1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * virtio transport for vsock
4 *
5 * Copyright (C) 2013-2015 Red Hat, Inc.
6 * Author: Asias He <asias@redhat.com>
7 * Stefan Hajnoczi <stefanha@redhat.com>
8 *
9 * Some of the code is take from Gerd Hoffmann <kraxel@redhat.com>'s
10 * early virtio-vsock proof-of-concept bits.
11 */
12 #include <linux/spinlock.h>
13 #include <linux/module.h>
14 #include <linux/list.h>
15 #include <linux/atomic.h>
16 #include <linux/virtio.h>
17 #include <linux/virtio_ids.h>
18 #include <linux/virtio_config.h>
19 #include <linux/virtio_vsock.h>
20 #include <linux/dma-mapping.h>
21 #include <net/sock.h>
22 #include <linux/mutex.h>
23 #include <net/af_vsock.h>
24
25 static struct workqueue_struct *virtio_vsock_workqueue;
26 static struct virtio_vsock __rcu *the_virtio_vsock;
27 static DEFINE_MUTEX(the_virtio_vsock_mutex); /* protects the_virtio_vsock */
28 static struct virtio_transport virtio_transport; /* forward declaration */
29
30 struct virtio_vsock {
31 struct virtio_device *vdev;
32 struct virtqueue *vqs[VSOCK_VQ_MAX];
33
34 /* Virtqueue processing is deferred to a workqueue */
35 struct work_struct tx_work;
36 struct work_struct rx_work;
37 struct work_struct event_work;
38
39 /* The following fields are protected by tx_lock. vqs[VSOCK_VQ_TX]
40 * must be accessed with tx_lock held.
41 */
42 struct mutex tx_lock;
43 bool tx_run;
44
45 struct work_struct send_pkt_work;
46 struct sk_buff_head send_pkt_queue;
47
48 atomic_t queued_replies;
49
50 /* The following fields are protected by rx_lock. vqs[VSOCK_VQ_RX]
51 * must be accessed with rx_lock held.
52 */
53 struct mutex rx_lock;
54 bool rx_run;
55 int rx_buf_nr;
56 int rx_buf_max_nr;
57
58 u32 guest_cid;
59 bool seqpacket_allow;
60
61 /* These fields are used only in tx path in function
62 * 'virtio_transport_send_pkt_work()', so to save
63 * stack space in it, place both of them here. Each
64 * pointer from 'out_sgs' points to the corresponding
65 * element in 'out_bufs' - this is initialized in
66 * 'virtio_vsock_probe()'. Both fields are protected
67 * by 'tx_lock'. +1 is needed for packet header.
68 */
69 struct scatterlist *out_sgs[MAX_SKB_FRAGS + 1];
70 struct scatterlist out_bufs[MAX_SKB_FRAGS + 1];
71
72 /* The following fields are protected by event_lock.
73 * vqs[VSOCK_VQ_EVENT] must be accessed with event_lock held.
74 */
75 struct mutex event_lock;
76 bool event_run;
77 __dma_from_device_group_begin();
78 struct virtio_vsock_event event_list[8];
79 __dma_from_device_group_end();
80 };
81
virtio_transport_get_local_cid(void)82 static u32 virtio_transport_get_local_cid(void)
83 {
84 struct virtio_vsock *vsock;
85 u32 ret;
86
87 rcu_read_lock();
88 vsock = rcu_dereference(the_virtio_vsock);
89 if (!vsock) {
90 ret = VMADDR_CID_ANY;
91 goto out_rcu;
92 }
93
94 ret = vsock->guest_cid;
95 out_rcu:
96 rcu_read_unlock();
97 return ret;
98 }
99
100 /* Caller need to hold vsock->tx_lock on vq */
virtio_transport_send_skb(struct sk_buff * skb,struct virtqueue * vq,struct virtio_vsock * vsock,gfp_t gfp)101 static int virtio_transport_send_skb(struct sk_buff *skb, struct virtqueue *vq,
102 struct virtio_vsock *vsock, gfp_t gfp)
103 {
104 int ret, in_sg = 0, out_sg = 0;
105 struct scatterlist **sgs;
106
107 sgs = vsock->out_sgs;
108 sg_init_one(sgs[out_sg], virtio_vsock_hdr(skb),
109 sizeof(*virtio_vsock_hdr(skb)));
110 out_sg++;
111
112 if (!skb_is_nonlinear(skb)) {
113 if (skb->len > 0) {
114 sg_init_one(sgs[out_sg], skb->data, skb->len);
115 out_sg++;
116 }
117 } else {
118 struct skb_shared_info *si;
119 int i;
120
121 /* If skb is nonlinear, then its buffer must contain
122 * only header and nothing more. Data is stored in
123 * the fragged part.
