_F_CSUM);
203
    features |= (1 << VIRTIO_NET_F_HOST_TSO4);
204
    features |= (1 << VIRTIO_NET_F_HOST_TSO6);
205
    features |= (1 << VIRTIO_NET_F_HOST_ECN);
206

    
207
    return features;
208
}
209

    
210
static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
211
{
212
    VirtIONet *n = to_virtio_net(vdev);
213

    
214
    n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
215

    
216
    if (n->has_vnet_hdr) {
217
        tap_set_offload(n->nic->nc.peer,
218
                        (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
219
                        (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
220
                        (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
221
                        (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
222
                        (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
223
    }
224
    if (!n->nic->nc.peer ||
225
        n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
226
        return;
227
    }
228
    if (!tap_get_vhost_net(n->nic->nc.peer)) {
229
        return;
230
    }
231
    vhost_net_ack_features(tap_get_vhost_net(n->nic->nc.peer), features);
232
}
233

    
234
static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
235
                                     VirtQueueElement *elem)
236
{
237
    uint8_t on;
238

    
239
    if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
240
        fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
241
        exit(1);
242
    }
243

    
244
    on = ldub_p(elem->out_sg[1].iov_base);
245

    
246
    if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
247
        n->promisc = on;
248
    else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
249
        n->allmulti = on;
250
    else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
251
        n->alluni = on;
252
    else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
253
        n->nomulti = on;
254
    else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
255
        n->nouni = on;
256
    else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
257
        n->nobcast = on;
258
    else
259
        return VIRTIO_NET_ERR;
260

    
261
    return VIRTIO_NET_OK;
262
}
263

    
264
static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
265
                                 VirtQueueElement *elem)
266
{
267
    struct virtio_net_ctrl_mac mac_data;
268

    
269
    if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
270
        elem->out_sg[1].iov_len < sizeof(mac_data) ||
271
        elem->out_sg[2].iov_len < sizeof(mac_data))
272
        return VIRTIO_NET_ERR;
273

    
274
    n->mac_table.in_use = 0;
275
    n->mac_table.first_multi = 0;
276
    n->mac_table.uni_overflow = 0;
277
    n->mac_table.multi_overflow = 0;
278
    memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
279

    
280
    mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
281

    
282
    if (sizeof(mac_data.entries) +
283
        (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
284
        return VIRTIO_NET_ERR;
285

    
286
    if (mac_data.entries <= MAC_TABLE_ENTRIES) {
287
        memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
288
               mac_data.entries * ETH_ALEN);
289
        n->mac_table.in_use += mac_data.entries;
290
    } else {
291
        n->mac_table.uni_overflow = 1;
292
    }
293

    
294
    n->mac_table.first_multi = n->mac_table.in_use;
295

    
296
    mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
297

    
298
    if (sizeof(mac_data.entries) +
299
        (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
300
        return VIRTIO_NET_ERR;
301

    
302
    if (mac_data.entries) {
303
        if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
304
            memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
305
                   elem->out_sg[2].iov_base + sizeof(mac_data),
306
                   mac_data.entries * ETH_ALEN);
307
            n->mac_table.in_use += mac_data.entries;
308
        } else {
309
            n->mac_table.multi_overflow = 1;
310
        }
311
    }
312

    
313
    return VIRTIO_NET_OK;
314
}
315

    
316
static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
317
                                        VirtQueueElement *elem)
318
{
319
    uint16_t vid;
320

    
321
    if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
322
        fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
323
        return VIRTIO_NET_ERR;
324
    }
325

    
326
    vid = lduw_le_p(elem->out_sg[1].iov_base);
327

    
328
    if (vid >= MAX_VLAN)
329
        return VIRTIO_NET_ERR;
330

    
331
    if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
332
        n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
333
    else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
334
        n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
335
    else
336
        return VIRTIO_NET_ERR;
337

