Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * vhost 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#include <linux/miscdevice.h>
10#include <linux/atomic.h>
11#include <linux/module.h>
12#include <linux/mutex.h>
13#include <linux/vmalloc.h>
14#include <net/sock.h>
15#include <linux/virtio_vsock.h>
16#include <linux/vhost.h>
17#include <linux/hashtable.h>
18
19#include <net/af_vsock.h>
20#include "vhost.h"
21
22#define VHOST_VSOCK_DEFAULT_HOST_CID 2
23/* Max number of bytes transferred before requeueing the job.
24 * Using this limit prevents one virtqueue from starving others. */
25#define VHOST_VSOCK_WEIGHT 0x80000
26/* Max number of packets transferred before requeueing the job.
27 * Using this limit prevents one virtqueue from starving others with
28 * small pkts.
29 */
30#define VHOST_VSOCK_PKT_WEIGHT 256
31
32enum {
33 VHOST_VSOCK_FEATURES = VHOST_FEATURES,
34};
35
36/* Used to track all the vhost_vsock instances on the system. */
37static DEFINE_MUTEX(vhost_vsock_mutex);
38static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
39
40struct vhost_vsock {
41 struct vhost_dev dev;
42 struct vhost_virtqueue vqs[2];
43
44 /* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */
45 struct hlist_node hash;
46
47 struct vhost_work send_pkt_work;
48 spinlock_t send_pkt_list_lock;
49 struct list_head send_pkt_list; /* host->guest pending packets */
50
51 atomic_t queued_replies;
52
53 u32 guest_cid;
54};
55
56static u32 vhost_transport_get_local_cid(void)
57{
58 return VHOST_VSOCK_DEFAULT_HOST_CID;
59}
60
61/* Callers that dereference the return value must hold vhost_vsock_mutex or the
62 * RCU read lock.
63 */
64static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
65{
66 struct vhost_vsock *vsock;
67
68 hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
69 u32 other_cid = vsock->guest_cid;
70
71 /* Skip instances that have no CID yet */
72 if (other_cid == 0)
73 continue;
74
75 if (other_cid == guest_cid)
76 return vsock;
77
78 }
79
80 return NULL;
81}
82
83static void
84vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
85 struct vhost_virtqueue *vq)
86{
87 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
88 int pkts = 0, total_len = 0;
89 bool added = false;
90 bool restart_tx = false;
91
92 mutex_lock(&vq->mutex);
93
94 if (!vhost_vq_get_backend(vq))
95 goto out;
96
97 /* Avoid further vmexits, we're already processing the virtqueue */
98 vhost_disable_notify(&vsock->dev, vq);
99
100 do {
101 struct virtio_vsock_pkt *pkt;
102 struct iov_iter iov_iter;
103 unsigned out, in;
104 size_t nbytes;
105 size_t iov_len, payload_len;
106 int head;
107
108 spin_lock_bh(&vsock->send_pkt_list_lock);
109 if (list_empty(&vsock->send_pkt_list)) {
110 spin_unlock_bh(&vsock->send_pkt_list_lock);
111 vhost_enable_notify(&vsock->dev, vq);
112 break;
113 }
114
115 pkt = list_first_entry(&vsock->send_pkt_list,
116 struct virtio_vsock_pkt, list);
117 list_del_init(&pkt->list);
118 spin_unlock_bh(&vsock->send_pkt_list_lock);
119
120 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
121 &out, &in, NULL, NULL);
122 if (head < 0) {
123 spin_lock_bh(&vsock->send_pkt_list_lock);
124 list_add(&pkt->list, &vsock->send_pkt_list);
125 spin_unlock_bh(&vsock->send_pkt_list_lock);
126 break;
127 }
128
129 if (head == vq->num) {
130 spin_lock_bh(&vsock->send_pkt_list_lock);
131 list_add(&pkt->list, &vsock->send_pkt_list);
132 spin_unlock_bh(&vsock->send_pkt_list_lock);
133
134 /* We cannot finish yet if more buffers snuck in while
135 * re-enabling notify.
136 */
137 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
138 vhost_disable_notify(&vsock->dev, vq);
139 continue;
140 }
141 break;
142 }
143
144 if (out) {
145 virtio_transport_free_pkt(pkt);
146 vq_err(vq, "Expected 0 output buffers, got %u\n", out);
147 break;
148 }
149
150 iov_len = iov_length(&vq->iov[out], in);
151 if (iov_len < sizeof(pkt->hdr)) {
152 virtio_transport_free_pkt(pkt);
153 vq_err(vq, "Buffer len [%zu] too small\n", iov_len);
154 break;
155 }
156
157 iov_iter_init(&iov_iter, READ, &vq->iov[out], in, iov_len);
158 payload_len = pkt->len - pkt->off;
159
160 /* If the packet is greater than the space available in the
161 * buffer, we split it using multiple buffers.
162 */
163 if (payload_len > iov_len - sizeof(pkt->hdr))
164 payload_len = iov_len - sizeof(pkt->hdr);
165
166 /* Set the correct length in the header */
167 pkt->hdr.len = cpu_to_le32(payload_len);
168
169 nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
170 if (nbytes != sizeof(pkt->hdr)) {
171 virtio_transport_free_pkt(pkt);
172 vq_err(vq, "Faulted on copying pkt hdr\n");
173 break;
174 }
175
176 nbytes = copy_to_iter(pkt->buf + pkt->off, payload_len,
177 &iov_iter);
178 if (nbytes != payload_len) {
179 virtio_transport_free_pkt(pkt);
180 vq_err(vq, "Faulted on copying pkt buf\n");
181 break;
182 }
183
184 /* Deliver to monitoring devices all packets that we
185 * will transmit.
186 */
187 virtio_transport_deliver_tap_pkt(pkt);
188
189 vhost_add_used(vq, head, sizeof(pkt->hdr) + payload_len);
190 added = true;
191
192 pkt->off += payload_len;
193 total_len += payload_len;
194
195 /* If we didn't send all the payload we can requeue the packet
196 * to send it with the next available buffer.
197 */
198 if (pkt->off < pkt->len) {
199 /* We are queueing the same virtio_vsock_pkt to handle
200 * the remaining bytes, and we want to deliver it
201 * to monitoring devices in the next iteration.