124 */
125 WARN_ON_ONCE(skb_headroom(skb) != sizeof(*virtio_vsock_hdr(skb)));
126
127 si = skb_shinfo(skb);
128
129 for (i = 0; i < si->nr_frags; i++) {
130 skb_frag_t *skb_frag = &si->frags[i];
131 void *va;
132
133 /* We will use 'page_to_virt()' for the userspace page
134 * here, because virtio or dma-mapping layers will call
135 * 'virt_to_phys()' later to fill the buffer descriptor.
136 * We don't touch memory at "virtual" address of this page.
137 */
138 va = page_to_virt(skb_frag_page(skb_frag));
139 sg_init_one(sgs[out_sg],
140 va + skb_frag_off(skb_frag),
141 skb_frag_size(skb_frag));
142 out_sg++;
143 }
144 }
145
146 ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, skb, gfp);
147 /* Usually this means that there is no more space available in
148 * the vq
149 */
150 if (ret < 0)
151 return ret;
152
153 virtio_transport_deliver_tap_pkt(skb);
154 return 0;
155 }
156
157 static void
virtio_transport_send_pkt_work(struct work_struct * work)158 virtio_transport_send_pkt_work(struct work_struct *work)
159 {
160 struct virtio_vsock *vsock =
161 container_of(work, struct virtio_vsock, send_pkt_work);
162 struct virtqueue *vq;
163 bool added = false;
164 bool restart_rx = false;
165
166 mutex_lock(&vsock->tx_lock);
167
168 if (!vsock->tx_run)
169 goto out;
170
171 vq = vsock->vqs[VSOCK_VQ_TX];
172
173 for (;;) {
174 struct sk_buff *skb;
175 bool reply;
176 int ret;
177
178 skb = virtio_vsock_skb_dequeue(&vsock->send_pkt_queue);
179 if (!skb)
180 break;
181
182 reply = virtio_vsock_skb_reply(skb);
183
184 ret = virtio_transport_send_skb(skb, vq, vsock, GFP_KERNEL);
185 if (ret < 0) {
186 virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
187 break;
188 }
189
190 if (reply) {
191 struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
192 int val;
193
194 val = atomic_dec_return(&vsock->queued_replies);
195
196 /* Do we now have resources to resume rx processing? */
197 if (val + 1 == virtqueue_get_vring_size(rx_vq))
198 restart_rx = true;
199 }
200
201 added = true;
202 }
203
204 if (added)
205 virtqueue_kick(vq);
206
207 out:
208 mutex_unlock(&vsock->tx_lock);
209
210 if (restart_rx)
211 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
212 }
213
214 /* Caller need to hold RCU for vsock.
215 * Returns 0 if the packet is successfully put on the vq.
216 */
virtio_transport_send_skb_fast_path(struct virtio_vsock * vsock,struct sk_buff * skb)217 static int virtio_transport_send_skb_fast_path(struct virtio_vsock *vsock, struct sk_buff *skb)
218 {
219 struct virtqueue *vq = vsock->vqs[VSOCK_VQ_TX];
220 int ret;
221
222 /* Inside RCU, can't sleep! */
223 ret = mutex_trylock(&vsock->tx_lock);
224 if (unlikely(ret == 0))
225 return -EBUSY;
226
227 ret = virtio_transport_send_skb(skb, vq, vsock, GFP_ATOMIC);
228 if (ret == 0)
229 virtqueue_kick(vq);
230
231 mutex_unlock(&vsock->tx_lock);
232
233 return ret;
234 }
235
236 static int
virtio_transport_send_pkt(struct sk_buff * skb,struct net * net)237 virtio_transport_send_pkt(struct sk_buff *skb, struct net *net)
238 {
239 struct virtio_vsock_hdr *hdr;
240 struct virtio_vsock *vsock;
241 int len = skb->len;
242
243 hdr = virtio_vsock_hdr(skb);
244
245 rcu_read_lock();
246 vsock = rcu_dereference(the_virtio_vsock);
247 if (!vsock) {
248 kfree_skb(skb);
249 len = -ENODEV;
250 goto out_rcu;
251 }
252
253 if (le64_to_cpu(hdr->dst_cid) == vsock->guest_cid) {
254 kfree_skb(skb);
255 len = -ENODEV;
256 goto out_rcu;
257 }
258
259 /* If send_pkt_queue is empty, we can safely bypass this queue
260 * because packet order is maintained and (try) to put the packet
261 * on the virtqueue using virtio_transport_send_skb_fast_path.
