1
2
3
4
5
6
7
8
9
10
11
12
13#include <linux/spinlock.h>
14#include <linux/module.h>
15#include <linux/list.h>
16#include <linux/atomic.h>
17#include <linux/virtio.h>
18#include <linux/virtio_ids.h>
19#include <linux/virtio_config.h>
20#include <linux/virtio_vsock.h>
21#include <net/sock.h>
22#include <linux/mutex.h>
23#include <net/af_vsock.h>
24
25static struct workqueue_struct *virtio_vsock_workqueue;
26static struct virtio_vsock *the_virtio_vsock;
27static DEFINE_MUTEX(the_virtio_vsock_mutex);
28
29struct virtio_vsock {
30 struct virtio_device *vdev;
31 struct virtqueue *vqs[VSOCK_VQ_MAX];
32
33
34 struct work_struct tx_work;
35 struct work_struct rx_work;
36 struct work_struct event_work;
37
38
39
40
41 struct mutex tx_lock;
42
43 struct work_struct send_pkt_work;
44 spinlock_t send_pkt_list_lock;
45 struct list_head send_pkt_list;
46
47 struct work_struct loopback_work;
48 spinlock_t loopback_list_lock;
49 struct list_head loopback_list;
50
51 atomic_t queued_replies;
52
53
54
55
56 struct mutex rx_lock;
57 int rx_buf_nr;
58 int rx_buf_max_nr;
59
60
61
62
63 struct mutex event_lock;
64 struct virtio_vsock_event event_list[8];
65
66 u32 guest_cid;
67};
68
69static struct virtio_vsock *virtio_vsock_get(void)
70{
71 return the_virtio_vsock;
72}
73
74static u32 virtio_transport_get_local_cid(void)
75{
76 struct virtio_vsock *vsock = virtio_vsock_get();
77
78 return vsock->guest_cid;
79}
80
81static void virtio_transport_loopback_work(struct work_struct *work)
82{
83 struct virtio_vsock *vsock =
84 container_of(work, struct virtio_vsock, loopback_work);
85 LIST_HEAD(pkts);
86
87 spin_lock_bh(&vsock->loopback_list_lock);
88 list_splice_init(&vsock->loopback_list, &pkts);
89 spin_unlock_bh(&vsock->loopback_list_lock);
90
91 mutex_lock(&vsock->rx_lock);
92 while (!list_empty(&pkts)) {
93 struct virtio_vsock_pkt *pkt;
94
95 pkt = list_first_entry(&pkts, struct virtio_vsock_pkt, list);
96 list_del_init(&pkt->list);
97
98 virtio_transport_recv_pkt(pkt);
99 }
100 mutex_unlock(&vsock->rx_lock);
101}
102
103static int virtio_transport_send_pkt_loopback(struct virtio_vsock *vsock,
104 struct virtio_vsock_pkt *pkt)
105{
106 int len = pkt->len;
107
108 spin_lock_bh(&vsock->loopback_list_lock);
109 list_add_tail(&pkt->list, &vsock->loopback_list);
110 spin_unlock_bh(&vsock->loopback_list_lock);
111
112 queue_work(virtio_vsock_workqueue, &vsock->loopback_work);
113
114 return len;
115}
116
117static void
118virtio_transport_send_pkt_work(struct work_struct *work)
119{
120 struct virtio_vsock *vsock =
121 container_of(work, struct virtio_vsock, send_pkt_work);
122 struct virtqueue *vq;
123 bool added = false;
124 bool restart_rx = false;
125
126 mutex_lock(&vsock->tx_lock);
127
128 vq = vsock->vqs[VSOCK_VQ_TX];
129
130 for (;;) {
131 struct virtio_vsock_pkt *pkt;
132 struct scatterlist hdr, buf, *sgs[2];
133 int ret, in_sg = 0, out_sg = 0;
134 bool reply;
135
136 spin_lock_bh(&vsock->send_pkt_list_lock);
137 if (list_empty(&vsock->send_pkt_list)) {
138 spin_unlock_bh(&vsock->send_pkt_list_lock);
139 break;
140 }
141
142 pkt = list_first_entry(&vsock->send_pkt_list,