202 */
203 pkt->tap_delivered = false;
204
205 spin_lock_bh(&vsock->send_pkt_list_lock);
206 list_add(&pkt->list, &vsock->send_pkt_list);
207 spin_unlock_bh(&vsock->send_pkt_list_lock);
208 } else {
209 if (pkt->reply) {
210 int val;
211
212 val = atomic_dec_return(&vsock->queued_replies);
213
214 /* Do we have resources to resume tx
215 * processing?
216 */
217 if (val + 1 == tx_vq->num)
218 restart_tx = true;
219 }
220
221 virtio_transport_free_pkt(pkt);
222 }
223 } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
224 if (added)
225 vhost_signal(&vsock->dev, vq);
226
227out:
228 mutex_unlock(&vq->mutex);
229
230 if (restart_tx)
231 vhost_poll_queue(&tx_vq->poll);
232}
233
234static void vhost_transport_send_pkt_work(struct vhost_work *work)
235{
236 struct vhost_virtqueue *vq;
237 struct vhost_vsock *vsock;
238
239 vsock = container_of(work, struct vhost_vsock, send_pkt_work);
240 vq = &vsock->vqs[VSOCK_VQ_RX];
241
242 vhost_transport_do_send_pkt(vsock, vq);
243}
244
245static int
246vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
247{
248 struct vhost_vsock *vsock;
249 int len = pkt->len;
250
251 rcu_read_lock();
252
253 /* Find the vhost_vsock according to guest context id */
254 vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
255 if (!vsock) {
256 rcu_read_unlock();
257 virtio_transport_free_pkt(pkt);
258 return -ENODEV;
259 }
260
261 if (pkt->reply)
262 atomic_inc(&vsock->queued_replies);
263
264 spin_lock_bh(&vsock->send_pkt_list_lock);
265 list_add_tail(&pkt->list, &vsock->send_pkt_list);
266 spin_unlock_bh(&vsock->send_pkt_list_lock);
267
268 vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
269
270 rcu_read_unlock();
271 return len;
272}
273
274static int
275vhost_transport_cancel_pkt(struct vsock_sock *vsk)
276{
277 struct vhost_vsock *vsock;
278 struct virtio_vsock_pkt *pkt, *n;
279 int cnt = 0;
280 int ret = -ENODEV;
281 LIST_HEAD(freeme);
282
283 rcu_read_lock();
284
285 /* Find the vhost_vsock according to guest context id */
286 vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
287 if (!vsock)
288 goto out;
289
290 spin_lock_bh(&vsock->send_pkt_list_lock);
291 list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
292 if (pkt->vsk != vsk)
293 continue;
294 list_move(&pkt->list, &freeme);
295 }
296 spin_unlock_bh(&vsock->send_pkt_list_lock);
297
298 list_for_each_entry_safe(pkt, n, &freeme, list) {
299 if (pkt->reply)
300 cnt++;
301 list_del(&pkt->list);
302 virtio_transport_free_pkt(pkt);
303 }
304
305 if (cnt) {
306 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
307 int new_cnt;
308
309 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
310 if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
311 vhost_poll_queue(&tx_vq->poll);
312 }
313
314 ret = 0;
315out:
316 rcu_read_unlock();
317 return ret;
318}
319
320static struct virtio_vsock_pkt *
321vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
322 unsigned int out, unsigned int in)
323{
324 struct virtio_vsock_pkt *pkt;
325 struct iov_iter iov_iter;
326 size_t nbytes;
327 size_t len;
328
329 if (in != 0) {
330 vq_err(vq, "Expected 0 input buffers, got %u\n", in);
331 return NULL;
332 }
333
334 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
335 if (!pkt)
336 return NULL;
337
338 len = iov_length(vq->iov, out);
339 iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
340
341 nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
342 if (nbytes != sizeof(pkt->hdr)) {
343 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
344 sizeof(pkt->hdr), nbytes);
345 kfree(pkt);
346 return NULL;
347 }
348
349 if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM)
350 pkt->len = le32_to_cpu(pkt->hdr.len);
351
352 /* No payload */
353 if (!pkt->len)
354 return pkt;
355
356 /* The pkt is too big */
357 if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
358 kfree(pkt);
359 return NULL;
360 }
361
362 pkt->buf = kmalloc(pkt->len, GFP_KERNEL);
363 if (!pkt->buf) {
364 kfree(pkt);
365 return NULL;
366 }
367
368 pkt->buf_len = pkt->len;
369
370 nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
371 if (nbytes != pkt->len) {
372 vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
373 pkt->len, nbytes);
374 virtio_transport_free_pkt(pkt);
375 return NULL;
376 }
377
378 return pkt;
379}
380
381/* Is there space left for replies to rx packets? */
382static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
383{
384 struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
385 int val;
386
387 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
388 val = atomic_read(&vsock->queued_replies);
389
390 return val < vq->num;
391}
392
393static struct virtio_transport vhost_transport = {
394 .transport = {
395 .module = THIS_MODULE,
396
397 .get_local_cid = vhost_transport_get_local_cid,
398
399 .init = virtio_transport_do_socket_init,
400 .destruct = virtio_transport_destruct,
401 .release = virtio_transport_release,
402 .connect = virtio_transport_connect,
403 .shutdown = virtio_transport_shutdown,
404 .cancel_pkt = vhost_transport_cancel_pkt,
405
406 .dgram_enqueue = virtio_transport_dgram_enqueue,
407 .dgram_dequeue = virtio_transport_dgram_dequeue,
408 .dgram_bind = virtio_transport_dgram_bind,
409 .dgram_allow = virtio_transport_dgram_allow,
410
411 .stream_enqueue = virtio_transport_stream_enqueue,
412 .stream_dequeue = virtio_transport_stream_dequeue,
413 .stream_has_data = virtio_transport_stream_has_data,
414 .stream_has_space = virtio_transport_stream_has_space,
415 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
416 .stream_is_active = virtio_transport_stream_is_active,
417 .stream_allow = virtio_transport_stream_allow,
418
419 .notify_poll_in = virtio_transport_notify_poll_in,
420 .notify_poll_out = virtio_transport_notify_poll_out,
421 .notify_recv_init = virtio_transport_notify_recv_init,
422 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
423 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
424 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
425 .notify_send_init = virtio_transport_notify_send_init,
426 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
427 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
428 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
429 .notify_buffer_size = virtio_transport_notify_buffer_size,
430
431 },
432
433 .send_pkt = vhost_transport_send_pkt,
434};
435
436static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
437{
438 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
439 poll.work);
440 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
441 dev);
442 struct virtio_vsock_pkt *pkt;
443 int head, pkts = 0, total_len = 0;
444 unsigned int out, in;
445 bool added = false;
446
447 mutex_lock(&vq->mutex);
448
449 if (!vhost_vq_get_backend(vq))
450 goto out;
451
452 vhost_disable_notify(&vsock->dev, vq);
453 do {
454 u32 len;
455
456 if (!vhost_vsock_more_replies(vsock)) {
457 /* Stop tx until the device processes already
458 * pending replies. Leave tx virtqueue
459 * callbacks disabled.