262 * If this fails we simply put the packet on the intermediate
263 * queue and schedule the worker.
264 */
265 if (!skb_queue_empty_lockless(&vsock->send_pkt_queue) ||
266 virtio_transport_send_skb_fast_path(vsock, skb)) {
267 if (virtio_vsock_skb_reply(skb))
268 atomic_inc(&vsock->queued_replies);
269
270 virtio_vsock_skb_queue_tail(&vsock->send_pkt_queue, skb);
271 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
272 }
273
274 out_rcu:
275 rcu_read_unlock();
276 return len;
277 }
278
279 static int
virtio_transport_cancel_pkt(struct vsock_sock * vsk)280 virtio_transport_cancel_pkt(struct vsock_sock *vsk)
281 {
282 struct virtio_vsock *vsock;
283 int cnt = 0, ret;
284
285 rcu_read_lock();
286 vsock = rcu_dereference(the_virtio_vsock);
287 if (!vsock) {
288 ret = -ENODEV;
289 goto out_rcu;
290 }
291
292 cnt = virtio_transport_purge_skbs(vsk, &vsock->send_pkt_queue);
293
294 if (cnt) {
295 struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
296 int new_cnt;
297
298 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
299 if (new_cnt + cnt >= virtqueue_get_vring_size(rx_vq) &&
300 new_cnt < virtqueue_get_vring_size(rx_vq))
301 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
302 }
303
304 ret = 0;
305
306 out_rcu:
307 rcu_read_unlock();
308 return ret;
309 }
310
virtio_vsock_rx_fill(struct virtio_vsock * vsock)311 static void virtio_vsock_rx_fill(struct virtio_vsock *vsock)
312 {
313 int total_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE;
314 struct scatterlist pkt, *p;
315 struct virtqueue *vq;
316 struct sk_buff *skb;
317 int ret;
318
319 vq = vsock->vqs[VSOCK_VQ_RX];
320
321 do {
322 skb = virtio_vsock_alloc_linear_skb(total_len, GFP_KERNEL);
323 if (!skb)
324 break;
325
326 memset(skb->head, 0, VIRTIO_VSOCK_SKB_HEADROOM);
327 sg_init_one(&pkt, virtio_vsock_hdr(skb), total_len);
328 p = &pkt;
329 ret = virtqueue_add_sgs(vq, &p, 0, 1, skb, GFP_KERNEL);
330 if (ret < 0) {
331 kfree_skb(skb);
332 break;
333 }
334
335 vsock->rx_buf_nr++;
336 } while (vq->num_free);
337 if (vsock->rx_buf_nr > vsock->rx_buf_max_nr)
338 vsock->rx_buf_max_nr = vsock->rx_buf_nr;
339 virtqueue_kick(vq);
340 }
341
virtio_transport_tx_work(struct work_struct * work)342 static void virtio_transport_tx_work(struct work_struct *work)
343 {
344 struct virtio_vsock *vsock =
345 container_of(work, struct virtio_vsock, tx_work);
346 struct virtqueue *vq;
347 bool added = false;
348
349 mutex_lock(&vsock->tx_lock);
350
351 if (!vsock->tx_run)
352 goto out;
353
354 vq = vsock->vqs[VSOCK_VQ_TX];
355
356 do {
357 struct sk_buff *skb;
358 unsigned int len;
359
360 virtqueue_disable_cb(vq);
361 while ((skb = virtqueue_get_buf(vq, &len)) != NULL) {
362 virtio_transport_consume_skb_sent(skb, true);
363 added = true;
364 }
365 } while (!virtqueue_enable_cb(vq));
366
367 out:
368 mutex_unlock(&vsock->tx_lock);
369
370 if (added)
371 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
372 }
373
374 /* Is there space left for replies to rx packets? */
virtio_transport_more_replies(struct virtio_vsock * vsock)375 static bool virtio_transport_more_replies(struct virtio_vsock *vsock)
376 {
377 struct virtqueue *vq = vsock->vqs[VSOCK_VQ_RX];
378 int val;
379
380 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
381 val = atomic_read(&vsock->queued_replies);
382
383 return val < virtqueue_get_vring_size(vq);
384 }
385
386 /* event_lock must be held */
virtio_vsock_event_fill_one(struct virtio_vsock * vsock,struct virtio_vsock_event * event)387 static 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_cache_clean(vq, &sg, 1, event, GFP_KERNEL);
398 }
399
400 /* event_lock must be held */
virtio_vsock_event_fill(struct virtio_vsock * vsock)401 static 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
virtio_vsock_reset_sock(struct sock * sk)414 static void virtio_vsock_reset_sock(struct sock *sk)
415 {
416 /* vmci_transport.c doesn't take sk_lock here either. At least we're
417 * under vsock_table_lock so the sock cannot disappear while we're
418 * executing.