    
338
    return VIRTIO_NET_OK;
339
}
340

    
341
static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
342
{
343
    VirtIONet *n = to_virtio_net(vdev);
344
    struct virtio_net_ctrl_hdr ctrl;
345
    virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
346
    VirtQueueElement elem;
347

    
348
    while (virtqueue_pop(vq, &elem)) {
349
        if ((elem.in_num < 1) || (elem.out_num < 1)) {
350
            fprintf(stderr, "virtio-net ctrl missing headers\n");
351
            exit(1);
352
        }
353

    
354
        if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
355
            elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
356
            fprintf(stderr, "virtio-net ctrl header not in correct element\n");
357
            exit(1);
358
        }
359

    
360
        ctrl.class = ldub_p(elem.out_sg[0].iov_base);
361
        ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
362

    
363
        if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
364
            status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
365
        else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
366
            status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
367
        else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
368
            status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
369

    
370
        stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
371

    
372
        virtqueue_push(vq, &elem, sizeof(status));
373
        virtio_notify(vdev, vq);
374
    }
375
}
376

    
377
/* RX */
378

    
379
static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
380
{
381
    VirtIONet *n = to_virtio_net(vdev);
382

    
383
    qemu_flush_queued_packets(&n->nic->nc);
384

    
385
    /* We now have RX buffers, signal to the IO thread to break out of the
386
     * select to re-poll the tap file descriptor */
387
    qemu_notify_event();
388
}
389

    
390
static int virtio_net_can_receive(VLANClientState *nc)
391
{
392
    VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
393

    
394
    if (!virtio_queue_ready(n->rx_vq) ||
395
        !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
396
        return 0;
397

    
398
    return 1;
399
}
400

    
401
static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
402
{
403
    if (virtio_queue_empty(n->rx_vq) ||
404
        (n->mergeable_rx_bufs &&
405
         !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
406
        virtio_queue_set_notification(n->rx_vq, 1);
407

    
408
        /* To avoid a race condition where the guest has made some buffers
409
         * available after the above check but before notification was
410
         * enabled, check for available buffers again.
411
         */
412
        if (virtio_queue_empty(n->rx_vq) ||
413
            (n->mergeable_rx_bufs &&
414
             !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
415
            return 0;
416
    }
417

    
418
    virtio_queue_set_notification(n->rx_vq, 0);
419
    return 1;
420
}
421

    
422
/* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
423
 * it never finds out that the packets don't have valid checksums.  This
424
 * causes dhclient to get upset.  Fedora's carried a patch for ages to
425
 * fix this with Xen but it hasn't appeared in an upstream release of
426
 * dhclient yet.
427
 *
428
 * To avoid breaking existing guests, we catch udp packets and add
429
 * checksums.  This is terrible but it's better than hacking the guest
430
 * kernels.
431
 *
432
 * N.B. if we introduce a zero-copy API, this operation is no longer free so
433
 * we should provide a mechanism to disable it to avoid polluting the host
434
 * cache.
435
 */
436
static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
437
                                        const uint8_t *buf, size_t size)
438
{
439
    if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
440
        (size > 27 && size < 1500) && /* normal sized MTU */
441
        (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
442
        (buf[23] == 17) && /* ip.protocol == UDP */
443
        (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
444
        /* FIXME this cast is evil */
445
        net_checksum_calculate((uint8_t *)buf, size);
446
        hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
447
    }
448
}
449

    
450
static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
451
                          const void *buf, size_t size, size_t hdr_len)
452
{
453
    struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
454
    int offset = 0;
455

    
456
    hdr->flags = 0;
457
    hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
458

    
459
    if (n->has_vnet_hdr) {
460
        memcpy(hdr, buf, sizeof(*hdr));
461
        offset = sizeof(*hdr);
462
        work_around_broken_dhclient(hdr, buf + offset, size - offset);
463
    }
464

    
465
    /* We only ever receive a struct virtio_net_hdr from the tapfd,
466
     * but we may be passing along a larger header to the guest.
467
     */
468
    iov[0].iov_base += hdr_len;
469
    iov[0].iov_len  -= hdr_len;
470