143 struct virtio_vsock_pkt, list);
144 list_del_init(&pkt->list);
145 spin_unlock_bh(&vsock->send_pkt_list_lock);
146
147 virtio_transport_deliver_tap_pkt(pkt);
148
149 reply = pkt->reply;
150
151 sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr));
152 sgs[out_sg++] = &hdr;
153 if (pkt->buf) {
154 sg_init_one(&buf, pkt->buf, pkt->len);
155 sgs[out_sg++] = &buf;
156 }
157
158 ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, pkt, GFP_KERNEL);
159
160
161
162 if (ret < 0) {
163 spin_lock_bh(&vsock->send_pkt_list_lock);
164 list_add(&pkt->list, &vsock->send_pkt_list);
165 spin_unlock_bh(&vsock->send_pkt_list_lock);
166 break;
167 }
168
169 if (reply) {
170 struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
171 int val;
172
173 val = atomic_dec_return(&vsock->queued_replies);
174
175
176 if (val + 1 == virtqueue_get_vring_size(rx_vq))
177 restart_rx = true;
178 }
179
180 added = true;
181 }
182
183 if (added)
184 virtqueue_kick(vq);
185
186 mutex_unlock(&vsock->tx_lock);
187
188 if (restart_rx)
189 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
190}
191
192static int
193virtio_transport_send_pkt(struct virtio_vsock_pkt *pkt)
194{
195 struct virtio_vsock *vsock;
196 int len = pkt->len;
197
198 vsock = virtio_vsock_get();
199 if (!vsock) {
200 virtio_transport_free_pkt(pkt);
201 return -ENODEV;
202 }
203
204 if (le32_to_cpu(pkt->hdr.dst_cid) == vsock->guest_cid)
205 return virtio_transport_send_pkt_loopback(vsock, pkt);
206
207 if (pkt->reply)
208 atomic_inc(&vsock->queued_replies);
209
210 spin_lock_bh(&vsock->send_pkt_list_lock);
211 list_add_tail(&pkt->list, &vsock->send_pkt_list);
212 spin_unlock_bh(&vsock->send_pkt_list_lock);
213
214 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
215 return len;
216}
217
218static int
219virtio_transport_cancel_pkt(struct vsock_sock *vsk)
220{
221 struct virtio_vsock *vsock;
222 struct virtio_vsock_pkt *pkt, *n;
223 int cnt = 0;
224 LIST_HEAD(freeme);
225
226 vsock = virtio_vsock_get();
227 if (!vsock) {
228 return -ENODEV;
229 }
230
231 spin_lock_bh(&vsock->send_pkt_list_lock);
232 list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
233 if (pkt->vsk != vsk)
234 continue;
235 list_move(&pkt->list, &freeme);
236 }
237 spin_unlock_bh(&vsock->send_pkt_list_lock);
238
239 list_for_each_entry_safe(pkt, n, &freeme, list) {
240 if (pkt->reply)
241 cnt++;
242 list_del(&pkt->list);
243 virtio_transport_free_pkt(pkt);
244 }
245
246 if (cnt) {
247 struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
248 int new_cnt;
249
250 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
251 if (new_cnt + cnt >= virtqueue_get_vring_size(rx_vq) &&
252 new_cnt < virtqueue_get_vring_size(rx_vq))
253 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
254 }
255
256 return 0;
257}
258
259static void virtio_vsock_rx_fill(struct virtio_vsock *vsock)
260{
261 int buf_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE;
262 struct virtio_vsock_pkt *pkt;
263 struct scatterlist hdr, buf, *sgs[2];
264 struct virtqueue *vq;
265 int ret;
266
267 vq = vsock->vqs[VSOCK_VQ_RX];
268