419 */
420
421 sk->sk_state = TCP_CLOSE;
422 sk->sk_err = ECONNRESET;
423 sk_error_report(sk);
424 }
425
virtio_vsock_update_guest_cid(struct virtio_vsock * vsock)426 static void virtio_vsock_update_guest_cid(struct virtio_vsock *vsock)
427 {
428 struct virtio_device *vdev = vsock->vdev;
429 __le64 guest_cid;
430
431 vdev->config->get(vdev, offsetof(struct virtio_vsock_config, guest_cid),
432 &guest_cid, sizeof(guest_cid));
433 vsock->guest_cid = le64_to_cpu(guest_cid);
434 }
435
436 /* event_lock must be held */
virtio_vsock_event_handle(struct virtio_vsock * vsock,struct virtio_vsock_event * event)437 static void virtio_vsock_event_handle(struct virtio_vsock *vsock,
438 struct virtio_vsock_event *event)
439 {
440 switch (le32_to_cpu(event->id)) {
441 case VIRTIO_VSOCK_EVENT_TRANSPORT_RESET:
442 virtio_vsock_update_guest_cid(vsock);
443 vsock_for_each_connected_socket(&virtio_transport.transport,
444 virtio_vsock_reset_sock);
445 break;
446 }
447 }
448
virtio_transport_event_work(struct work_struct * work)449 static void virtio_transport_event_work(struct work_struct *work)
450 {
451 struct virtio_vsock *vsock =
452 container_of(work, struct virtio_vsock, event_work);
453 struct virtqueue *vq;
454
455 mutex_lock(&vsock->event_lock);
456
457 if (!vsock->event_run)
458 goto out;
459
460 vq = vsock->vqs[VSOCK_VQ_EVENT];
461
462 do {
463 struct virtio_vsock_event *event;
464 unsigned int len;
465
466 virtqueue_disable_cb(vq);
467 while ((event = virtqueue_get_buf(vq, &len)) != NULL) {
468 if (len == sizeof(*event))
469 virtio_vsock_event_handle(vsock, event);
470
471 virtio_vsock_event_fill_one(vsock, event);
472 }
473 } while (!virtqueue_enable_cb(vq));
474
475 virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
476 out:
477 mutex_unlock(&vsock->event_lock);
478 }
479
virtio_vsock_event_done(struct virtqueue * vq)480 static void virtio_vsock_event_done(struct virtqueue *vq)
481 {
482 struct virtio_vsock *vsock = vq->vdev->priv;
483
484 if (!vsock)
485 return;
486 queue_work(virtio_vsock_workqueue, &vsock->event_work);
487 }
488
virtio_vsock_tx_done(struct virtqueue * vq)489 static void virtio_vsock_tx_done(struct virtqueue *vq)
490 {
491 struct virtio_vsock *vsock = vq->vdev->priv;
492
493 if (!vsock)
494 return;
495 queue_work(virtio_vsock_workqueue, &vsock->tx_work);
496 }
497
virtio_vsock_rx_done(struct virtqueue * vq)498 static void virtio_vsock_rx_done(struct virtqueue *vq)
499 {
500 struct virtio_vsock *vsock = vq->vdev->priv;
501
502 if (!vsock)
503 return;
504 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
505 }
506
virtio_transport_can_msgzerocopy(int bufs_num)507 static bool virtio_transport_can_msgzerocopy(int bufs_num)
508 {
509 struct virtio_vsock *vsock;
510 bool res = false;
511
512 rcu_read_lock();
513
514 vsock = rcu_dereference(the_virtio_vsock);
515 if (vsock) {
516 struct virtqueue *vq = vsock->vqs[VSOCK_VQ_TX];
517
518 /* Check that tx queue is large enough to keep whole
519 * data to send. This is needed, because when there is
520 * not enough free space in the queue, current skb to
521 * send will be reinserted to the head of tx list of
522 * the socket to retry transmission later, so if skb
523 * is bigger than whole queue, it will be reinserted
524 * again and again, thus blocking other skbs to be sent.