460 */
461 goto no_more_replies;
462 }
463
464 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
465 &out, &in, NULL, NULL);
466 if (head < 0)
467 break;
468
469 if (head == vq->num) {
470 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
471 vhost_disable_notify(&vsock->dev, vq);
472 continue;
473 }
474 break;
475 }
476
477 pkt = vhost_vsock_alloc_pkt(vq, out, in);
478 if (!pkt) {
479 vq_err(vq, "Faulted on pkt\n");
480 continue;
481 }
482
483 len = pkt->len;
484
485 /* Deliver to monitoring devices all received packets */
486 virtio_transport_deliver_tap_pkt(pkt);
487
488 /* Only accept correctly addressed packets */
489 if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid &&
490 le64_to_cpu(pkt->hdr.dst_cid) ==
491 vhost_transport_get_local_cid())
492 virtio_transport_recv_pkt(&vhost_transport, pkt);
493 else
494 virtio_transport_free_pkt(pkt);
495
496 len += sizeof(pkt->hdr);
497 vhost_add_used(vq, head, len);
498 total_len += len;
499 added = true;
500 } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
501
502no_more_replies:
503 if (added)
504 vhost_signal(&vsock->dev, vq);
505
506out:
507 mutex_unlock(&vq->mutex);
508}
509
510static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
511{
512 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
513 poll.work);
514 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
515 dev);
516
517 vhost_transport_do_send_pkt(vsock, vq);
518}
519
520static int vhost_vsock_start(struct vhost_vsock *vsock)
521{
522 struct vhost_virtqueue *vq;
523 size_t i;
524 int ret;
525
526 mutex_lock(&vsock->dev.mutex);
527
528 ret = vhost_dev_check_owner(&vsock->dev);
529 if (ret)
530 goto err;
531
532 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
533 vq = &vsock->vqs[i];
534
535 mutex_lock(&vq->mutex);
536
537 if (!vhost_vq_access_ok(vq)) {
538 ret = -EFAULT;
539 goto err_vq;
540 }
541
542 if (!vhost_vq_get_backend(vq)) {
543 vhost_vq_set_backend(vq, vsock);
544 ret = vhost_vq_init_access(vq);
545 if (ret)
546 goto err_vq;
547 }
548
549 mutex_unlock(&vq->mutex);
550 }
551
552 /* Some packets may have been queued before the device was started,
553 * let's kick the send worker to send them.
554 */
555 vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
556
557 mutex_unlock(&vsock->dev.mutex);
558 return 0;
559
560err_vq:
561 vhost_vq_set_backend(vq, NULL);
562 mutex_unlock(&vq->mutex);
563
564 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
565 vq = &vsock->vqs[i];
566
567 mutex_lock(&vq->mutex);
568 vhost_vq_set_backend(vq, NULL);
569 mutex_unlock(&vq->mutex);
570 }
571err:
572 mutex_unlock(&vsock->dev.mutex);
573 return ret;
574}
575
576static int vhost_vsock_stop(struct vhost_vsock *vsock)
577{
578 size_t i;
579 int ret;
580
581 mutex_lock(&vsock->dev.mutex);
582
583 ret = vhost_dev_check_owner(&vsock->dev);
584 if (ret)
585 goto err;
586
587 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
588 struct vhost_virtqueue *vq = &vsock->vqs[i];
589
590 mutex_lock(&vq->mutex);
591 vhost_vq_set_backend(vq, NULL);
592 mutex_unlock(&vq->mutex);
593 }
594
595err:
596 mutex_unlock(&vsock->dev.mutex);
597 return ret;
598}
599
600static void vhost_vsock_free(struct vhost_vsock *vsock)
601{
602 kvfree(vsock);
603}
604
605static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
606{
607 struct vhost_virtqueue **vqs;
608 struct vhost_vsock *vsock;
609 int ret;
610
611 /* This struct is large and allocation could fail, fall back to vmalloc
612 * if there is no other way.
613 */
614 vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
615 if (!vsock)
616 return -ENOMEM;
617
618 vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
619 if (!vqs) {
620 ret = -ENOMEM;
621 goto out;
622 }
623
624 vsock->guest_cid = 0; /* no CID assigned yet */
625
626 atomic_set(&vsock->queued_replies, 0);
627
628 vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
629 vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
630 vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
631 vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
632
633 vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
634 UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
635 VHOST_VSOCK_WEIGHT, true, NULL);
636
637 file->private_data = vsock;
638 spin_lock_init(&vsock->send_pkt_list_lock);
639 INIT_LIST_HEAD(&vsock->send_pkt_list);
640 vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
641 return 0;
642
643out:
644 vhost_vsock_free(vsock);
645 return ret;
646}
647
648static void vhost_vsock_flush(struct vhost_vsock *vsock)
649{
650 int i;
651
652 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
653 if (vsock->vqs[i].handle_kick)
654 vhost_poll_flush(&vsock->vqs[i].poll);
655 vhost_work_flush(&vsock->dev, &vsock->send_pkt_work);
656}
657
658static void vhost_vsock_reset_orphans(struct sock *sk)
659{
660 struct vsock_sock *vsk = vsock_sk(sk);
661
662 /* vmci_transport.c doesn't take sk_lock here either. At least we're
663 * under vsock_table_lock so the sock cannot disappear while we're
664 * executing.
665 */
666
667 /* If the peer is still valid, no need to reset connection */
668 if (vhost_vsock_get(vsk->remote_addr.svm_cid))
669 return;
670
671 /* If the close timeout is pending, let it expire. This avoids races
672 * with the timeout callback.
