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 (1ULL << VIRTIO_F_ACCESS_PLATFORM) |
35 (1ULL << VIRTIO_VSOCK_F_SEQPACKET)
36};
37
38enum {
39 VHOST_VSOCK_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)
40};
41
42/* Used to track all the vhost_vsock instances on the system. */
43static DEFINE_MUTEX(vhost_vsock_mutex);
44static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
45
46struct vhost_vsock {
47 struct vhost_dev dev;
48 struct vhost_virtqueue vqs[2];
49
50 /* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */
51 struct hlist_node hash;
52
53 struct vhost_work send_pkt_work;
54 struct sk_buff_head send_pkt_queue; /* host->guest pending packets */
55
56 atomic_t queued_replies;
57
58 u32 guest_cid;
59 bool seqpacket_allow;
60};
61
62static u32 vhost_transport_get_local_cid(void)
63{
64 return VHOST_VSOCK_DEFAULT_HOST_CID;
65}
66
67/* Callers that dereference the return value must hold vhost_vsock_mutex or the
68 * RCU read lock.
69 */
70static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
71{
72 struct vhost_vsock *vsock;
73
74 hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
75 u32 other_cid = vsock->guest_cid;
76
77 /* Skip instances that have no CID yet */
78 if (other_cid == 0)
79 continue;
80
81 if (other_cid == guest_cid)
82 return vsock;
83
84 }
85
86 return NULL;
87}
88
89static void
90vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
91 struct vhost_virtqueue *vq)
92{
93 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
94 int pkts = 0, total_len = 0;
95 bool added = false;
96 bool restart_tx = false;
97
98 mutex_lock(&vq->mutex);
99
100 if (!vhost_vq_get_backend(vq))
101 goto out;
102
103 if (!vq_meta_prefetch(vq))
104 goto out;
105
106 /* Avoid further vmexits, we're already processing the virtqueue */
107 vhost_disable_notify(&vsock->dev, vq);
108
109 do {
110 struct virtio_vsock_hdr *hdr;
111 size_t iov_len, payload_len;
112 struct iov_iter iov_iter;
113 u32 flags_to_restore = 0;
114 struct sk_buff *skb;
115 unsigned out, in;
116 size_t nbytes;
117 u32 offset;
118 int head;
119
120 skb = virtio_vsock_skb_dequeue(&vsock->send_pkt_queue);
121
122 if (!skb) {
123 vhost_enable_notify(&vsock->dev, vq);
124 break;
125 }
126
127 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
128 &out, &in, NULL, NULL);
129 if (head < 0) {
130 virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
131 break;
132 }
133
134 if (head == vq->num) {
135 virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
136 /* We cannot finish yet if more buffers snuck in while
137 * re-enabling notify.
138 */
139 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
140 vhost_disable_notify(&vsock->dev, vq);
141 continue;
142 }
143 break;
144 }
145
146 if (out) {
147 kfree_skb(skb);
148 vq_err(vq, "Expected 0 output buffers, got %u\n", out);
149 break;
150 }
151
152 iov_len = iov_length(&vq->iov[out], in);
153 if (iov_len < sizeof(*hdr)) {
154 kfree_skb(skb);
155 vq_err(vq, "Buffer len [%zu] too small\n", iov_len);
156 break;
157 }
158
159 iov_iter_init(&iov_iter, ITER_DEST, &vq->iov[out], in, iov_len);
160 offset = VIRTIO_VSOCK_SKB_CB(skb)->offset;
161 payload_len = skb->len - offset;
162 hdr = virtio_vsock_hdr(skb);
163
164 /* If the packet is greater than the space available in the
165 * buffer, we split it using multiple buffers.
166 */
167 if (payload_len > iov_len - sizeof(*hdr)) {
168 payload_len = iov_len - sizeof(*hdr);
169
170 /* As we are copying pieces of large packet's buffer to
171 * small rx buffers, headers of packets in rx queue are
172 * created dynamically and are initialized with header
173 * of current packet(except length). But in case of
174 * SOCK_SEQPACKET, we also must clear message delimeter
175 * bit (VIRTIO_VSOCK_SEQ_EOM) and MSG_EOR bit
176 * (VIRTIO_VSOCK_SEQ_EOR) if set. Otherwise,
177 * there will be sequence of packets with these
178 * bits set. After initialized header will be copied to
179 * rx buffer, these required bits will be restored.
180 */
181 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
182 hdr->flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM);
183 flags_to_restore |= VIRTIO_VSOCK_SEQ_EOM;
184
185 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) {
186 hdr->flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
187 flags_to_restore |= VIRTIO_VSOCK_SEQ_EOR;
188 }
189 }
190 }
191
192 /* Set the correct length in the header */
193 hdr->len = cpu_to_le32(payload_len);
194
195 nbytes = copy_to_iter(hdr, sizeof(*hdr), &iov_iter);
196 if (nbytes != sizeof(*hdr)) {
197 kfree_skb(skb);
198 vq_err(vq, "Faulted on copying pkt hdr\n");
199 break;
200 }
201
202 if (skb_copy_datagram_iter(skb,
203 offset,
204 &iov_iter,
205 payload_len)) {
206 kfree_skb(skb);
207 vq_err(vq, "Faulted on copying pkt buf\n");
208 break;
209 }
210
211 /* Deliver to monitoring devices all packets that we
212 * will transmit.
213 */
214 virtio_transport_deliver_tap_pkt(skb);
215
216 vhost_add_used(vq, head, sizeof(*hdr) + payload_len);
217 added = true;
218
219 VIRTIO_VSOCK_SKB_CB(skb)->offset += payload_len;
220 total_len += payload_len;
221
222 /* If we didn't send all the payload we can requeue the packet
223 * to send it with the next available buffer.
224 */
225 if (VIRTIO_VSOCK_SKB_CB(skb)->offset < skb->len) {
226 hdr->flags |= cpu_to_le32(flags_to_restore);
227
228 /* We are queueing the same skb to handle
229 * the remaining bytes, and we want to deliver it
230 * to monitoring devices in the next iteration.
231 */
232 virtio_vsock_skb_clear_tap_delivered(skb);
233 virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
234 } else {
235 if (virtio_vsock_skb_reply(skb)) {
236 int val;
237
238 val = atomic_dec_return(&vsock->queued_replies);
239
240 /* Do we have resources to resume tx
241 * processing?