525 * Each page of the user provided buffer will be added
526 * as a single buffer to the tx virtqueue, so compare
527 * number of pages against maximum capacity of the queue.
528 */
529 if (bufs_num <= vq->num_max)
530 res = true;
531 }
532
533 rcu_read_unlock();
534
535 return res;
536 }
537
virtio_transport_msgzerocopy_allow(void)538 static bool virtio_transport_msgzerocopy_allow(void)
539 {
540 return true;
541 }
542
virtio_transport_stream_allow(struct vsock_sock * vsk,u32 cid,u32 port)543 bool virtio_transport_stream_allow(struct vsock_sock *vsk, u32 cid, u32 port)
544 {
545 return vsock_net_mode_global(vsk);
546 }
547
548 static bool virtio_transport_seqpacket_allow(struct vsock_sock *vsk,
549 u32 remote_cid);
550
virtio_transport_has_remote_cid(struct vsock_sock * vsk,u32 cid)551 static bool virtio_transport_has_remote_cid(struct vsock_sock *vsk, u32 cid)
552 {
553 /* The CID could be implemented by the host. Always assume it is. */
554 return true;
555 }
556
557 static struct virtio_transport virtio_transport = {
558 .transport = {
559 .module = THIS_MODULE,
560
561 .get_local_cid = virtio_transport_get_local_cid,
562 .has_remote_cid = virtio_transport_has_remote_cid,
563
564 .init = virtio_transport_do_socket_init,
565 .destruct = virtio_transport_destruct,
566 .release = virtio_transport_release,
567 .connect = virtio_transport_connect,
568 .shutdown = virtio_transport_shutdown,
569 .cancel_pkt = virtio_transport_cancel_pkt,
570
571 .dgram_bind = virtio_transport_dgram_bind,
572 .dgram_dequeue = virtio_transport_dgram_dequeue,
573 .dgram_enqueue = virtio_transport_dgram_enqueue,
574 .dgram_allow = virtio_transport_dgram_allow,
575
576 .stream_dequeue = virtio_transport_stream_dequeue,
577 .stream_enqueue = virtio_transport_stream_enqueue,
578 .stream_has_data = virtio_transport_stream_has_data,
579 .stream_has_space = virtio_transport_stream_has_space,
580 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
581 .stream_is_active = virtio_transport_stream_is_active,
582 .stream_allow = virtio_transport_stream_allow,
583
584 .seqpacket_dequeue = virtio_transport_seqpacket_dequeue,
585 .seqpacket_enqueue = virtio_transport_seqpacket_enqueue,
586 .seqpacket_allow = virtio_transport_seqpacket_allow,
587 .seqpacket_has_data = virtio_transport_seqpacket_has_data,
588
589 .msgzerocopy_allow = virtio_transport_msgzerocopy_allow,
590
591 .notify_poll_in = virtio_transport_notify_poll_in,
592 .notify_poll_out = virtio_transport_notify_poll_out,
593 .notify_recv_init = virtio_transport_notify_recv_init,
594 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
595 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
596 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
597 .notify_send_init = virtio_transport_notify_send_init,
598 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
599 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
600 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
601 .notify_buffer_size = virtio_transport_notify_buffer_size,
602 .notify_set_rcvlowat = virtio_transport_notify_set_rcvlowat,
603
604 .unsent_bytes = virtio_transport_unsent_bytes,
605
606 .read_skb = virtio_transport_read_skb,
607 },
608
609 .send_pkt = virtio_transport_send_pkt,
610 .can_msgzerocopy = virtio_transport_can_msgzerocopy,
611 };
612
613 static bool
virtio_transport_seqpacket_allow(struct vsock_sock * vsk,u32 remote_cid)614 virtio_transport_seqpacket_allow(struct vsock_sock *vsk, u32 remote_cid)
615 {
616 struct virtio_vsock *vsock;
617 bool seqpacket_allow;
618
619 if (!vsock_net_mode_global(vsk))
620 return false;
621
622 seqpacket_allow = false;
623 rcu_read_lock();
624 vsock = rcu_dereference(the_virtio_vsock);
625 if (vsock)
626 seqpacket_allow = vsock->seqpacket_allow;
627 rcu_read_unlock();
628
629 return seqpacket_allow;
630 }
631
virtio_transport_rx_work(struct work_struct * work)632 static void virtio_transport_rx_work(struct work_struct *work)
633 {
634 struct virtio_vsock *vsock =
635 container_of(work, struct virtio_vsock, rx_work);
636 struct virtqueue *vq;
637
638 mutex_lock(&vsock->rx_lock);
639
640 if (!vsock->rx_run)
641 goto out_nofill;
642
643 vq = vsock->vqs[VSOCK_VQ_RX];
644
645 do {
646 virtqueue_disable_cb(vq);
647 for (;;) {
648 unsigned int len, payload_len;
649 struct virtio_vsock_hdr *hdr;
650 struct sk_buff *skb;
651
652 if (!virtio_transport_more_replies(vsock)) {
653 /* Stop rx until the device processes already
654 * pending replies. Leave rx virtqueue
655 * callbacks disabled.