673 */
674 if (vsk->close_work_scheduled)
675 return;
676
677 sock_set_flag(sk, SOCK_DONE);
678 vsk->peer_shutdown = SHUTDOWN_MASK;
679 sk->sk_state = SS_UNCONNECTED;
680 sk->sk_err = ECONNRESET;
681 sk->sk_error_report(sk);
682}
683
684static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
685{
686 struct vhost_vsock *vsock = file->private_data;
687
688 mutex_lock(&vhost_vsock_mutex);
689 if (vsock->guest_cid)
690 hash_del_rcu(&vsock->hash);
691 mutex_unlock(&vhost_vsock_mutex);
692
693 /* Wait for other CPUs to finish using vsock */
694 synchronize_rcu();
695
696 /* Iterating over all connections for all CIDs to find orphans is
697 * inefficient. Room for improvement here. */
698 vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
699
700 vhost_vsock_stop(vsock);
701 vhost_vsock_flush(vsock);
702 vhost_dev_stop(&vsock->dev);
703
704 spin_lock_bh(&vsock->send_pkt_list_lock);
705 while (!list_empty(&vsock->send_pkt_list)) {
706 struct virtio_vsock_pkt *pkt;
707
708 pkt = list_first_entry(&vsock->send_pkt_list,
709 struct virtio_vsock_pkt, list);
710 list_del_init(&pkt->list);
711 virtio_transport_free_pkt(pkt);
712 }
713 spin_unlock_bh(&vsock->send_pkt_list_lock);
714
715 vhost_dev_cleanup(&vsock->dev);
716 kfree(vsock->dev.vqs);
717 vhost_vsock_free(vsock);
718 return 0;
719}
720
721static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
722{
723 struct vhost_vsock *other;
724
725 /* Refuse reserved CIDs */
726 if (guest_cid <= VMADDR_CID_HOST ||
727 guest_cid == U32_MAX)
728 return -EINVAL;
729
730 /* 64-bit CIDs are not yet supported */
731 if (guest_cid > U32_MAX)
732 return -EINVAL;
733
734 /* Refuse if CID is assigned to the guest->host transport (i.e. nested
735 * VM), to make the loopback work.
736 */
737 if (vsock_find_cid(guest_cid))
738 return -EADDRINUSE;
739
740 /* Refuse if CID is already in use */
741 mutex_lock(&vhost_vsock_mutex);
742 other = vhost_vsock_get(guest_cid);
743 if (other && other != vsock) {
744 mutex_unlock(&vhost_vsock_mutex);
745 return -EADDRINUSE;
746 }
747
748 if (vsock->guest_cid)
749 hash_del_rcu(&vsock->hash);
750
751 vsock->guest_cid = guest_cid;
752 hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
753 mutex_unlock(&vhost_vsock_mutex);
754
755 return 0;
756}
757
758static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
759{
760 struct vhost_virtqueue *vq;
761 int i;
762
763 if (features & ~VHOST_VSOCK_FEATURES)
764 return -EOPNOTSUPP;
765
766 mutex_lock(&vsock->dev.mutex);
767 if ((features & (1 << VHOST_F_LOG_ALL)) &&
768 !vhost_log_access_ok(&vsock->dev)) {
769 mutex_unlock(&vsock->dev.mutex);
770 return -EFAULT;
771 }
772
773 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
774 vq = &vsock->vqs[i];
775 mutex_lock(&vq->mutex);
776 vq->acked_features = features;
777 mutex_unlock(&vq->mutex);
778 }
779 mutex_unlock(&vsock->dev.mutex);
780 return 0;
781}
782
783static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
784 unsigned long arg)
785{
786 struct vhost_vsock *vsock = f->private_data;
787 void __user *argp = (void __user *)arg;
788 u64 guest_cid;
789 u64 features;
790 int start;
791 int r;
792
793 switch (ioctl) {
794 case VHOST_VSOCK_SET_GUEST_CID:
795 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
796 return -EFAULT;
797 return vhost_vsock_set_cid(vsock, guest_cid);
798 case VHOST_VSOCK_SET_RUNNING:
799 if (copy_from_user(&start, argp, sizeof(start)))
800 return -EFAULT;
801 if (start)
802 return vhost_vsock_start(vsock);
803 else
804 return vhost_vsock_stop(vsock);
805 case VHOST_GET_FEATURES:
806 features = VHOST_VSOCK_FEATURES;
807 if (copy_to_user(argp, &features, sizeof(features)))
808 return -EFAULT;
809 return 0;
810 case VHOST_SET_FEATURES:
811 if (copy_from_user(&features, argp, sizeof(features)))
812 return -EFAULT;
813 return vhost_vsock_set_features(vsock, features);
814 default:
815 mutex_lock(&vsock->dev.mutex);
816 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
817 if (r == -ENOIOCTLCMD)
818 r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
819 else
820 vhost_vsock_flush(vsock);
821 mutex_unlock(&vsock->dev.mutex);
822 return r;
823 }
824}
825
826static const struct file_operations vhost_vsock_fops = {
827 .owner = THIS_MODULE,
828 .open = vhost_vsock_dev_open,
829 .release = vhost_vsock_dev_release,
830 .llseek = noop_llseek,
831 .unlocked_ioctl = vhost_vsock_dev_ioctl,
832 .compat_ioctl = compat_ptr_ioctl,
833};
834
835static struct miscdevice vhost_vsock_misc = {
836 .minor = VHOST_VSOCK_MINOR,
837 .name = "vhost-vsock",
838 .fops = &vhost_vsock_fops,
839};
840
841static int __init vhost_vsock_init(void)
842{
843 int ret;
844
845 ret = vsock_core_register(&vhost_transport.transport,
846 VSOCK_TRANSPORT_F_H2G);
847 if (ret < 0)
848 return ret;
849 return misc_register(&vhost_vsock_misc);
850};
851
852static void __exit vhost_vsock_exit(void)
853{
854 misc_deregister(&vhost_vsock_misc);
855 vsock_core_unregister(&vhost_transport.transport);
856};
857
858module_init(vhost_vsock_init);
859module_exit(vhost_vsock_exit);
860MODULE_LICENSE("GPL v2");
861MODULE_AUTHOR("Asias He");
862MODULE_DESCRIPTION("vhost transport for vsock ");
863MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
864MODULE_ALIAS("devname:vhost-vsock");
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * vhost 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#include <linux/miscdevice.h>
10#include <linux/atomic.h>
11#include <linux/module.h>
12#include <linux/mutex.h>
13#include <linux/vmalloc.h>
14#include <net/sock.h>
15#include <linux/virtio_vsock.h>
16#include <linux/vhost.h>
17#include <linux/hashtable.h>
18
19#include <net/af_vsock.h>
20#include "vhost.h"
21
22#define VHOST_VSOCK_DEFAULT_HOST_CID 2
23/* Max number of bytes transferred before requeueing the job.