    
471
    return offset;
472
}
473

    
474
static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
475
{
476
    static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
477
    static const uint8_t vlan[] = {0x81, 0x00};
478
    uint8_t *ptr = (uint8_t *)buf;
479
    int i;
480

    
481
    if (n->promisc)
482
        return 1;
483

    
484
    if (n->has_vnet_hdr) {
485
        ptr += sizeof(struct virtio_net_hdr);
486
    }
487

    
488
    if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
489
        int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
490
        if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
491
            return 0;
492
    }
493

    
494
    if (ptr[0] & 1) { // multicast
495
        if (!memcmp(ptr, bcast, sizeof(bcast))) {
496
            return !n->nobcast;
497
        } else if (n->nomulti) {
498
            return 0;
499
        } else if (n->allmulti || n->mac_table.multi_overflow) {
500
            return 1;
501
        }
502

    
503
        for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
504
            if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
505
                return 1;
506
            }
507
        }
508
    } else { // unicast
509
        if (n->nouni) {
510
            return 0;
511
        } else if (n->alluni || n->mac_table.uni_overflow) {
512
            return 1;
513
        } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
514
            return 1;
515
        }
516

    
517
        for (i = 0; i < n->mac_table.first_multi; i++) {
518
            if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
519
                return 1;
520
            }
521
        }
522
    }
523

    
524
    return 0;
525
}
526

    
527
static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
528
{
529
    VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
530
    struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
531
    size_t hdr_len, offset, i;
532

    
533
    if (!virtio_net_can_receive(&n->nic->nc))
534
        return -1;
535

    
536
    /* hdr_len refers to the header we supply to the guest */
537
    hdr_len = n->mergeable_rx_bufs ?
538
        sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
539

    
540

    
541
    if (!virtio_net_has_buffers(n, size + hdr_len))
542
        return 0;
543

    
544
    if (!receive_filter(n, buf, size))
545
        return size;
546

    
547
    offset = i = 0;
548

    
549
    while (offset < size) {
550
        VirtQueueElement elem;
551
        int len, total;
552
        struct iovec sg[VIRTQUEUE_MAX_SIZE];
553

    
554
        total = 0;
555

    
556
        if ((i != 0 && !n->mergeable_rx_bufs) ||
557
            virtqueue_pop(n->rx_vq, &elem) == 0) {
558
            if (i == 0)
559
                return -1;
560
            fprintf(stderr, "virtio-net truncating packet: "
561
                    "offset %zd, size %zd, hdr_len %zd\n",
562
                    offset, size, hdr_len);
563
            exit(1);
564
        }
565

    
566
        if (elem.in_num < 1) {
567
            fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
568
            exit(1);
569
        }
570

    
571
        if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != hdr_len) {
572
            fprintf(stderr, "virtio-net header not in first element\n");
573
            exit(1);
574
        }
575

    
576
        memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
577

    
578
        if (i == 0) {
579
            if (n->mergeable_rx_bufs)
580
                mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
581

    
582
            offset += receive_header(n, sg, elem.in_num,
583
                                     buf + offset, size - offset, hdr_len);
584
            total += hdr_len;
585
        }
586

    
587
        /* copy in packet.  ugh */
588
        len = iov_from_buf(sg, elem.in_num,
589
                           buf + offset, size - offset);
590
        total += len;
591

    
592
        /* signal other side */
593
        virtqueue_fill(n->rx_vq, &elem, total, i++);
594

    
595
        offset += len;
596
    }
597

    
598
    if (mhdr)
599
        mhdr->num_buffers = i;
600

    
601
    virtqueue_flush(n->rx_vq, i);
602
    virtio_notify(&n->vdev, n->rx_vq);
603

    
604
    return size;
605
}
606

    
607
static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
608

    
609
static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
610
{
611
    VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
612