656 */
657 goto out;
658 }
659
660 skb = virtqueue_get_buf(vq, &len);
661 if (!skb)
662 break;
663
664 vsock->rx_buf_nr--;
665
666 /* Drop short/long packets */
667 if (unlikely(len < sizeof(*hdr) ||
668 len > virtio_vsock_skb_len(skb))) {
669 kfree_skb(skb);
670 continue;
671 }
672
673 hdr = virtio_vsock_hdr(skb);
674 payload_len = le32_to_cpu(hdr->len);
675 if (unlikely(payload_len > len - sizeof(*hdr))) {
676 kfree_skb(skb);
677 continue;
678 }
679
680 if (payload_len)
681 virtio_vsock_skb_put(skb, payload_len);
682
683 virtio_transport_deliver_tap_pkt(skb);
684
685 /* Force virtio-transport into global mode since it
686 * does not yet support local-mode namespacing.
687 */
688 virtio_transport_recv_pkt(&virtio_transport, skb, NULL);
689 }
690 } while (!virtqueue_enable_cb(vq));
691
692 out:
693 if (vsock->rx_buf_nr < vsock->rx_buf_max_nr / 2)
694 virtio_vsock_rx_fill(vsock);
695 out_nofill:
696 mutex_unlock(&vsock->rx_lock);
697 }
698
virtio_vsock_vqs_init(struct virtio_vsock * vsock)699 static int virtio_vsock_vqs_init(struct virtio_vsock *vsock)
700 {
701 struct virtio_device *vdev = vsock->vdev;
702 struct virtqueue_info vqs_info[] = {
703 { "rx", virtio_vsock_rx_done },
704 { "tx", virtio_vsock_tx_done },
705 { "event", virtio_vsock_event_done },
706 };
707 int ret;
708
709 mutex_lock(&vsock->rx_lock);
710 vsock->rx_buf_nr = 0;
711 vsock->rx_buf_max_nr = 0;
712 mutex_unlock(&vsock->rx_lock);
713
714 atomic_set(&vsock->queued_replies, 0);
715
716 ret = virtio_find_vqs(vdev, VSOCK_VQ_MAX, vsock->vqs, vqs_info, NULL);
717 if (ret < 0)
718 return ret;
719
720 virtio_vsock_update_guest_cid(vsock);
721
722 virtio_device_ready(vdev);
723
724 return 0;
725 }
726
virtio_vsock_vqs_start(struct virtio_vsock * vsock)727 static void virtio_vsock_vqs_start(struct virtio_vsock *vsock)
728 {
729 mutex_lock(&vsock->tx_lock);
730 vsock->tx_run = true;
731 mutex_unlock(&vsock->tx_lock);
732
733 mutex_lock(&vsock->rx_lock);
734 virtio_vsock_rx_fill(vsock);
735 vsock->rx_run = true;
736 mutex_unlock(&vsock->rx_lock);
737
738 mutex_lock(&vsock->event_lock);
739 virtio_vsock_event_fill(vsock);
740 vsock->event_run = true;
741 mutex_unlock(&vsock->event_lock);
742
743 /* virtio_transport_send_pkt() can queue packets once
744 * the_virtio_vsock is set, but they won't be processed until
745 * vsock->tx_run is set to true. We queue vsock->send_pkt_work
746 * when initialization finishes to send those packets queued
747 * earlier.