24 * Using this limit prevents one virtqueue from starving others. */
25#define VHOST_VSOCK_WEIGHT 0x80000
26/* Max number of packets transferred before requeueing the job.
27 * Using this limit prevents one virtqueue from starving others with
28 * small pkts.
29 */
30#define VHOST_VSOCK_PKT_WEIGHT 256
31
32enum {
33 VHOST_VSOCK_FEATURES = VHOST_FEATURES,
34};
35
36/* Used to track all the vhost_vsock instances on the system. */
37static DEFINE_MUTEX(vhost_vsock_mutex);
38static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
39
40struct vhost_vsock {
41 struct vhost_dev dev;
42 struct vhost_virtqueue vqs[2];
43
44 /* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */
45 struct hlist_node hash;
46
47 struct vhost_work send_pkt_work;
48 spinlock_t send_pkt_list_lock;
49 struct list_head send_pkt_list; /* host->guest pending packets */
50
51 atomic_t queued_replies;
52
53 u32 guest_cid;
54};
55
56static u32 vhost_transport_get_local_cid(void)
57{
58 return VHOST_VSOCK_DEFAULT_HOST_CID;
59}
60
61/* Callers that dereference the return value must hold vhost_vsock_mutex or the
62 * RCU read lock.
63 */
64static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
65{
66 struct vhost_vsock *vsock;
67
68 hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
69 u32 other_cid = vsock->guest_cid;
70
71 /* Skip instances that have no CID yet */
72 if (other_cid == 0)
73 continue;
74
75 if (other_cid == guest_cid)
76 return vsock;
77
78 }
79
80 return NULL;
81}
82
83static void
84vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
85 struct vhost_virtqueue *vq)
86{
87 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
88 int pkts = 0, total_len = 0;
89 bool added = false;
90 bool restart_tx = false;
91
92 mutex_lock(&vq->mutex);
93
94 if (!vq->private_data)
95 goto out;
96
97 /* Avoid further vmexits, we're already processing the virtqueue */
98 vhost_disable_notify(&vsock->dev, vq);
99
100 do {
101 struct virtio_vsock_pkt *pkt;
102 struct iov_iter iov_iter;
103 unsigned out, in;
104 size_t nbytes;
105 size_t iov_len, payload_len;
106 int head;
107
108 spin_lock_bh(&vsock->send_pkt_list_lock);
109 if (list_empty(&vsock->send_pkt_list)) {
110 spin_unlock_bh(&vsock->send_pkt_list_lock);
111 vhost_enable_notify(&vsock->dev, vq);
112 break;
113 }
114
115 pkt = list_first_entry(&vsock->send_pkt_list,
116 struct virtio_vsock_pkt, list);
117 list_del_init(&pkt->list);
118 spin_unlock_bh(&vsock->send_pkt_list_lock);
119
120 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
121 &out, &in, NULL, NULL);
122 if (head < 0) {
123 spin_lock_bh(&vsock->send_pkt_list_lock);
124 list_add(&pkt->list, &vsock->send_pkt_list);
125 spin_unlock_bh(&vsock->send_pkt_list_lock);
126 break;
127 }
128
129 if (head == vq->num) {
130 spin_lock_bh(&vsock->send_pkt_list_lock);
131 list_add(&pkt->list, &vsock->send_pkt_list);
132 spin_unlock_bh(&vsock->send_pkt_list_lock);
133
134 /* We cannot finish yet if more buffers snuck in while
135 * re-enabling notify.
136 */
137 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
138 vhost_disable_notify(&vsock->dev, vq);
139 continue;
140 }
141 break;
142 }
143
144 if (out) {
145 virtio_transport_free_pkt(pkt);
146 vq_err(vq, "Expected 0 output buffers, got %u\n", out);
147 break;
148 }
149
150 iov_len = iov_length(&vq->iov[out], in);
151 if (iov_len < sizeof(pkt->hdr)) {
152 virtio_transport_free_pkt(pkt);
153 vq_err(vq, "Buffer len [%zu] too small\n", iov_len);
154 break;
155 }
156
157 iov_iter_init(&iov_iter, READ, &vq->iov[out], in, iov_len);
158 payload_len = pkt->len - pkt->off;
159
160 /* If the packet is greater than the space available in the
161 * buffer, we split it using multiple buffers.
162 */
163 if (payload_len > iov_len - sizeof(pkt->hdr))
164 payload_len = iov_len - sizeof(pkt->hdr);
165
166 /* Set the correct length in the header */
167 pkt->hdr.len = cpu_to_le32(payload_len);
168
169 nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
170 if (nbytes != sizeof(pkt->hdr)) {
171 virtio_transport_free_pkt(pkt);
172 vq_err(vq, "Faulted on copying pkt hdr\n");
173 break;
174 }
175
176 nbytes = copy_to_iter(pkt->buf + pkt->off, payload_len,
177 &iov_iter);
178 if (nbytes != payload_len) {
179 virtio_transport_free_pkt(pkt);
180 vq_err(vq, "Faulted on copying pkt buf\n");
181 break;
182 }
183
184 vhost_add_used(vq, head, sizeof(pkt->hdr) + payload_len);
185 added = true;
186
187 /* Deliver to monitoring devices all correctly transmitted
188 * packets.
189 */
190 virtio_transport_deliver_tap_pkt(pkt);
191
192 pkt->off += payload_len;
193 total_len += payload_len;
194
195 /* If we didn't send all the payload we can requeue the packet
196 * to send it with the next available buffer.