269 do {
270 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
271 if (!pkt)
272 break;
273
274 pkt->buf = kmalloc(buf_len, GFP_KERNEL);
275 if (!pkt->buf) {
276 virtio_transport_free_pkt(pkt);
277 break;
278 }
279
280 pkt->len = buf_len;
281
282 sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr));
283 sgs[0] = &hdr;
284
285 sg_init_one(&buf, pkt->buf, buf_len);
286 sgs[1] = &buf;
287 ret = virtqueue_add_sgs(vq, sgs, 0, 2, pkt, GFP_KERNEL);
288 if (ret) {
289 virtio_transport_free_pkt(pkt);
290 break;
291 }
292 vsock->rx_buf_nr++;
293 } while (vq->num_free);
294 if (vsock->rx_buf_nr > vsock->rx_buf_max_nr)
295 vsock->rx_buf_max_nr = vsock->rx_buf_nr;
296 virtqueue_kick(vq);
297}
298
299static void virtio_transport_tx_work(struct work_struct *work)
300{
301 struct virtio_vsock *vsock =
302 container_of(work, struct virtio_vsock, tx_work);
303 struct virtqueue *vq;
304 bool added = false;
305
306 vq = vsock->vqs[VSOCK_VQ_TX];
307 mutex_lock(&vsock->tx_lock);
308 do {
309 struct virtio_vsock_pkt *pkt;
310 unsigned int len;
311
312 virtqueue_disable_cb(vq);
313 while ((pkt = virtqueue_get_buf(vq, &len)) != NULL) {
314 virtio_transport_free_pkt(pkt);
315 added = true;
316 }
317 } while (!virtqueue_enable_cb(vq));
318 mutex_unlock(&vsock->tx_lock);
319
320 if (added)
321 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
322}
323
324
325static bool virtio_transport_more_replies(struct virtio_vsock *vsock)
326{
327 struct virtqueue *vq = vsock->vqs[VSOCK_VQ_RX];
328 int val;
329
330 smp_rmb();
331 val = atomic_read(&vsock->queued_replies);
332
333 return val < virtqueue_get_vring_size(vq);
334}
335
336static void virtio_transport_rx_work(struct work_struct *work)
337{
338 struct virtio_vsock *vsock =
339 container_of(work, struct virtio_vsock, rx_work);
340 struct virtqueue *vq;
341
342 vq = vsock->vqs[VSOCK_VQ_RX];
343
344 mutex_lock(&vsock->rx_lock);
345
346 do {
347 virtqueue_disable_cb(vq);
348 for (;;) {
349 struct virtio_vsock_pkt *pkt;
350 unsigned int len;
351
352 if (!virtio_transport_more_replies(vsock)) {
353
354
355
356
357 goto out;
358 }
359
360 pkt = virtqueue_get_buf(vq, &len);
361 if (!pkt) {
362 break;
363 }
364
365 vsock->rx_buf_nr--;
366
367
368 if (unlikely(len < sizeof(pkt->hdr) ||
369 len > sizeof(pkt->hdr) + pkt->len)) {
370 virtio_transport_free_pkt(pkt);
371 continue;
372 }
373
374 pkt->len = len - sizeof(pkt->hdr);
375 virtio_transport_deliver_tap_pkt(pkt);
376 virtio_transport_recv_pkt(pkt);
377 }
378 } while (!virtqueue_enable_cb(vq));
379
380out:
381 if (vsock->rx_buf_nr < vsock->rx_buf_max_nr / 2)
382 virtio_vsock_rx_fill(vsock);
383 mutex_unlock(&vsock->rx_lock);
384}
385
386
387static int virtio_vsock_event_fill_one(struct virtio_vsock *vsock,
388 struct virtio_vsock_event *event)
389{
390 struct scatterlist sg;
391 struct virtqueue *vq;
392
393 vq = vsock->vqs[VSOCK_VQ_EVENT];
394
395 sg_init_one(&sg, event, sizeof(*event));
396
397 return virtqueue_add_inbuf(vq, &sg, 1, event, GFP_KERNEL);
398}
399
400
401static void virtio_vsock_event_fill(struct virtio_vsock *vsock)
402{
403 size_t i;
404