748 * We don't need to queue the other workers (rx, event) because
749 * as long as we don't fill the queues with empty buffers, the
750 * host can't send us any notification.
751 */
752 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
753 }
754
virtio_vsock_vqs_del(struct virtio_vsock * vsock)755 static void virtio_vsock_vqs_del(struct virtio_vsock *vsock)
756 {
757 struct virtio_device *vdev = vsock->vdev;
758 struct sk_buff *skb;
759
760 /* Reset all connected sockets when the VQs disappear */
761 vsock_for_each_connected_socket(&virtio_transport.transport,
762 virtio_vsock_reset_sock);
763
764 /* Stop all work handlers to make sure no one is accessing the device,
765 * so we can safely call virtio_reset_device().
766 */
767 mutex_lock(&vsock->rx_lock);
768 vsock->rx_run = false;
769 mutex_unlock(&vsock->rx_lock);
770
771 mutex_lock(&vsock->tx_lock);
772 vsock->tx_run = false;
773 mutex_unlock(&vsock->tx_lock);
774
775 mutex_lock(&vsock->event_lock);
776 vsock->event_run = false;
777 mutex_unlock(&vsock->event_lock);
778
779 /* Flush all device writes and interrupts, device will not use any
780 * more buffers.
781 */
782 virtio_reset_device(vdev);
783
784 mutex_lock(&vsock->rx_lock);
785 while ((skb = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX])))
786 kfree_skb(skb);
787 mutex_unlock(&vsock->rx_lock);
788
789 mutex_lock(&vsock->tx_lock);
790 while ((skb = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX])))
791 kfree_skb(skb);
792 mutex_unlock(&vsock->tx_lock);
793
794 virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue);
795
796 /* Delete virtqueues and flush outstanding callbacks if any */
797 vdev->config->del_vqs(vdev);
798 }
799
virtio_vsock_probe(struct virtio_device * vdev)800 static int virtio_vsock_probe(struct virtio_device *vdev)
801 {
802 struct virtio_vsock *vsock = NULL;
803 int ret;
804 int i;
805
806 ret = mutex_lock_interruptible(&the_virtio_vsock_mutex);
807 if (ret)
808 return ret;
809
810 /* Only one virtio-vsock device per guest is supported */
811 if (rcu_dereference_protected(the_virtio_vsock,
812 lockdep_is_held(&the_virtio_vsock_mutex))) {
813 ret = -EBUSY;
814 goto out;
815 }
816
817 vsock = kzalloc_obj(*vsock);
818 if (!vsock) {
819 ret = -ENOMEM;
820 goto out;
821 }
822
823 vsock->vdev = vdev;
824
825
826 mutex_init(&vsock->tx_lock);
827 mutex_init(&vsock->rx_lock);
828 mutex_init(&vsock->event_lock);
829 skb_queue_head_init(&vsock->send_pkt_queue);
830 INIT_WORK(&vsock->rx_work, virtio_transport_rx_work);
831 INIT_WORK(&vsock->tx_work, virtio_transport_tx_work);
832 INIT_WORK(&vsock->event_work, virtio_transport_event_work);
833 INIT_WORK(&vsock->send_pkt_work, virtio_transport_send_pkt_work);
834
835 if (virtio_has_feature(vdev, VIRTIO_VSOCK_F_SEQPACKET))
836 vsock->seqpacket_allow = true;
837
838 vdev->priv = vsock;
839
840 ret = virtio_vsock_vqs_init(vsock);
841 if (ret < 0)
842 goto out;
843
844 for (i = 0; i < ARRAY_SIZE(vsock->out_sgs); i++)
845 vsock->out_sgs[i] = &vsock->out_bufs[i];
846
847 rcu_assign_pointer(the_virtio_vsock, vsock);
848 virtio_vsock_vqs_start(vsock);
849
850 mutex_unlock(&the_virtio_vsock_mutex);
851
852 return 0;
853
854 out:
855 kfree(vsock);
856 mutex_unlock(&the_virtio_vsock_mutex);
857 return ret;
858 }
859
virtio_vsock_remove(struct virtio_device * vdev)860 static void virtio_vsock_remove(struct virtio_device *vdev)
861 {
862 struct virtio_vsock *vsock = vdev->priv;
863
864 mutex_lock(&the_virtio_vsock_mutex);
865
866 vdev->priv = NULL;
867 rcu_assign_pointer(the_virtio_vsock, NULL);
868 synchronize_rcu();
869
870 virtio_vsock_vqs_del(vsock);
871
872 /* Other works can be queued before 'config->del_vqs()', so we flush
873 * all works before to free the vsock object to avoid use after free.