197 */
198 if (pkt->off < pkt->len) {
199 spin_lock_bh(&vsock->send_pkt_list_lock);
200 list_add(&pkt->list, &vsock->send_pkt_list);
201 spin_unlock_bh(&vsock->send_pkt_list_lock);
202 } else {
203 if (pkt->reply) {
204 int val;
205
206 val = atomic_dec_return(&vsock->queued_replies);
207
208 /* Do we have resources to resume tx
209 * processing?
210 */
211 if (val + 1 == tx_vq->num)
212 restart_tx = true;
213 }
214
215 virtio_transport_free_pkt(pkt);
216 }
217 } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
218 if (added)
219 vhost_signal(&vsock->dev, vq);
220
221out:
222 mutex_unlock(&vq->mutex);
223
224 if (restart_tx)
225 vhost_poll_queue(&tx_vq->poll);
226}
227
228static void vhost_transport_send_pkt_work(struct vhost_work *work)
229{
230 struct vhost_virtqueue *vq;
231 struct vhost_vsock *vsock;
232
233 vsock = container_of(work, struct vhost_vsock, send_pkt_work);
234 vq = &vsock->vqs[VSOCK_VQ_RX];
235
236 vhost_transport_do_send_pkt(vsock, vq);
237}
238
239static int
240vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
241{
242 struct vhost_vsock *vsock;
243 int len = pkt->len;
244
245 rcu_read_lock();
246
247 /* Find the vhost_vsock according to guest context id */
248 vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
249 if (!vsock) {
250 rcu_read_unlock();
251 virtio_transport_free_pkt(pkt);
252 return -ENODEV;
253 }
254
255 if (pkt->reply)
256 atomic_inc(&vsock->queued_replies);
257
258 spin_lock_bh(&vsock->send_pkt_list_lock);
259 list_add_tail(&pkt->list, &vsock->send_pkt_list);
260 spin_unlock_bh(&vsock->send_pkt_list_lock);
261
262 vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
263
264 rcu_read_unlock();
265 return len;
266}
267
268static int
269vhost_transport_cancel_pkt(struct vsock_sock *vsk)
270{
271 struct vhost_vsock *vsock;
272 struct virtio_vsock_pkt *pkt, *n;
273 int cnt = 0;
274 int ret = -ENODEV;
275 LIST_HEAD(freeme);
276
277 rcu_read_lock();
278
279 /* Find the vhost_vsock according to guest context id */
280 vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
281 if (!vsock)
282 goto out;
283
284 spin_lock_bh(&vsock->send_pkt_list_lock);
285 list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
286 if (pkt->vsk != vsk)
287 continue;
288 list_move(&pkt->list, &freeme);
289 }
290 spin_unlock_bh(&vsock->send_pkt_list_lock);
291
292 list_for_each_entry_safe(pkt, n, &freeme, list) {
293 if (pkt->reply)
294 cnt++;
295 list_del(&pkt->list);
296 virtio_transport_free_pkt(pkt);
297 }
298
299 if (cnt) {
300 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
301 int new_cnt;
302
303 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
304 if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
305 vhost_poll_queue(&tx_vq->poll);
306 }
307
308 ret = 0;
309out:
310 rcu_read_unlock();
311 return ret;
312}
313
314static struct virtio_vsock_pkt *
315vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
316 unsigned int out, unsigned int in)
317{
318 struct virtio_vsock_pkt *pkt;
319 struct iov_iter iov_iter;
320 size_t nbytes;
321 size_t len;
322
323 if (in != 0) {
324 vq_err(vq, "Expected 0 input buffers, got %u\n", in);
325 return NULL;
326 }
327
328 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
329 if (!pkt)
330 return NULL;
331
332 len = iov_length(vq->iov, out);
333 iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
334
335 nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
336 if (nbytes != sizeof(pkt->hdr)) {
337 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
338 sizeof(pkt->hdr), nbytes);
339 kfree(pkt);
340 return NULL;
341 }
342
343 if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM)
344 pkt->len = le32_to_cpu(pkt->hdr.len);
345
346 /* No payload */
347 if (!pkt->len)
348 return pkt;
349
350 /* The pkt is too big */
351 if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
352 kfree(pkt);
353 return NULL;
354 }
355
356 pkt->buf = kmalloc(pkt->len, GFP_KERNEL);
357 if (!pkt->buf) {
358 kfree(pkt);
359 return NULL;
360 }
361
362 pkt->buf_len = pkt->len;
363
364 nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
365 if (nbytes != pkt->len) {
366 vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
367 pkt->len, nbytes);
368 virtio_transport_free_pkt(pkt);
369 return NULL;
370 }
371
372 return pkt;
373}
374
375/* Is there space left for replies to rx packets? */
376static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
377{
378 struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
379 int val;
380
381 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
382 val = atomic_read(&vsock->queued_replies);
383
384 return val < vq->num;
385}
386
387static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
388{
389 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
390 poll.work);
391 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
392 dev);
393 struct virtio_vsock_pkt *pkt;
394 int head, pkts = 0, total_len = 0;
395 unsigned int out, in;
396 bool added = false;
397
398 mutex_lock(&vq->mutex);
399
400 if (!vq->private_data)
401 goto out;
402
403 vhost_disable_notify(&vsock->dev, vq);
404 do {
405 u32 len;
406
407 if (!vhost_vsock_more_replies(vsock)) {
408 /* Stop tx until the device processes already
409 * pending replies. Leave tx virtqueue
410 * callbacks disabled.