242 */
243 if (val + 1 == tx_vq->num)
244 restart_tx = true;
245 }
246
247 consume_skb(skb);
248 }
249 } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
250 if (added)
251 vhost_signal(&vsock->dev, vq);
252
253out:
254 mutex_unlock(&vq->mutex);
255
256 if (restart_tx)
257 vhost_poll_queue(&tx_vq->poll);
258}
259
260static void vhost_transport_send_pkt_work(struct vhost_work *work)
261{
262 struct vhost_virtqueue *vq;
263 struct vhost_vsock *vsock;
264
265 vsock = container_of(work, struct vhost_vsock, send_pkt_work);
266 vq = &vsock->vqs[VSOCK_VQ_RX];
267
268 vhost_transport_do_send_pkt(vsock, vq);
269}
270
271static int
272vhost_transport_send_pkt(struct sk_buff *skb)
273{
274 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
275 struct vhost_vsock *vsock;
276 int len = skb->len;
277
278 rcu_read_lock();
279
280 /* Find the vhost_vsock according to guest context id */
281 vsock = vhost_vsock_get(le64_to_cpu(hdr->dst_cid));
282 if (!vsock) {
283 rcu_read_unlock();
284 kfree_skb(skb);
285 return -ENODEV;
286 }
287
288 if (virtio_vsock_skb_reply(skb))
289 atomic_inc(&vsock->queued_replies);
290
291 virtio_vsock_skb_queue_tail(&vsock->send_pkt_queue, skb);
292 vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &vsock->send_pkt_work);
293
294 rcu_read_unlock();
295 return len;
296}
297
298static int
299vhost_transport_cancel_pkt(struct vsock_sock *vsk)
300{
301 struct vhost_vsock *vsock;
302 int cnt = 0;
303 int ret = -ENODEV;
304
305 rcu_read_lock();
306
307 /* Find the vhost_vsock according to guest context id */
308 vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
309 if (!vsock)
310 goto out;
311
312 cnt = virtio_transport_purge_skbs(vsk, &vsock->send_pkt_queue);
313
314 if (cnt) {
315 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
316 int new_cnt;
317
318 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
319 if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
320 vhost_poll_queue(&tx_vq->poll);
321 }
322
323 ret = 0;
324out:
325 rcu_read_unlock();
326 return ret;
327}
328
329static struct sk_buff *
330vhost_vsock_alloc_skb(struct vhost_virtqueue *vq,
331 unsigned int out, unsigned int in)
332{
333 struct virtio_vsock_hdr *hdr;
334 struct iov_iter iov_iter;
335 struct sk_buff *skb;
336 size_t payload_len;
337 size_t nbytes;
338 size_t len;
339
340 if (in != 0) {
341 vq_err(vq, "Expected 0 input buffers, got %u\n", in);
342 return NULL;
343 }
344
345 len = iov_length(vq->iov, out);
346
347 /* len contains both payload and hdr */
348 skb = virtio_vsock_alloc_skb(len, GFP_KERNEL);
349 if (!skb)
350 return NULL;
351
352 iov_iter_init(&iov_iter, ITER_SOURCE, vq->iov, out, len);
353
354 hdr = virtio_vsock_hdr(skb);
355 nbytes = copy_from_iter(hdr, sizeof(*hdr), &iov_iter);
356 if (nbytes != sizeof(*hdr)) {
357 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
358 sizeof(*hdr), nbytes);
359 kfree_skb(skb);
360 return NULL;
361 }
362
363 payload_len = le32_to_cpu(hdr->len);
364
365 /* No payload */
366 if (!payload_len)
367 return skb;
368
369 /* The pkt is too big or the length in the header is invalid */
370 if (payload_len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE ||
371 payload_len + sizeof(*hdr) > len) {
372 kfree_skb(skb);
373 return NULL;
374 }
375
376 virtio_vsock_skb_rx_put(skb);
377
378 nbytes = copy_from_iter(skb->data, payload_len, &iov_iter);
379 if (nbytes != payload_len) {
380 vq_err(vq, "Expected %zu byte payload, got %zu bytes\n",
381 payload_len, nbytes);
382 kfree_skb(skb);
383 return NULL;
384 }
385
386 return skb;
387}
388
389/* Is there space left for replies to rx packets? */
390static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
391{
392 struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
393 int val;
394
395 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
396 val = atomic_read(&vsock->queued_replies);
397
398 return val < vq->num;
399}
400
401static bool vhost_transport_msgzerocopy_allow(void)
402{
403 return true;
404}
405
406static bool vhost_transport_seqpacket_allow(u32 remote_cid);
407
408static struct virtio_transport vhost_transport = {
409 .transport = {
410 .module = THIS_MODULE,
411
412 .get_local_cid = vhost_transport_get_local_cid,
413
414 .init = virtio_transport_do_socket_init,
415 .destruct = virtio_transport_destruct,
416 .release = virtio_transport_release,
417 .connect = virtio_transport_connect,
418 .shutdown = virtio_transport_shutdown,
419 .cancel_pkt = vhost_transport_cancel_pkt,
420
421 .dgram_enqueue = virtio_transport_dgram_enqueue,
422 .dgram_dequeue = virtio_transport_dgram_dequeue,
423 .dgram_bind = virtio_transport_dgram_bind,
424 .dgram_allow = virtio_transport_dgram_allow,
425
426 .stream_enqueue = virtio_transport_stream_enqueue,
427 .stream_dequeue = virtio_transport_stream_dequeue,
428 .stream_has_data = virtio_transport_stream_has_data,
429 .stream_has_space = virtio_transport_stream_has_space,
430 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
431 .stream_is_active = virtio_transport_stream_is_active,
432 .stream_allow = virtio_transport_stream_allow,
433
434 .seqpacket_dequeue = virtio_transport_seqpacket_dequeue,
435 .seqpacket_enqueue = virtio_transport_seqpacket_enqueue,
436 .seqpacket_allow = vhost_transport_seqpacket_allow,
437 .seqpacket_has_data = virtio_transport_seqpacket_has_data,
438
439 .msgzerocopy_allow = vhost_transport_msgzerocopy_allow,
440
441 .notify_poll_in = virtio_transport_notify_poll_in,
442 .notify_poll_out = virtio_transport_notify_poll_out,
443 .notify_recv_init = virtio_transport_notify_recv_init,
444 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
445 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
446 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
447 .notify_send_init = virtio_transport_notify_send_init,
448 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
449 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
450 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
451 .notify_buffer_size = virtio_transport_notify_buffer_size,
452 .notify_set_rcvlowat = virtio_transport_notify_set_rcvlowat,
453
454 .read_skb = virtio_transport_read_skb,
455 },
456
457 .send_pkt = vhost_transport_send_pkt,
458};
459
460static bool vhost_transport_seqpacket_allow(u32 remote_cid)
461{
462 struct vhost_vsock *vsock;
463 bool seqpacket_allow = false;
464
465 rcu_read_lock();
466 vsock = vhost_vsock_get(remote_cid);
467
468 if (vsock)
469 seqpacket_allow = vsock->seqpacket_allow;
470
471 rcu_read_unlock();
472
473 return seqpacket_allow;
474}
475
476static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
477{
478 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
479 poll.work);
480 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
481 dev);
482 int head, pkts = 0, total_len = 0;
483 unsigned int out, in;
484 struct sk_buff *skb;
485 bool added = false;
486
487 mutex_lock(&vq->mutex);
488
489 if (!vhost_vq_get_backend(vq))
490 goto out;
491
492 if (!vq_meta_prefetch(vq))
493 goto out;
494
495 vhost_disable_notify(&vsock->dev, vq);
496 do {
497 struct virtio_vsock_hdr *hdr;
498
499 if (!vhost_vsock_more_replies(vsock)) {
500 /* Stop tx until the device processes already
501 * pending replies. Leave tx virtqueue
502 * callbacks disabled.