405 for (i = 0; i < ARRAY_SIZE(vsock->event_list); i++) {
406 struct virtio_vsock_event *event = &vsock->event_list[i];
407
408 virtio_vsock_event_fill_one(vsock, event);
409 }
410
411 virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
412}
413
414static void virtio_vsock_reset_sock(struct sock *sk)
415{
416 lock_sock(sk);
417 sk->sk_state = TCP_CLOSE;
418 sk->sk_err = ECONNRESET;
419 sk->sk_error_report(sk);
420 release_sock(sk);
421}
422
423static void virtio_vsock_update_guest_cid(struct virtio_vsock *vsock)
424{
425 struct virtio_device *vdev = vsock->vdev;
426 __le64 guest_cid;
427
428 vdev->config->get(vdev, offsetof(struct virtio_vsock_config, guest_cid),
429 &guest_cid, sizeof(guest_cid));
430 vsock->guest_cid = le64_to_cpu(guest_cid);
431}
432
433
434static void virtio_vsock_event_handle(struct virtio_vsock *vsock,
435 struct virtio_vsock_event *event)
436{
437 switch (le32_to_cpu(event->id)) {
438 case VIRTIO_VSOCK_EVENT_TRANSPORT_RESET:
439 virtio_vsock_update_guest_cid(vsock);
440 vsock_for_each_connected_socket(virtio_vsock_reset_sock);
441 break;
442 }
443}
444
445static void virtio_transport_event_work(struct work_struct *work)
446{
447 struct virtio_vsock *vsock =
448 container_of(work, struct virtio_vsock, event_work);
449 struct virtqueue *vq;
450
451 vq = vsock->vqs[VSOCK_VQ_EVENT];
452
453 mutex_lock(&vsock->event_lock);
454
455 do {
456 struct virtio_vsock_event *event;
457 unsigned int len;
458
459 virtqueue_disable_cb(vq);
460 while ((event = virtqueue_get_buf(vq, &len)) != NULL) {
461 if (len == sizeof(*event))
462 virtio_vsock_event_handle(vsock, event);
463
464 virtio_vsock_event_fill_one(vsock, event);
465 }
466 } while (!virtqueue_enable_cb(vq));
467
468 virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
469
470 mutex_unlock(&vsock->event_lock);
471}
472
473static void virtio_vsock_event_done(struct virtqueue *vq)
474{
475 struct virtio_vsock *vsock = vq->vdev->priv;
476
477 if (!vsock)
478 return;
479 queue_work(virtio_vsock_workqueue, &vsock->event_work);
480}
481
482static void virtio_vsock_tx_done(struct virtqueue *vq)
483{
484 struct virtio_vsock *vsock = vq->vdev->priv;
485
486 if (!vsock)
487 return;
488 queue_work(virtio_vsock_workqueue, &vsock->tx_work);
489}
490
491static void virtio_vsock_rx_done(struct virtqueue *vq)
492{
493 struct virtio_vsock *vsock = vq->vdev->priv;
494
495 if (!vsock)
496 return;
497 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
498}
499
500static struct virtio_transport virtio_transport = {
501 .transport = {
502 .get_local_cid = virtio_transport_get_local_cid,
503
504 .init = virtio_transport_do_socket_init,
505 .destruct = virtio_transport_destruct,
506 .release = virtio_transport_release,
507 .connect = virtio_transport_connect,
508 .shutdown = virtio_transport_shutdown,
509 .cancel_pkt = virtio_transport_cancel_pkt,
510
511 .dgram_bind = virtio_transport_dgram_bind,
512 .dgram_dequeue = virtio_transport_dgram_dequeue,
513 .dgram_enqueue = virtio_transport_dgram_enqueue,
514 .dgram_allow = virtio_transport_dgram_allow,
515
516 .stream_dequeue = virtio_transport_stream_dequeue,