411 */
412 goto no_more_replies;
413 }
414
415 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
416 &out, &in, NULL, NULL);
417 if (head < 0)
418 break;
419
420 if (head == vq->num) {
421 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
422 vhost_disable_notify(&vsock->dev, vq);
423 continue;
424 }
425 break;
426 }
427
428 pkt = vhost_vsock_alloc_pkt(vq, out, in);
429 if (!pkt) {
430 vq_err(vq, "Faulted on pkt\n");
431 continue;
432 }
433
434 len = pkt->len;
435
436 /* Deliver to monitoring devices all received packets */
437 virtio_transport_deliver_tap_pkt(pkt);
438
439 /* Only accept correctly addressed packets */
440 if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid)
441 virtio_transport_recv_pkt(pkt);
442 else
443 virtio_transport_free_pkt(pkt);
444
445 len += sizeof(pkt->hdr);
446 vhost_add_used(vq, head, len);
447 total_len += len;
448 added = true;
449 } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
450
451no_more_replies:
452 if (added)
453 vhost_signal(&vsock->dev, vq);
454
455out:
456 mutex_unlock(&vq->mutex);
457}
458
459static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
460{
461 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
462 poll.work);
463 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
464 dev);
465
466 vhost_transport_do_send_pkt(vsock, vq);
467}
468
469static int vhost_vsock_start(struct vhost_vsock *vsock)
470{
471 struct vhost_virtqueue *vq;
472 size_t i;
473 int ret;
474
475 mutex_lock(&vsock->dev.mutex);
476
477 ret = vhost_dev_check_owner(&vsock->dev);
478 if (ret)
479 goto err;
480
481 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
482 vq = &vsock->vqs[i];
483
484 mutex_lock(&vq->mutex);
485
486 if (!vhost_vq_access_ok(vq)) {
487 ret = -EFAULT;
488 goto err_vq;
489 }
490
491 if (!vq->private_data) {
492 vq->private_data = vsock;
493 ret = vhost_vq_init_access(vq);
494 if (ret)
495 goto err_vq;
496 }
497
498 mutex_unlock(&vq->mutex);
499 }
500
501 mutex_unlock(&vsock->dev.mutex);
502 return 0;
503
504err_vq:
505 vq->private_data = NULL;
506 mutex_unlock(&vq->mutex);
507
508 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
509 vq = &vsock->vqs[i];
510
511 mutex_lock(&vq->mutex);
512 vq->private_data = NULL;
513 mutex_unlock(&vq->mutex);
514 }
515err:
516 mutex_unlock(&vsock->dev.mutex);
517 return ret;
518}
519
520static int vhost_vsock_stop(struct vhost_vsock *vsock)
521{
522 size_t i;
523 int ret;
524
525 mutex_lock(&vsock->dev.mutex);
526
527 ret = vhost_dev_check_owner(&vsock->dev);
528 if (ret)
529 goto err;
530
531 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
532 struct vhost_virtqueue *vq = &vsock->vqs[i];
533
534 mutex_lock(&vq->mutex);
535 vq->private_data = NULL;
536 mutex_unlock(&vq->mutex);
537 }
538
539err:
540 mutex_unlock(&vsock->dev.mutex);
541 return ret;
542}
543
544static void vhost_vsock_free(struct vhost_vsock *vsock)
545{
546 kvfree(vsock);
547}
548
549static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
550{
551 struct vhost_virtqueue **vqs;
552 struct vhost_vsock *vsock;
553 int ret;
554
555 /* This struct is large and allocation could fail, fall back to vmalloc
556 * if there is no other way.
557 */
558 vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
559 if (!vsock)
560 return -ENOMEM;
561
562 vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
563 if (!vqs) {
564 ret = -ENOMEM;
565 goto out;
566 }
567
568 vsock->guest_cid = 0; /* no CID assigned yet */
569
570 atomic_set(&vsock->queued_replies, 0);
571
572 vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
573 vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
574 vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
575 vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
576
577 vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
578 UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
579 VHOST_VSOCK_WEIGHT);
580
581 file->private_data = vsock;
582 spin_lock_init(&vsock->send_pkt_list_lock);
583 INIT_LIST_HEAD(&vsock->send_pkt_list);
584 vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
585 return 0;
586
587out:
588 vhost_vsock_free(vsock);
589 return ret;
590}
591
592static void vhost_vsock_flush(struct vhost_vsock *vsock)
593{
594 int i;
595
596 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
597 if (vsock->vqs[i].handle_kick)
598 vhost_poll_flush(&vsock->vqs[i].poll);
599 vhost_work_flush(&vsock->dev, &vsock->send_pkt_work);
600}
601
602static void vhost_vsock_reset_orphans(struct sock *sk)
603{
604 struct vsock_sock *vsk = vsock_sk(sk);
605
606 /* vmci_transport.c doesn't take sk_lock here either. At least we're
607 * under vsock_table_lock so the sock cannot disappear while we're
608 * executing.
609 */
610
611 /* If the peer is still valid, no need to reset connection */
612 if (vhost_vsock_get(vsk->remote_addr.svm_cid))
613 return;
614
615 /* If the close timeout is pending, let it expire. This avoids races
616 * with the timeout callback.
617 */
618 if (vsk->close_work_scheduled)
619 return;
620
621 sock_set_flag(sk, SOCK_DONE);
622 vsk->peer_shutdown = SHUTDOWN_MASK;
623 sk->sk_state = SS_UNCONNECTED;
624 sk->sk_err = ECONNRESET;
625 sk->sk_error_report(sk);
626}
627
628static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
629{
630 struct vhost_vsock *vsock = file->private_data;
631
632 mutex_lock(&vhost_vsock_mutex);
633 if (vsock->guest_cid)
634 hash_del_rcu(&vsock->hash);
635 mutex_unlock(&vhost_vsock_mutex);
636
637 /* Wait for other CPUs to finish using vsock */
638 synchronize_rcu();
639
640 /* Iterating over all connections for all CIDs to find orphans is
641 * inefficient. Room for improvement here. */
642 vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
643
644 vhost_vsock_stop(vsock);
645 vhost_vsock_flush(vsock);
646 vhost_dev_stop(&vsock->dev);
647
648 spin_lock_bh(&vsock->send_pkt_list_lock);
649 while (!list_empty(&vsock->send_pkt_list)) {
650 struct virtio_vsock_pkt *pkt;
651
652 pkt = list_first_entry(&vsock->send_pkt_list,
653 struct virtio_vsock_pkt, list);
654 list_del_init(&pkt->list);
655 virtio_transport_free_pkt(pkt);
656 }