503 */
504 goto no_more_replies;
505 }
506
507 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
508 &out, &in, NULL, NULL);
509 if (head < 0)
510 break;
511
512 if (head == vq->num) {
513 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
514 vhost_disable_notify(&vsock->dev, vq);
515 continue;
516 }
517 break;
518 }
519
520 skb = vhost_vsock_alloc_skb(vq, out, in);
521 if (!skb) {
522 vq_err(vq, "Faulted on pkt\n");
523 continue;
524 }
525
526 total_len += sizeof(*hdr) + skb->len;
527
528 /* Deliver to monitoring devices all received packets */
529 virtio_transport_deliver_tap_pkt(skb);
530
531 hdr = virtio_vsock_hdr(skb);
532
533 /* Only accept correctly addressed packets */
534 if (le64_to_cpu(hdr->src_cid) == vsock->guest_cid &&
535 le64_to_cpu(hdr->dst_cid) ==
536 vhost_transport_get_local_cid())
537 virtio_transport_recv_pkt(&vhost_transport, skb);
538 else
539 kfree_skb(skb);
540
541 vhost_add_used(vq, head, 0);
542 added = true;
543 } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
544
545no_more_replies:
546 if (added)
547 vhost_signal(&vsock->dev, vq);
548
549out:
550 mutex_unlock(&vq->mutex);
551}
552
553static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
554{
555 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
556 poll.work);
557 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
558 dev);
559
560 vhost_transport_do_send_pkt(vsock, vq);
561}
562
563static int vhost_vsock_start(struct vhost_vsock *vsock)
564{
565 struct vhost_virtqueue *vq;
566 size_t i;
567 int ret;
568
569 mutex_lock(&vsock->dev.mutex);
570
571 ret = vhost_dev_check_owner(&vsock->dev);
572 if (ret)
573 goto err;
574
575 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
576 vq = &vsock->vqs[i];
577
578 mutex_lock(&vq->mutex);
579
580 if (!vhost_vq_access_ok(vq)) {
581 ret = -EFAULT;
582 goto err_vq;
583 }
584
585 if (!vhost_vq_get_backend(vq)) {
586 vhost_vq_set_backend(vq, vsock);
587 ret = vhost_vq_init_access(vq);
588 if (ret)
589 goto err_vq;
590 }
591
592 mutex_unlock(&vq->mutex);
593 }
594
595 /* Some packets may have been queued before the device was started,
596 * let's kick the send worker to send them.
597 */
598 vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &vsock->send_pkt_work);
599
600 mutex_unlock(&vsock->dev.mutex);
601 return 0;
602
603err_vq:
604 vhost_vq_set_backend(vq, NULL);
605 mutex_unlock(&vq->mutex);
606
607 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
608 vq = &vsock->vqs[i];
609
610 mutex_lock(&vq->mutex);
611 vhost_vq_set_backend(vq, NULL);
612 mutex_unlock(&vq->mutex);
613 }
614err:
615 mutex_unlock(&vsock->dev.mutex);
616 return ret;
617}
618
619static int vhost_vsock_stop(struct vhost_vsock *vsock, bool check_owner)
620{
621 size_t i;
622 int ret = 0;
623
624 mutex_lock(&vsock->dev.mutex);
625
626 if (check_owner) {
627 ret = vhost_dev_check_owner(&vsock->dev);
628 if (ret)
629 goto err;
630 }
631
632 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
633 struct vhost_virtqueue *vq = &vsock->vqs[i];
634
635 mutex_lock(&vq->mutex);
636 vhost_vq_set_backend(vq, NULL);
637 mutex_unlock(&vq->mutex);
638 }
639
640err:
641 mutex_unlock(&vsock->dev.mutex);
642 return ret;
643}
644
645static void vhost_vsock_free(struct vhost_vsock *vsock)
646{
647 kvfree(vsock);
648}
649
650static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
651{
652 struct vhost_virtqueue **vqs;
653 struct vhost_vsock *vsock;
654 int ret;
655
656 /* This struct is large and allocation could fail, fall back to vmalloc
657 * if there is no other way.
658 */
659 vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
660 if (!vsock)
661 return -ENOMEM;
662
663 vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
664 if (!vqs) {
665 ret = -ENOMEM;
666 goto out;
667 }
668
669 vsock->guest_cid = 0; /* no CID assigned yet */
670
671 atomic_set(&vsock->queued_replies, 0);
672
673 vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
674 vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
675 vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
676 vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
677
678 vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
679 UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
680 VHOST_VSOCK_WEIGHT, true, NULL);
681
682 file->private_data = vsock;
683 skb_queue_head_init(&vsock->send_pkt_queue);
684 vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
685 return 0;
686
687out:
688 vhost_vsock_free(vsock);
689 return ret;
690}
691
692static void vhost_vsock_flush(struct vhost_vsock *vsock)
693{
694 vhost_dev_flush(&vsock->dev);
695}
696
697static void vhost_vsock_reset_orphans(struct sock *sk)
698{
699 struct vsock_sock *vsk = vsock_sk(sk);
700
701 /* vmci_transport.c doesn't take sk_lock here either. At least we're
702 * under vsock_table_lock so the sock cannot disappear while we're
703 * executing.
704 */
705
706 /* If the peer is still valid, no need to reset connection */
707 if (vhost_vsock_get(vsk->remote_addr.svm_cid))
708 return;
709
710 /* If the close timeout is pending, let it expire. This avoids races
711 * with the timeout callback.