    
613
    virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
614
    virtio_notify(&n->vdev, n->tx_vq);
615

    
616
    n->async_tx.elem.out_num = n->async_tx.len = 0;
617

    
618
    virtio_queue_set_notification(n->tx_vq, 1);
619
    virtio_net_flush_tx(n, n->tx_vq);
620
}
621

    
622
/* TX */
623
static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
624
{
625
    VirtQueueElement elem;
626

    
627
    if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
628
        return;
629

    
630
    if (n->async_tx.elem.out_num) {
631
        virtio_queue_set_notification(n->tx_vq, 0);
632
        return;
633
    }
634

    
635
    while (virtqueue_pop(vq, &elem)) {
636
        ssize_t ret, len = 0;
637
        unsigned int out_num = elem.out_num;
638
        struct iovec *out_sg = &elem.out_sg[0];
639
        unsigned hdr_len;
640

    
641
        /* hdr_len refers to the header received from the guest */
642
        hdr_len = n->mergeable_rx_bufs ?
643
            sizeof(struct virtio_net_hdr_mrg_rxbuf) :
644
            sizeof(struct virtio_net_hdr);
645

    
646
        if (out_num < 1 || out_sg->iov_len != hdr_len) {
647
            fprintf(stderr, "virtio-net header not in first element\n");
648
            exit(1);
649
        }
650

    
651
        /* ignore the header if GSO is not supported */
652
        if (!n->has_vnet_hdr) {
653
            out_num--;
654
            out_sg++;
655
            len += hdr_len;
656
        } else if (n->mergeable_rx_bufs) {
657
            /* tapfd expects a struct virtio_net_hdr */
658
            hdr_len -= sizeof(struct virtio_net_hdr);
659
            out_sg->iov_len -= hdr_len;
660
            len += hdr_len;
661
        }
662

    
663
        ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
664
                                      virtio_net_tx_complete);
665
        if (ret == 0) {
666
            virtio_queue_set_notification(n->tx_vq, 0);
667
            n->async_tx.elem = elem;
668
            n->async_tx.len  = len;
669
            return;
670
        }
671

    
672
        len += ret;
673

    
674
        virtqueue_push(vq, &elem, len);
675
        virtio_notify(&n->vdev, vq);
676
    }
677
}
678

    
679
static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
680
{
681
    VirtIONet *n = to_virtio_net(vdev);
682

    
683
    if (n->tx_timer_active) {
684
        virtio_queue_set_notification(vq, 1);
685
        qemu_del_timer(n->tx_timer);
686
        n->tx_timer_active = 0;
687
        virtio_net_flush_tx(n, vq);
688
    } else {
689
        qemu_mod_timer(n->tx_timer,
690
                       qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
691
        n->tx_timer_active = 1;
692
        virtio_queue_set_notification(vq, 0);
693
    }
694
}
695

    
696
static void virtio_net_tx_timer(void *opaque)
697
{
698
    VirtIONet *n = opaque;
699

    
700
    n->tx_timer_active = 0;
701

    
702
    /* Just in case the driver is not ready on more */
703
    if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
704
        return;
705

    
706
    virtio_queue_set_notification(n->tx_vq, 1);
707
    virtio_net_flush_tx(n, n->tx_vq);
708
}
709

    
710
static void virtio_net_save(QEMUFile *f, void *opaque)
711
{
712
    VirtIONet *n = opaque;
713

    
714
    if (n->vhost_started) {
715
        /* TODO: should we really stop the backend?
716
         * If we don't, it might keep writing to memory. */
717
        vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
718
        n->vhost_started = 0;
719
    }
720
    virtio_save(&n->vdev, f);
721