657 spin_unlock_bh(&vsock->send_pkt_list_lock);
658
659 vhost_dev_cleanup(&vsock->dev);
660 kfree(vsock->dev.vqs);
661 vhost_vsock_free(vsock);
662 return 0;
663}
664
665static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
666{
667 struct vhost_vsock *other;
668
669 /* Refuse reserved CIDs */
670 if (guest_cid <= VMADDR_CID_HOST ||
671 guest_cid == U32_MAX)
672 return -EINVAL;
673
674 /* 64-bit CIDs are not yet supported */
675 if (guest_cid > U32_MAX)
676 return -EINVAL;
677
678 /* Refuse if CID is already in use */
679 mutex_lock(&vhost_vsock_mutex);
680 other = vhost_vsock_get(guest_cid);
681 if (other && other != vsock) {
682 mutex_unlock(&vhost_vsock_mutex);
683 return -EADDRINUSE;
684 }
685
686 if (vsock->guest_cid)
687 hash_del_rcu(&vsock->hash);
688
689 vsock->guest_cid = guest_cid;
690 hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
691 mutex_unlock(&vhost_vsock_mutex);
692
693 return 0;
694}
695
696static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
697{
698 struct vhost_virtqueue *vq;
699 int i;
700
701 if (features & ~VHOST_VSOCK_FEATURES)
702 return -EOPNOTSUPP;
703
704 mutex_lock(&vsock->dev.mutex);
705 if ((features & (1 << VHOST_F_LOG_ALL)) &&
706 !vhost_log_access_ok(&vsock->dev)) {
707 mutex_unlock(&vsock->dev.mutex);
708 return -EFAULT;
709 }
710
711 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
712 vq = &vsock->vqs[i];
713 mutex_lock(&vq->mutex);
714 vq->acked_features = features;
715 mutex_unlock(&vq->mutex);
716 }
717 mutex_unlock(&vsock->dev.mutex);
718 return 0;
719}
720
721static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
722 unsigned long arg)
723{
724 struct vhost_vsock *vsock = f->private_data;
725 void __user *argp = (void __user *)arg;
726 u64 guest_cid;
727 u64 features;
728 int start;
729 int r;
730
731 switch (ioctl) {
732 case VHOST_VSOCK_SET_GUEST_CID:
733 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
734 return -EFAULT;
735 return vhost_vsock_set_cid(vsock, guest_cid);
736 case VHOST_VSOCK_SET_RUNNING:
737 if (copy_from_user(&start, argp, sizeof(start)))
738 return -EFAULT;
739 if (start)
740 return vhost_vsock_start(vsock);
741 else
742 return vhost_vsock_stop(vsock);
743 case VHOST_GET_FEATURES:
744 features = VHOST_VSOCK_FEATURES;
745 if (copy_to_user(argp, &features, sizeof(features)))
746 return -EFAULT;
747 return 0;
748 case VHOST_SET_FEATURES:
749 if (copy_from_user(&features, argp, sizeof(features)))
750 return -EFAULT;
751 return vhost_vsock_set_features(vsock, features);
752 default:
753 mutex_lock(&vsock->dev.mutex);
754 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
755 if (r == -ENOIOCTLCMD)
756 r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
757 else
758 vhost_vsock_flush(vsock);
759 mutex_unlock(&vsock->dev.mutex);
760 return r;
761 }
762}
763
764#ifdef CONFIG_COMPAT
765static long vhost_vsock_dev_compat_ioctl(struct file *f, unsigned int ioctl,
766 unsigned long arg)
767{
768 return vhost_vsock_dev_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
769}
770#endif
771
772static const struct file_operations vhost_vsock_fops = {
773 .owner = THIS_MODULE,
774 .open = vhost_vsock_dev_open,
775 .release = vhost_vsock_dev_release,
776 .llseek = noop_llseek,
777 .unlocked_ioctl = vhost_vsock_dev_ioctl,
778#ifdef CONFIG_COMPAT
779 .compat_ioctl = vhost_vsock_dev_compat_ioctl,
780#endif
781};
782
783static struct miscdevice vhost_vsock_misc = {
784 .minor = VHOST_VSOCK_MINOR,
785 .name = "vhost-vsock",
786 .fops = &vhost_vsock_fops,
787};
788
789static struct virtio_transport vhost_transport = {
790 .transport = {
791 .get_local_cid = vhost_transport_get_local_cid,
792
793 .init = virtio_transport_do_socket_init,
794 .destruct = virtio_transport_destruct,
795 .release = virtio_transport_release,
796 .connect = virtio_transport_connect,
797 .shutdown = virtio_transport_shutdown,
798 .cancel_pkt = vhost_transport_cancel_pkt,
799
800 .dgram_enqueue = virtio_transport_dgram_enqueue,
801 .dgram_dequeue = virtio_transport_dgram_dequeue,
802 .dgram_bind = virtio_transport_dgram_bind,
803 .dgram_allow = virtio_transport_dgram_allow,
804
805 .stream_enqueue = virtio_transport_stream_enqueue,
806 .stream_dequeue = virtio_transport_stream_dequeue,
807 .stream_has_data = virtio_transport_stream_has_data,
808 .stream_has_space = virtio_transport_stream_has_space,
809 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
810 .stream_is_active = virtio_transport_stream_is_active,
811 .stream_allow = virtio_transport_stream_allow,
812
813 .notify_poll_in = virtio_transport_notify_poll_in,
814 .notify_poll_out = virtio_transport_notify_poll_out,
815 .notify_recv_init = virtio_transport_notify_recv_init,
816 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
817 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
818 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
819 .notify_send_init = virtio_transport_notify_send_init,
820 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
821 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
822 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
823
824 .set_buffer_size = virtio_transport_set_buffer_size,
825 .set_min_buffer_size = virtio_transport_set_min_buffer_size,
826 .set_max_buffer_size = virtio_transport_set_max_buffer_size,
827 .get_buffer_size = virtio_transport_get_buffer_size,
828 .get_min_buffer_size = virtio_transport_get_min_buffer_size,
829 .get_max_buffer_size = virtio_transport_get_max_buffer_size,
830 },
831
832 .send_pkt = vhost_transport_send_pkt,
833};
834
835static int __init vhost_vsock_init(void)
836{
837 int ret;
838
839 ret = vsock_core_init(&vhost_transport.transport);
840 if (ret < 0)
841 return ret;
842 return misc_register(&vhost_vsock_misc);
843};
844
845static void __exit vhost_vsock_exit(void)
846{
847 misc_deregister(&vhost_vsock_misc);
848 vsock_core_exit();
849};
850
851module_init(vhost_vsock_init);
852module_exit(vhost_vsock_exit);
853MODULE_LICENSE("GPL v2");
854MODULE_AUTHOR("Asias He");
855MODULE_DESCRIPTION("vhost transport for vsock ");
856MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
857MODULE_ALIAS("devname:vhost-vsock");