517 .stream_enqueue = virtio_transport_stream_enqueue,
518 .stream_has_data = virtio_transport_stream_has_data,
519 .stream_has_space = virtio_transport_stream_has_space,
520 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
521 .stream_is_active = virtio_transport_stream_is_active,
522 .stream_allow = virtio_transport_stream_allow,
523
524 .notify_poll_in = virtio_transport_notify_poll_in,
525 .notify_poll_out = virtio_transport_notify_poll_out,
526 .notify_recv_init = virtio_transport_notify_recv_init,
527 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
528 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
529 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
530 .notify_send_init = virtio_transport_notify_send_init,
531 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
532 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
533 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
534
535 .set_buffer_size = virtio_transport_set_buffer_size,
536 .set_min_buffer_size = virtio_transport_set_min_buffer_size,
537 .set_max_buffer_size = virtio_transport_set_max_buffer_size,
538 .get_buffer_size = virtio_transport_get_buffer_size,
539 .get_min_buffer_size = virtio_transport_get_min_buffer_size,
540 .get_max_buffer_size = virtio_transport_get_max_buffer_size,
541 },
542
543 .send_pkt = virtio_transport_send_pkt,
544};
545
546static int virtio_vsock_probe(struct virtio_device *vdev)
547{
548 vq_callback_t *callbacks[] = {
549 virtio_vsock_rx_done,
550 virtio_vsock_tx_done,
551 virtio_vsock_event_done,
552 };
553 static const char * const names[] = {
554 "rx",
555 "tx",
556 "event",
557 };
558 struct virtio_vsock *vsock = NULL;
559 int ret;
560
561 ret = mutex_lock_interruptible(&the_virtio_vsock_mutex);
562 if (ret)
563 return ret;
564
565
566 if (the_virtio_vsock) {
567 ret = -EBUSY;
568 goto out;
569 }
570
571 vsock = kzalloc(sizeof(*vsock), GFP_KERNEL);
572 if (!vsock) {
573 ret = -ENOMEM;
574 goto out;
575 }
576
577 vsock->vdev = vdev;
578
579 ret = virtio_find_vqs(vsock->vdev, VSOCK_VQ_MAX,
580 vsock->vqs, callbacks, names,
581 NULL);
582 if (ret < 0)
583 goto out;
584
585 virtio_vsock_update_guest_cid(vsock);
586
587 ret = vsock_core_init(&virtio_transport.transport);
588 if (ret < 0)
589 goto out_vqs;
590
591 vsock->rx_buf_nr = 0;
592 vsock->rx_buf_max_nr = 0;
593 atomic_set(&vsock->queued_replies, 0);
594
595 vdev->priv = vsock;
596 the_virtio_vsock = vsock;
597 mutex_init(&vsock->tx_lock);
598 mutex_init(&vsock->rx_lock);
599 mutex_init(&vsock->event_lock);
600 spin_lock_init(&vsock->send_pkt_list_lock);
601 INIT_LIST_HEAD(&vsock->send_pkt_list);
602 spin_lock_init(&vsock->loopback_list_lock);
603 INIT_LIST_HEAD(&vsock->loopback_list);
604 INIT_WORK(&vsock->rx_work, virtio_transport_rx_work);
605 INIT_WORK(&vsock->tx_work, virtio_transport_tx_work);
606 INIT_WORK(&vsock->event_work, virtio_transport_event_work);
607 INIT_WORK(&vsock->send_pkt_work, virtio_transport_send_pkt_work);
608 INIT_WORK(&vsock->loopback_work, virtio_transport_loopback_work);
609
610 mutex_lock(&vsock->rx_lock);