    
722
    qemu_put_buffer(f, n->mac, ETH_ALEN);
723
    qemu_put_be32(f, n->tx_timer_active);
724
    qemu_put_be32(f, n->mergeable_rx_bufs);
725
    qemu_put_be16(f, n->status);
726
    qemu_put_byte(f, n->promisc);
727
    qemu_put_byte(f, n->allmulti);
728
    qemu_put_be32(f, n->mac_table.in_use);
729
    qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
730
    qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
731
    qemu_put_be32(f, n->has_vnet_hdr);
732
    qemu_put_byte(f, n->mac_table.multi_overflow);
733
    qemu_put_byte(f, n->mac_table.uni_overflow);
734
    qemu_put_byte(f, n->alluni);
735
    qemu_put_byte(f, n->nomulti);
736
    qemu_put_byte(f, n->nouni);
737
    qemu_put_byte(f, n->nobcast);
738
    qemu_put_byte(f, n->has_ufo);
739
}
740

    
741
static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
742
{
743
    VirtIONet *n = opaque;
744
    int i;
745

    
746
    if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
747
        return -EINVAL;
748

    
749
    virtio_load(&n->vdev, f);
750

    
751
    qemu_get_buffer(f, n->mac, ETH_ALEN);
752
    n->tx_timer_active = qemu_get_be32(f);
753
    n->mergeable_rx_bufs = qemu_get_be32(f);
754

    
755
    if (version_id >= 3)
756
        n->status = qemu_get_be16(f);
757

    
758
    if (version_id >= 4) {
759
        if (version_id < 8) {
760
            n->promisc = qemu_get_be32(f);
761
            n->allmulti = qemu_get_be32(f);
762
        } else {
763
            n->promisc = qemu_get_byte(f);
764
            n->allmulti = qemu_get_byte(f);
765
        }
766
    }
767

    
768
    if (version_id >= 5) {
769
        n->mac_table.in_use = qemu_get_be32(f);
770
        /* MAC_TABLE_ENTRIES may be different from the saved image */
771
        if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
772
            qemu_get_buffer(f, n->mac_table.macs,
773
                            n->mac_table.in_use * ETH_ALEN);
774
        } else if (n->mac_table.in_use) {
775
            qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
776
            n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
777
            n->mac_table.in_use = 0;
778
        }
779
    }
780
 
781
    if (version_id >= 6)
782
        qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
783

    
784
    if (version_id >= 7) {
785
        if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
786
            error_report("virtio-net: saved image requires vnet_hdr=on");
787
            return -1;
788
        }
789

    
790
        if (n->has_vnet_hdr) {
791
            tap_using_vnet_hdr(n->nic->nc.peer, 1);
792
            tap_set_offload(n->nic->nc.peer,
793
                    (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
794
                    (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
795
                    (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
796
                    (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
797
                    (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
798
        }
799
    }
800

    
801
    if (version_id >= 9) {
802
        n->mac_table.multi_overflow = qemu_get_byte(f);
803
        n->mac_table.uni_overflow = qemu_get_byte(f);
804
    }
805

    
806
    if (version_id >= 10) {
807
        n->alluni = qemu_get_byte(f);
808
        n->nomulti = qemu_get_byte(f);
809
        n->nouni = qemu_get_byte(f);
810
        n->nobcast = qemu_get_byte(f);
811
    }
812

    
813
    if (version_id >= 11) {
814
        if (qemu_get_byte(f) && !peer_has_ufo(n)) {
815
            error_report("virtio-net: saved image requires TUN_F_UFO support");
816
            return -1;
817
        }
818
    }
819

    
820
    /* Find the first multicast entry in the saved MAC filter */
821
    for (i = 0; i < n->mac_table.in_use; i++) {
822
        if (n->mac_table.macs[i * ETH_ALEN] & 1) {
823
            break;
824
        }
825
    }
826
    n->mac_table.first_multi = i;
827

    
828
    if (n->tx_timer_active) {
829
        qemu_mod_timer(n->tx_timer,
830
                       qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
831
    }
832
    return 0;
833
}
834

    
835
static void virtio_net_cleanup(VLANClientState *nc)
836
{
837
    VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
838