712 */
713 if (vsk->close_work_scheduled)
714 return;
715
716 sock_set_flag(sk, SOCK_DONE);
717 vsk->peer_shutdown = SHUTDOWN_MASK;
718 sk->sk_state = SS_UNCONNECTED;
719 sk->sk_err = ECONNRESET;
720 sk_error_report(sk);
721}
722
723static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
724{
725 struct vhost_vsock *vsock = file->private_data;
726
727 mutex_lock(&vhost_vsock_mutex);
728 if (vsock->guest_cid)
729 hash_del_rcu(&vsock->hash);
730 mutex_unlock(&vhost_vsock_mutex);
731
732 /* Wait for other CPUs to finish using vsock */
733 synchronize_rcu();
734
735 /* Iterating over all connections for all CIDs to find orphans is
736 * inefficient. Room for improvement here. */
737 vsock_for_each_connected_socket(&vhost_transport.transport,
738 vhost_vsock_reset_orphans);
739
740 /* Don't check the owner, because we are in the release path, so we
741 * need to stop the vsock device in any case.
742 * vhost_vsock_stop() can not fail in this case, so we don't need to
743 * check the return code.
744 */
745 vhost_vsock_stop(vsock, false);
746 vhost_vsock_flush(vsock);
747 vhost_dev_stop(&vsock->dev);
748
749 virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue);
750
751 vhost_dev_cleanup(&vsock->dev);
752 kfree(vsock->dev.vqs);
753 vhost_vsock_free(vsock);
754 return 0;
755}
756
757static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
758{
759 struct vhost_vsock *other;
760
761 /* Refuse reserved CIDs */
762 if (guest_cid <= VMADDR_CID_HOST ||
763 guest_cid == U32_MAX)
764 return -EINVAL;
765
766 /* 64-bit CIDs are not yet supported */
767 if (guest_cid > U32_MAX)
768 return -EINVAL;
769
770 /* Refuse if CID is assigned to the guest->host transport (i.e. nested
771 * VM), to make the loopback work.
772 */
773 if (vsock_find_cid(guest_cid))
774 return -EADDRINUSE;
775
776 /* Refuse if CID is already in use */
777 mutex_lock(&vhost_vsock_mutex);
778 other = vhost_vsock_get(guest_cid);
779 if (other && other != vsock) {
780 mutex_unlock(&vhost_vsock_mutex);
781 return -EADDRINUSE;
782 }
783
784 if (vsock->guest_cid)
785 hash_del_rcu(&vsock->hash);
786
787 vsock->guest_cid = guest_cid;
788 hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
789 mutex_unlock(&vhost_vsock_mutex);
790
791 return 0;
792}
793
794static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
795{
796 struct vhost_virtqueue *vq;
797 int i;
798
799 if (features & ~VHOST_VSOCK_FEATURES)
800 return -EOPNOTSUPP;
801
802 mutex_lock(&vsock->dev.mutex);
803 if ((features & (1 << VHOST_F_LOG_ALL)) &&
804 !vhost_log_access_ok(&vsock->dev)) {
805 goto err;
806 }
807
808 if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) {
809 if (vhost_init_device_iotlb(&vsock->dev))
810 goto err;
811 }
812
813 if (features & (1ULL << VIRTIO_VSOCK_F_SEQPACKET))
814 vsock->seqpacket_allow = true;
815
816 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
817 vq = &vsock->vqs[i];
818 mutex_lock(&vq->mutex);
819 vq->acked_features = features;
820 mutex_unlock(&vq->mutex);
821 }
822 mutex_unlock(&vsock->dev.mutex);
823 return 0;
824
825err:
826 mutex_unlock(&vsock->dev.mutex);
827 return -EFAULT;
828}
829
830static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
831 unsigned long arg)
832{
833 struct vhost_vsock *vsock = f->private_data;
834 void __user *argp = (void __user *)arg;
835 u64 guest_cid;
836 u64 features;
837 int start;
838 int r;
839
840 switch (ioctl) {
841 case VHOST_VSOCK_SET_GUEST_CID:
842 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
843 return -EFAULT;
844 return vhost_vsock_set_cid(vsock, guest_cid);
845 case VHOST_VSOCK_SET_RUNNING:
846 if (copy_from_user(&start, argp, sizeof(start)))
847 return -EFAULT;
848 if (start)
849 return vhost_vsock_start(vsock);
850 else
851 return vhost_vsock_stop(vsock, true);
852 case VHOST_GET_FEATURES:
853 features = VHOST_VSOCK_FEATURES;
854 if (copy_to_user(argp, &features, sizeof(features)))
855 return -EFAULT;
856 return 0;
857 case VHOST_SET_FEATURES:
858 if (copy_from_user(&features, argp, sizeof(features)))
859 return -EFAULT;
860 return vhost_vsock_set_features(vsock, features);
861 case VHOST_GET_BACKEND_FEATURES:
862 features = VHOST_VSOCK_BACKEND_FEATURES;
863 if (copy_to_user(argp, &features, sizeof(features)))
864 return -EFAULT;
865 return 0;
866 case VHOST_SET_BACKEND_FEATURES:
867 if (copy_from_user(&features, argp, sizeof(features)))
868 return -EFAULT;
869 if (features & ~VHOST_VSOCK_BACKEND_FEATURES)
870 return -EOPNOTSUPP;
871 vhost_set_backend_features(&vsock->dev, features);
872 return 0;
873 default:
874 mutex_lock(&vsock->dev.mutex);
875 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
876 if (r == -ENOIOCTLCMD)
877 r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
878 else
879 vhost_vsock_flush(vsock);
880 mutex_unlock(&vsock->dev.mutex);
881 return r;
882 }
883}
884
885static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
886{
887 struct file *file = iocb->ki_filp;
888 struct vhost_vsock *vsock = file->private_data;
889 struct vhost_dev *dev = &vsock->dev;
890 int noblock = file->f_flags & O_NONBLOCK;
891
892 return vhost_chr_read_iter(dev, to, noblock);
893}
894
895static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb,
896 struct iov_iter *from)
897{
898 struct file *file = iocb->ki_filp;
899 struct vhost_vsock *vsock = file->private_data;
900 struct vhost_dev *dev = &vsock->dev;
901
902 return vhost_chr_write_iter(dev, from);
903}
904
905static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait)
906{
907 struct vhost_vsock *vsock = file->private_data;
908 struct vhost_dev *dev = &vsock->dev;
909
910 return vhost_chr_poll(file, dev, wait);
911}
912
913static const struct file_operations vhost_vsock_fops = {
914 .owner = THIS_MODULE,
915 .open = vhost_vsock_dev_open,
916 .release = vhost_vsock_dev_release,
917 .llseek = noop_llseek,
918 .unlocked_ioctl = vhost_vsock_dev_ioctl,
919 .compat_ioctl = compat_ptr_ioctl,
920 .read_iter = vhost_vsock_chr_read_iter,
921 .write_iter = vhost_vsock_chr_write_iter,
922 .poll = vhost_vsock_chr_poll,