611 virtio_vsock_rx_fill(vsock);
612 mutex_unlock(&vsock->rx_lock);
613
614 mutex_lock(&vsock->event_lock);
615 virtio_vsock_event_fill(vsock);
616 mutex_unlock(&vsock->event_lock);
617
618 mutex_unlock(&the_virtio_vsock_mutex);
619 return 0;
620
621out_vqs:
622 vsock->vdev->config->del_vqs(vsock->vdev);
623out:
624 kfree(vsock);
625 mutex_unlock(&the_virtio_vsock_mutex);
626 return ret;
627}
628
629static void virtio_vsock_remove(struct virtio_device *vdev)
630{
631 struct virtio_vsock *vsock = vdev->priv;
632 struct virtio_vsock_pkt *pkt;
633
634 flush_work(&vsock->loopback_work);
635 flush_work(&vsock->rx_work);
636 flush_work(&vsock->tx_work);
637 flush_work(&vsock->event_work);
638 flush_work(&vsock->send_pkt_work);
639
640 vdev->config->reset(vdev);
641
642 mutex_lock(&vsock->rx_lock);
643 while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX])))
644 virtio_transport_free_pkt(pkt);
645 mutex_unlock(&vsock->rx_lock);
646
647 mutex_lock(&vsock->tx_lock);
648 while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX])))
649 virtio_transport_free_pkt(pkt);
650 mutex_unlock(&vsock->tx_lock);
651
652 spin_lock_bh(&vsock->send_pkt_list_lock);
653 while (!list_empty(&vsock->send_pkt_list)) {
654 pkt = list_first_entry(&vsock->send_pkt_list,
655 struct virtio_vsock_pkt, list);
656 list_del(&pkt->list);
657 virtio_transport_free_pkt(pkt);
658 }
659 spin_unlock_bh(&vsock->send_pkt_list_lock);
660
661 spin_lock_bh(&vsock->loopback_list_lock);
662 while (!list_empty(&vsock->loopback_list)) {
663 pkt = list_first_entry(&vsock->loopback_list,
664 struct virtio_vsock_pkt, list);
665 list_del(&pkt->list);
666 virtio_transport_free_pkt(pkt);
667 }
668 spin_unlock_bh(&vsock->loopback_list_lock);
669
670 mutex_lock(&the_virtio_vsock_mutex);
671 the_virtio_vsock = NULL;
672 vsock_core_exit();
673 mutex_unlock(&the_virtio_vsock_mutex);
674
675 vdev->config->del_vqs(vdev);
676
677 kfree(vsock);
678}
679
680static struct virtio_device_id id_table[] = {
681 { VIRTIO_ID_VSOCK, VIRTIO_DEV_ANY_ID },
682 { 0 },
683};
684
685static unsigned int features[] = {
686};
687
688static struct virtio_driver virtio_vsock_driver = {
689 .feature_table = features,
690 .feature_table_size = ARRAY_SIZE(features),
691 .driver.name = KBUILD_MODNAME,
692 .driver.owner = THIS_MODULE,
693 .id_table = id_table,
694 .probe = virtio_vsock_probe,
695 .remove = virtio_vsock_remove,
696};
697
698static int __init virtio_vsock_init(void)
699{
700 int ret;
701
702 virtio_vsock_workqueue = alloc_workqueue("virtio_vsock", 0, 0);
703 if (!virtio_vsock_workqueue)
704 return -ENOMEM;
705 ret = register_virtio_driver(&virtio_vsock_driver);
706 if (ret)
707 destroy_workqueue(virtio_vsock_workqueue);
708 return ret;
709}
710
711static void __exit virtio_vsock_exit(void)
712{
713 unregister_virtio_driver(&virtio_vsock_driver);
714 destroy_workqueue(virtio_vsock_workqueue);
715}
716
717module_init(virtio_vsock_init);
718module_exit(virtio_vsock_exit);
719MODULE_LICENSE("GPL v2");
720MODULE_AUTHOR("Asias He");
721MODULE_DESCRIPTION("virtio transport for vsock");
722MODULE_DEVICE_TABLE(virtio, id_table);
723