874 */
875 flush_work(&vsock->tx_work);
876 flush_work(&vsock->event_work);
877 flush_work(&vsock->send_pkt_work);
878 flush_work(&vsock->rx_work);
879
880 mutex_unlock(&the_virtio_vsock_mutex);
881
882 kfree(vsock);
883 }
884
885 #ifdef CONFIG_PM_SLEEP
virtio_vsock_freeze(struct virtio_device * vdev)886 static int virtio_vsock_freeze(struct virtio_device *vdev)
887 {
888 struct virtio_vsock *vsock = vdev->priv;
889
890 mutex_lock(&the_virtio_vsock_mutex);
891
892 rcu_assign_pointer(the_virtio_vsock, NULL);
893 synchronize_rcu();
894
895 virtio_vsock_vqs_del(vsock);
896
897 mutex_unlock(&the_virtio_vsock_mutex);
898
899 return 0;
900 }
901
virtio_vsock_restore(struct virtio_device * vdev)902 static int virtio_vsock_restore(struct virtio_device *vdev)
903 {
904 struct virtio_vsock *vsock = vdev->priv;
905 int ret;
906
907 mutex_lock(&the_virtio_vsock_mutex);
908
909 /* Only one virtio-vsock device per guest is supported */
910 if (rcu_dereference_protected(the_virtio_vsock,
911 lockdep_is_held(&the_virtio_vsock_mutex))) {
912 ret = -EBUSY;
913 goto out;
914 }
915
916 ret = virtio_vsock_vqs_init(vsock);
917 if (ret < 0)
918 goto out;
919
920 rcu_assign_pointer(the_virtio_vsock, vsock);
921 virtio_vsock_vqs_start(vsock);
922
923 out:
924 mutex_unlock(&the_virtio_vsock_mutex);
925 return ret;
926 }
927 #endif /* CONFIG_PM_SLEEP */
928
929 static struct virtio_device_id id_table[] = {
930 { VIRTIO_ID_VSOCK, VIRTIO_DEV_ANY_ID },
931 { 0 },
932 };
933
934 static unsigned int features[] = {
935 VIRTIO_VSOCK_F_SEQPACKET
936 };
937
938 static struct virtio_driver virtio_vsock_driver = {
939 .feature_table = features,
940 .feature_table_size = ARRAY_SIZE(features),
941 .driver.name = KBUILD_MODNAME,
942 .id_table = id_table,
943 .probe = virtio_vsock_probe,
944 .remove = virtio_vsock_remove,
945 #ifdef CONFIG_PM_SLEEP
946 .freeze = virtio_vsock_freeze,
947 .restore = virtio_vsock_restore,
948 #endif
949 };
950
virtio_vsock_init(void)951 static int __init virtio_vsock_init(void)
952 {
953 int ret;
954
955 virtio_vsock_workqueue = alloc_workqueue("virtio_vsock", WQ_PERCPU, 0);
956 if (!virtio_vsock_workqueue)
957 return -ENOMEM;
958
959 ret = vsock_core_register(&virtio_transport.transport,
960 VSOCK_TRANSPORT_F_G2H);
961 if (ret)
962 goto out_wq;
963
964 ret = register_virtio_driver(&virtio_vsock_driver);
965 if (ret)
966 goto out_vci;
967
968 return 0;
969
970 out_vci:
971 vsock_core_unregister(&virtio_transport.transport);
972 out_wq:
973 destroy_workqueue(virtio_vsock_workqueue);
974 return ret;
975 }
976
virtio_vsock_exit(void)977 static void __exit virtio_vsock_exit(void)
978 {
979 unregister_virtio_driver(&virtio_vsock_driver);
980 vsock_core_unregister(&virtio_transport.transport);
981 destroy_workqueue(virtio_vsock_workqueue);
982 }
983
984 module_init(virtio_vsock_init);
985 module_exit(virtio_vsock_exit);
986 MODULE_LICENSE("GPL v2");
987 MODULE_AUTHOR("Asias He");
988 MODULE_DESCRIPTION("virtio transport for vsock");
989 MODULE_DEVICE_TABLE(virtio, id_table);
990