923};
924
925static struct miscdevice vhost_vsock_misc = {
926 .minor = VHOST_VSOCK_MINOR,
927 .name = "vhost-vsock",
928 .fops = &vhost_vsock_fops,
929};
930
931static int __init vhost_vsock_init(void)
932{
933 int ret;
934
935 ret = vsock_core_register(&vhost_transport.transport,
936 VSOCK_TRANSPORT_F_H2G);
937 if (ret < 0)
938 return ret;
939
940 ret = misc_register(&vhost_vsock_misc);
941 if (ret) {
942 vsock_core_unregister(&vhost_transport.transport);
943 return ret;
944 }
945
946 return 0;
947};
948
949static void __exit vhost_vsock_exit(void)
950{
951 misc_deregister(&vhost_vsock_misc);
952 vsock_core_unregister(&vhost_transport.transport);
953};
954
955module_init(vhost_vsock_init);
956module_exit(vhost_vsock_exit);
957MODULE_LICENSE("GPL v2");
958MODULE_AUTHOR("Asias He");
959MODULE_DESCRIPTION("vhost transport for vsock ");
960MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
961MODULE_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 (1ULL << VIRTIO_F_ACCESS_PLATFORM) |
35 (1ULL << VIRTIO_VSOCK_F_SEQPACKET)
36};
37
38enum {
39 VHOST_VSOCK_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)
40};
41
42/* Used to track all the vhost_vsock instances on the system. */
43static DEFINE_MUTEX(vhost_vsock_mutex);
44static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
45
46struct vhost_vsock {
47 struct vhost_dev dev;
48 struct vhost_virtqueue vqs[2];
49
50 /* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */
51 struct hlist_node hash;
52
53 struct vhost_work send_pkt_work;
54 spinlock_t send_pkt_list_lock;
55 struct list_head send_pkt_list; /* host->guest pending packets */
56
57 atomic_t queued_replies;
58
59 u32 guest_cid;
60 bool seqpacket_allow;
61};
62
63static u32 vhost_transport_get_local_cid(void)
64{
65 return VHOST_VSOCK_DEFAULT_HOST_CID;
66}
67
68/* Callers that dereference the return value must hold vhost_vsock_mutex or the
69 * RCU read lock.
70 */
71static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
72{
73 struct vhost_vsock *vsock;
74
75 hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
76 u32 other_cid = vsock->guest_cid;
77
78 /* Skip instances that have no CID yet */
79 if (other_cid == 0)
80 continue;
81
82 if (other_cid == guest_cid)
83 return vsock;
84
85 }
86
87 return NULL;
88}
89
90static void
91vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
92 struct vhost_virtqueue *vq)
93{
94 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
95 int pkts = 0, total_len = 0;
96 bool added = false;
97 bool restart_tx = false;
98
99 mutex_lock(&vq->mutex);
100
101 if (!vhost_vq_get_backend(vq))
102 goto out;
103
104 if (!vq_meta_prefetch(vq))
105 goto out;
106
107 /* Avoid further vmexits, we're already processing the virtqueue */
108 vhost_disable_notify(&vsock->dev, vq);
109
110 do {
111 struct virtio_vsock_pkt *pkt;
112 struct iov_iter iov_iter;
113 unsigned out, in;
114 size_t nbytes;
115 size_t iov_len, payload_len;
116 int head;
117 bool restore_flag = false;
118
119 spin_lock_bh(&vsock->send_pkt_list_lock);
120 if (list_empty(&vsock->send_pkt_list)) {
121 spin_unlock_bh(&vsock->send_pkt_list_lock);
122 vhost_enable_notify(&vsock->dev, vq);
123 break;
124 }
125
126 pkt = list_first_entry(&vsock->send_pkt_list,
127 struct virtio_vsock_pkt, list);
128 list_del_init(&pkt->list);
129 spin_unlock_bh(&vsock->send_pkt_list_lock);
130
131 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
132 &out, &in, NULL, NULL);
133 if (head < 0) {
134 spin_lock_bh(&vsock->send_pkt_list_lock);
135 list_add(&pkt->list, &vsock->send_pkt_list);
136 spin_unlock_bh(&vsock->send_pkt_list_lock);
137 break;
138 }
139
140 if (head == vq->num) {
141 spin_lock_bh(&vsock->send_pkt_list_lock);
142 list_add(&pkt->list, &vsock->send_pkt_list);
143 spin_unlock_bh(&vsock->send_pkt_list_lock);
144
145 /* We cannot finish yet if more buffers snuck in while
146 * re-enabling notify.
147 */
148 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
149 vhost_disable_notify(&vsock->dev, vq);
150 continue;
151 }
152 break;
153 }
154
155 if (out) {
156 virtio_transport_free_pkt(pkt);
157 vq_err(vq, "Expected 0 output buffers, got %u\n", out);
158 break;
159 }
160
161 iov_len = iov_length(&vq->iov[out], in);
162 if (iov_len < sizeof(pkt->hdr)) {
163 virtio_transport_free_pkt(pkt);
164 vq_err(vq, "Buffer len [%zu] too small\n", iov_len);
165 break;
166 }
167
168 iov_iter_init(&iov_iter, READ, &vq->iov[out], in, iov_len);
169 payload_len = pkt->len - pkt->off;
170
171 /* If the packet is greater than the space available in the
172 * buffer, we split it using multiple buffers.
173 */
174 if (payload_len > iov_len - sizeof(pkt->hdr)) {
175 payload_len = iov_len - sizeof(pkt->hdr);
176
177 /* As we are copying pieces of large packet's buffer to
178 * small rx buffers, headers of packets in rx queue are
179 * created dynamically and are initialized with header
180 * of current packet(except length). But in case of
181 * SOCK_SEQPACKET, we also must clear record delimeter
182 * bit(VIRTIO_VSOCK_SEQ_EOR). Otherwise, instead of one
183 * packet with delimeter(which marks end of record),
184 * there will be sequence of packets with delimeter
185 * bit set. After initialized header will be copied to
186 * rx buffer, this bit will be restored.
187 */
188 if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR) {
189 pkt->hdr.flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
190 restore_flag = true;
191 }
192 }
193
194 /* Set the correct length in the header */
195 pkt->hdr.len = cpu_to_le32(payload_len);
196
197 nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
198 if (nbytes != sizeof(pkt->hdr)) {
199 virtio_transport_free_pkt(pkt);
200 vq_err(vq, "Faulted on copying pkt hdr\n");
201 break;
202 }
203
204 nbytes = copy_to_iter(pkt->buf + pkt->off, payload_len,
205 &iov_iter);
206 if (nbytes != payload_len) {
207 virtio_transport_free_pkt(pkt);
208 vq_err(vq, "Faulted on copying pkt buf\n");
209 break;
210 }
211
212 /* Deliver to monitoring devices all packets that we
213 * will transmit.