    
839
    n->nic = NULL;
840
}
841

    
842
static NetClientInfo net_virtio_info = {
843
    .type = NET_CLIENT_TYPE_NIC,
844
    .size = sizeof(NICState),
845
    .can_receive = virtio_net_can_receive,
846
    .receive = virtio_net_receive,
847
        .cleanup = virtio_net_cleanup,
848
    .link_status_changed = virtio_net_set_link_status,
849
};
850

    
851
static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
852
{
853
    VirtIONet *n = to_virtio_net(vdev);
854
    if (!n->nic->nc.peer) {
855
        return;
856
    }
857
    if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
858
        return;
859
    }
860

    
861
    if (!tap_get_vhost_net(n->nic->nc.peer)) {
862
        return;
863
    }
864
    if (!!n->vhost_started == !!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
865
        return;
866
    }
867
    if (status & VIRTIO_CONFIG_S_DRIVER_OK) {
868
        int r = vhost_net_start(tap_get_vhost_net(n->nic->nc.peer), vdev);
869
        if (r < 0) {
870
            fprintf(stderr, "unable to start vhost net: %d: "
871
                    "falling back on userspace virtio\n", -r);
872
        } else {
873
            n->vhost_started = 1;
874
        }
875
    } else {
876
        vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
877
        n->vhost_started = 0;
878
    }
879
}
880

    
881
static void virtio_net_vmstate_change(void *opaque, int running, int reason)
882
{
883
    VirtIONet *n = opaque;
884
    uint8_t status = running ? VIRTIO_CONFIG_S_DRIVER_OK : 0;
885
    /* This is called when vm is started/stopped,
886
     * it will start/stop vhost backend if * appropriate
887
     * e.g. after migration. */
888
    virtio_net_set_status(&n->vdev, n->vdev.status & status);
889
}
890

    
891
VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf)
892
{
893
    VirtIONet *n;
894

    
895
    n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
896
                                        sizeof(struct virtio_net_config),
897
                                        sizeof(VirtIONet));
898

    
899
    n->vdev.get_config = virtio_net_get_config;
900
    n->vdev.set_config = virtio_net_set_config;
901
    n->vdev.get_features = virtio_net_get_features;
902
    n->vdev.set_features = virtio_net_set_features;
903
    n->vdev.bad_features = virtio_net_bad_features;
904
    n->vdev.reset = virtio_net_reset;
905
    n->vdev.set_status = virtio_net_set_status;
906
    n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
907
    n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
908
    n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
909
    qemu_macaddr_default_if_unset(&conf->macaddr);
910
    memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
911
    n->status = VIRTIO_NET_S_LINK_UP;
912

    
913
    n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
914

    
915
    qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
916

    
917
    n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
918
    n->tx_timer_active = 0;
919
    n->mergeable_rx_bufs = 0;
920
    n->promisc = 1; /* for compatibility */
921

    
922
    n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
923

    
924
    n->vlans = qemu_mallocz(MAX_VLAN >> 3);
925

    
926
    n->qdev = dev;
927
    register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
928
                    virtio_net_save, virtio_net_load, n);
929
    n->vmstate = qemu_add_vm_change_state_handler(virtio_net_vmstate_change, n);
930

    
931
    return &n->vdev;
932
}
933

    
934
void virtio_net_exit(VirtIODevice *vdev)
935
{
936
    VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
937
    qemu_del_vm_change_state_handler(n->vmstate);
938

    
939
    if (n->vhost_started) {
940
        vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
941
    }
942

    
943
    qemu_purge_queued_packets(&n->nic->nc);
944

    
945
    unregister_savevm(n->qdev, "virtio-net", n);
946

    
947
    qemu_free(n->mac_table.macs);
948
    qemu_free(n->vlans);
949

    
950
    qemu_del_timer(n->tx_timer);
951
    qemu_free_timer(n->tx_timer);
952

    
953
    virtio_cleanup(&n->vdev);
954
    qemu_del_vlan_client(&n->nic->nc);
955
}