214 */
215 virtio_transport_deliver_tap_pkt(pkt);
216
217 vhost_add_used(vq, head, sizeof(pkt->hdr) + payload_len);
218 added = true;
219
220 pkt->off += payload_len;
221 total_len += payload_len;
222
223 /* If we didn't send all the payload we can requeue the packet
224 * to send it with the next available buffer.
225 */
226 if (pkt->off < pkt->len) {
227 if (restore_flag)
228 pkt->hdr.flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
229
230 /* We are queueing the same virtio_vsock_pkt to handle
231 * the remaining bytes, and we want to deliver it
232 * to monitoring devices in the next iteration.
233 */
234 pkt->tap_delivered = false;
235
236 spin_lock_bh(&vsock->send_pkt_list_lock);
237 list_add(&pkt->list, &vsock->send_pkt_list);
238 spin_unlock_bh(&vsock->send_pkt_list_lock);
239 } else {
240 if (pkt->reply) {
241 int val;
242
243 val = atomic_dec_return(&vsock->queued_replies);
244
245 /* Do we have resources to resume tx
246 * processing?
247 */
248 if (val + 1 == tx_vq->num)
249 restart_tx = true;
250 }
251
252 virtio_transport_free_pkt(pkt);
253 }
254 } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
255 if (added)
256 vhost_signal(&vsock->dev, vq);
257
258out:
259 mutex_unlock(&vq->mutex);
260
261 if (restart_tx)
262 vhost_poll_queue(&tx_vq->poll);
263}
264
265static void vhost_transport_send_pkt_work(struct vhost_work *work)
266{
267 struct vhost_virtqueue *vq;
268 struct vhost_vsock *vsock;
269
270 vsock = container_of(work, struct vhost_vsock, send_pkt_work);
271 vq = &vsock->vqs[VSOCK_VQ_RX];
272
273 vhost_transport_do_send_pkt(vsock, vq);
274}
275
276static int
277vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
278{
279 struct vhost_vsock *vsock;
280 int len = pkt->len;
281
282 rcu_read_lock();
283
284 /* Find the vhost_vsock according to guest context id */
285 vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
286 if (!vsock) {
287 rcu_read_unlock();
288 virtio_transport_free_pkt(pkt);
289 return -ENODEV;
290 }
291
292 if (pkt->reply)
293 atomic_inc(&vsock->queued_replies);
294
295 spin_lock_bh(&vsock->send_pkt_list_lock);
296 list_add_tail(&pkt->list, &vsock->send_pkt_list);
297 spin_unlock_bh(&vsock->send_pkt_list_lock);
298
299 vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
300
301 rcu_read_unlock();
302 return len;
303}
304
305static int
306vhost_transport_cancel_pkt(struct vsock_sock *vsk)
307{
308 struct vhost_vsock *vsock;
309 struct virtio_vsock_pkt *pkt, *n;
310 int cnt = 0;
311 int ret = -ENODEV;
312 LIST_HEAD(freeme);
313
314 rcu_read_lock();
315
316 /* Find the vhost_vsock according to guest context id */
317 vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
318 if (!vsock)
319 goto out;
320
321 spin_lock_bh(&vsock->send_pkt_list_lock);
322 list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
323 if (pkt->vsk != vsk)
324 continue;
325 list_move(&pkt->list, &freeme);
326 }
327 spin_unlock_bh(&vsock->send_pkt_list_lock);
328
329 list_for_each_entry_safe(pkt, n, &freeme, list) {
330 if (pkt->reply)
331 cnt++;
332 list_del(&pkt->list);
333 virtio_transport_free_pkt(pkt);
334 }
335
336 if (cnt) {
337 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
338 int new_cnt;
339
340 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
341 if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
342 vhost_poll_queue(&tx_vq->poll);
343 }
344
345 ret = 0;
346out:
347 rcu_read_unlock();
348 return ret;
349}
350
351static struct virtio_vsock_pkt *
352vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
353 unsigned int out, unsigned int in)
354{
355 struct virtio_vsock_pkt *pkt;
356 struct iov_iter iov_iter;
357 size_t nbytes;
358 size_t len;
359
360 if (in != 0) {
361 vq_err(vq, "Expected 0 input buffers, got %u\n", in);
362 return NULL;
363 }
364
365 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
366 if (!pkt)
367 return NULL;
368
369 len = iov_length(vq->iov, out);
370 iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
371
372 nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
373 if (nbytes != sizeof(pkt->hdr)) {
374 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
375 sizeof(pkt->hdr), nbytes);
376 kfree(pkt);
377 return NULL;
378 }
379
380 pkt->len = le32_to_cpu(pkt->hdr.len);
381
382 /* No payload */
383 if (!pkt->len)
384 return pkt;
385
386 /* The pkt is too big */
387 if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
388 kfree(pkt);
389 return NULL;
390 }
391
392 pkt->buf = kmalloc(pkt->len, GFP_KERNEL);
393 if (!pkt->buf) {
394 kfree(pkt);
395 return NULL;
396 }
397
398 pkt->buf_len = pkt->len;
399
400 nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
401 if (nbytes != pkt->len) {
402 vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
403 pkt->len, nbytes);
404 virtio_transport_free_pkt(pkt);
405 return NULL;
406 }
407
408 return pkt;
409}
410
411/* Is there space left for replies to rx packets? */
412static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
413{
414 struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
415 int val;
416
417 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
418 val = atomic_read(&vsock->queued_replies);
419
420 return val < vq->num;
421}
422
423static bool vhost_transport_seqpacket_allow(u32 remote_cid);
424
425static struct virtio_transport vhost_transport = {
426 .transport = {
427 .module = THIS_MODULE,
428
429 .get_local_cid = vhost_transport_get_local_cid,
430
431 .init = virtio_transport_do_socket_init,
432 .destruct = virtio_transport_destruct,
433 .release = virtio_transport_release,
434 .connect = virtio_transport_connect,
435 .shutdown = virtio_transport_shutdown,
436 .cancel_pkt = vhost_transport_cancel_pkt,
437
438 .dgram_enqueue = virtio_transport_dgram_enqueue,
439 .dgram_dequeue = virtio_transport_dgram_dequeue,
440 .dgram_bind = virtio_transport_dgram_bind,
441 .dgram_allow = virtio_transport_dgram_allow,
442
443 .stream_enqueue = virtio_transport_stream_enqueue,
444 .stream_dequeue = virtio_transport_stream_dequeue,
445 .stream_has_data = virtio_transport_stream_has_data,
446 .stream_has_space = virtio_transport_stream_has_space,
447 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
448 .stream_is_active = virtio_transport_stream_is_active,
449 .stream_allow = virtio_transport_stream_allow,
450
451 .seqpacket_dequeue = virtio_transport_seqpacket_dequeue,
452 .seqpacket_enqueue = virtio_transport_seqpacket_enqueue,
453 .seqpacket_allow = vhost_transport_seqpacket_allow,
454 .seqpacket_has_data = virtio_transport_seqpacket_has_data,
455
456 .notify_poll_in = virtio_transport_notify_poll_in,
457 .notify_poll_out = virtio_transport_notify_poll_out,
458 .notify_recv_init = virtio_transport_notify_recv_init,
459 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
460 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
461 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
462 .notify_send_init = virtio_transport_notify_send_init,
463 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
464 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
465 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
466 .notify_buffer_size = virtio_transport_notify_buffer_size,
467
468 },
469
470 .send_pkt = vhost_transport_send_pkt,
471};
472
473static bool vhost_transport_seqpacket_allow(u32 remote_cid)
474{
475 struct vhost_vsock *vsock;
476 bool seqpacket_allow = false;
477
478 rcu_read_lock();
479 vsock = vhost_vsock_get(remote_cid);
480
481 if (vsock)
482 seqpacket_allow = vsock->seqpacket_allow;
483
484 rcu_read_unlock();
485
486 return seqpacket_allow;
487}
488
489static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
490{
491 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
492 poll.work);
493 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
494 dev);
495 struct virtio_vsock_pkt *pkt;
496 int head, pkts = 0, total_len = 0;
497 unsigned int out, in;
498 bool added = false;
499
500 mutex_lock(&vq->mutex);
501
502 if (!vhost_vq_get_backend(vq))
503 goto out;
504
505 if (!vq_meta_prefetch(vq))
506 goto out;
507
508 vhost_disable_notify(&vsock->dev, vq);
509 do {
510 u32 len;
511
512 if (!vhost_vsock_more_replies(vsock)) {
513 /* Stop tx until the device processes already
514 * pending replies. Leave tx virtqueue
515 * callbacks disabled.
516 */
517 goto no_more_replies;
518 }
519
520 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
521 &out, &in, NULL, NULL);
522 if (head < 0)
523 break;
524
525 if (head == vq->num) {
526 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
527 vhost_disable_notify(&vsock->dev, vq);
528 continue;
529 }
530 break;
531 }
532
533 pkt = vhost_vsock_alloc_pkt(vq, out, in);
534 if (!pkt) {
535 vq_err(vq, "Faulted on pkt\n");
536 continue;
537 }
538
539 len = pkt->len;
540
541 /* Deliver to monitoring devices all received packets */
542 virtio_transport_deliver_tap_pkt(pkt);
543
544 /* Only accept correctly addressed packets */
545 if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid &&
546 le64_to_cpu(pkt->hdr.dst_cid) ==
547 vhost_transport_get_local_cid())
548 virtio_transport_recv_pkt(&vhost_transport, pkt);
549 else
550 virtio_transport_free_pkt(pkt);
551
552 len += sizeof(pkt->hdr);
553 vhost_add_used(vq, head, len);
554 total_len += len;
555 added = true;
556 } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
557
558no_more_replies:
559 if (added)
560 vhost_signal(&vsock->dev, vq);
561
562out:
563 mutex_unlock(&vq->mutex);
564}
565
566static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
567{
568 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
569 poll.work);
570 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
571 dev);
572
573 vhost_transport_do_send_pkt(vsock, vq);
574}
575
576static int vhost_vsock_start(struct vhost_vsock *vsock)
577{
578 struct vhost_virtqueue *vq;
579 size_t i;
580 int ret;
581
582 mutex_lock(&vsock->dev.mutex);
583
584 ret = vhost_dev_check_owner(&vsock->dev);
585 if (ret)
586 goto err;
587
588 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
589 vq = &vsock->vqs[i];
590
591 mutex_lock(&vq->mutex);
592
593 if (!vhost_vq_access_ok(vq)) {
594 ret = -EFAULT;
595 goto err_vq;
596 }
597
598 if (!vhost_vq_get_backend(vq)) {
599 vhost_vq_set_backend(vq, vsock);
600 ret = vhost_vq_init_access(vq);
601 if (ret)
602 goto err_vq;
603 }
604
605 mutex_unlock(&vq->mutex);
606 }
607
608 /* Some packets may have been queued before the device was started,
609 * let's kick the send worker to send them.
610 */
611 vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
612
613 mutex_unlock(&vsock->dev.mutex);
614 return 0;
615
616err_vq:
617 vhost_vq_set_backend(vq, NULL);
618 mutex_unlock(&vq->mutex);
619
620 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
621 vq = &vsock->vqs[i];
622
623 mutex_lock(&vq->mutex);
624 vhost_vq_set_backend(vq, NULL);
625 mutex_unlock(&vq->mutex);
626 }
627err:
628 mutex_unlock(&vsock->dev.mutex);
629 return ret;
630}
631
632static int vhost_vsock_stop(struct vhost_vsock *vsock)
633{
634 size_t i;
635 int ret;
636
637 mutex_lock(&vsock->dev.mutex);
638
639 ret = vhost_dev_check_owner(&vsock->dev);
640 if (ret)
641 goto err;
642
643 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
644 struct vhost_virtqueue *vq = &vsock->vqs[i];
645
646 mutex_lock(&vq->mutex);
647 vhost_vq_set_backend(vq, NULL);
648 mutex_unlock(&vq->mutex);
649 }
650
651err:
652 mutex_unlock(&vsock->dev.mutex);
653 return ret;
654}
655
656static void vhost_vsock_free(struct vhost_vsock *vsock)
657{
658 kvfree(vsock);
659}
660
661static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
662{
663 struct vhost_virtqueue **vqs;
664 struct vhost_vsock *vsock;
665 int ret;
666
667 /* This struct is large and allocation could fail, fall back to vmalloc
668 * if there is no other way.
669 */
670 vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
671 if (!vsock)
672 return -ENOMEM;
673
674 vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
675 if (!vqs) {
676 ret = -ENOMEM;
677 goto out;
678 }
679
680 vsock->guest_cid = 0; /* no CID assigned yet */
681
682 atomic_set(&vsock->queued_replies, 0);
683
684 vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
685 vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
686 vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
687 vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
688
689 vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
690 UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
691 VHOST_VSOCK_WEIGHT, true, NULL);
692
693 file->private_data = vsock;
694 spin_lock_init(&vsock->send_pkt_list_lock);
695 INIT_LIST_HEAD(&vsock->send_pkt_list);
696 vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
697 return 0;
698
699out:
700 vhost_vsock_free(vsock);
701 return ret;
702}
703
704static void vhost_vsock_flush(struct vhost_vsock *vsock)
705{
706 int i;
707
708 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
709 if (vsock->vqs[i].handle_kick)
710 vhost_poll_flush(&vsock->vqs[i].poll);
711 vhost_work_dev_flush(&vsock->dev);
712}
713
714static void vhost_vsock_reset_orphans(struct sock *sk)
715{
716 struct vsock_sock *vsk = vsock_sk(sk);
717
718 /* vmci_transport.c doesn't take sk_lock here either. At least we're
719 * under vsock_table_lock so the sock cannot disappear while we're
720 * executing.
721 */
722
723 /* If the peer is still valid, no need to reset connection */
724 if (vhost_vsock_get(vsk->remote_addr.svm_cid))
725 return;
726
727 /* If the close timeout is pending, let it expire. This avoids races
728 * with the timeout callback.
729 */
730 if (vsk->close_work_scheduled)
731 return;
732
733 sock_set_flag(sk, SOCK_DONE);
734 vsk->peer_shutdown = SHUTDOWN_MASK;
735 sk->sk_state = SS_UNCONNECTED;
736 sk->sk_err = ECONNRESET;
737 sk_error_report(sk);
738}
739
740static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
741{
742 struct vhost_vsock *vsock = file->private_data;
743
744 mutex_lock(&vhost_vsock_mutex);
745 if (vsock->guest_cid)
746 hash_del_rcu(&vsock->hash);
747 mutex_unlock(&vhost_vsock_mutex);
748
749 /* Wait for other CPUs to finish using vsock */
750 synchronize_rcu();
751
752 /* Iterating over all connections for all CIDs to find orphans is
753 * inefficient. Room for improvement here. */
754 vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
755
756 vhost_vsock_stop(vsock);
757 vhost_vsock_flush(vsock);
758 vhost_dev_stop(&vsock->dev);
759
760 spin_lock_bh(&vsock->send_pkt_list_lock);
761 while (!list_empty(&vsock->send_pkt_list)) {
762 struct virtio_vsock_pkt *pkt;
763
764 pkt = list_first_entry(&vsock->send_pkt_list,
765 struct virtio_vsock_pkt, list);
766 list_del_init(&pkt->list);
767 virtio_transport_free_pkt(pkt);
768 }
769 spin_unlock_bh(&vsock->send_pkt_list_lock);
770
771 vhost_dev_cleanup(&vsock->dev);
772 kfree(vsock->dev.vqs);
773 vhost_vsock_free(vsock);
774 return 0;
775}
776
777static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
778{
779 struct vhost_vsock *other;
780
781 /* Refuse reserved CIDs */
782 if (guest_cid <= VMADDR_CID_HOST ||
783 guest_cid == U32_MAX)
784 return -EINVAL;
785
786 /* 64-bit CIDs are not yet supported */
787 if (guest_cid > U32_MAX)
788 return -EINVAL;
789
790 /* Refuse if CID is assigned to the guest->host transport (i.e. nested
791 * VM), to make the loopback work.
792 */
793 if (vsock_find_cid(guest_cid))
794 return -EADDRINUSE;
795
796 /* Refuse if CID is already in use */
797 mutex_lock(&vhost_vsock_mutex);
798 other = vhost_vsock_get(guest_cid);
799 if (other && other != vsock) {
800 mutex_unlock(&vhost_vsock_mutex);
801 return -EADDRINUSE;
802 }
803
804 if (vsock->guest_cid)
805 hash_del_rcu(&vsock->hash);
806
807 vsock->guest_cid = guest_cid;
808 hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
809 mutex_unlock(&vhost_vsock_mutex);
810
811 return 0;
812}
813
814static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
815{
816 struct vhost_virtqueue *vq;
817 int i;
818
819 if (features & ~VHOST_VSOCK_FEATURES)
820 return -EOPNOTSUPP;
821
822 mutex_lock(&vsock->dev.mutex);
823 if ((features & (1 << VHOST_F_LOG_ALL)) &&
824 !vhost_log_access_ok(&vsock->dev)) {
825 goto err;
826 }
827
828 if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) {
829 if (vhost_init_device_iotlb(&vsock->dev, true))
830 goto err;
831 }
832
833 if (features & (1ULL << VIRTIO_VSOCK_F_SEQPACKET))
834 vsock->seqpacket_allow = true;
835
836 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
837 vq = &vsock->vqs[i];
838 mutex_lock(&vq->mutex);
839 vq->acked_features = features;
840 mutex_unlock(&vq->mutex);
841 }
842 mutex_unlock(&vsock->dev.mutex);
843 return 0;
844
845err:
846 mutex_unlock(&vsock->dev.mutex);
847 return -EFAULT;
848}
849
850static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
851 unsigned long arg)
852{
853 struct vhost_vsock *vsock = f->private_data;
854 void __user *argp = (void __user *)arg;
855 u64 guest_cid;
856 u64 features;
857 int start;
858 int r;
859
860 switch (ioctl) {
861 case VHOST_VSOCK_SET_GUEST_CID:
862 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
863 return -EFAULT;
864 return vhost_vsock_set_cid(vsock, guest_cid);
865 case VHOST_VSOCK_SET_RUNNING:
866 if (copy_from_user(&start, argp, sizeof(start)))
867 return -EFAULT;
868 if (start)
869 return vhost_vsock_start(vsock);
870 else
871 return vhost_vsock_stop(vsock);
872 case VHOST_GET_FEATURES:
873 features = VHOST_VSOCK_FEATURES;
874 if (copy_to_user(argp, &features, sizeof(features)))
875 return -EFAULT;
876 return 0;
877 case VHOST_SET_FEATURES:
878 if (copy_from_user(&features, argp, sizeof(features)))
879 return -EFAULT;
880 return vhost_vsock_set_features(vsock, features);
881 case VHOST_GET_BACKEND_FEATURES:
882 features = VHOST_VSOCK_BACKEND_FEATURES;
883 if (copy_to_user(argp, &features, sizeof(features)))
884 return -EFAULT;
885 return 0;
886 case VHOST_SET_BACKEND_FEATURES:
887 if (copy_from_user(&features, argp, sizeof(features)))
888 return -EFAULT;
889 if (features & ~VHOST_VSOCK_BACKEND_FEATURES)
890 return -EOPNOTSUPP;
891 vhost_set_backend_features(&vsock->dev, features);
892 return 0;
893 default:
894 mutex_lock(&vsock->dev.mutex);
895 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
896 if (r == -ENOIOCTLCMD)
897 r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
898 else
899 vhost_vsock_flush(vsock);
900 mutex_unlock(&vsock->dev.mutex);
901 return r;
902 }
903}
904
905static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
906{
907 struct file *file = iocb->ki_filp;
908 struct vhost_vsock *vsock = file->private_data;
909 struct vhost_dev *dev = &vsock->dev;
910 int noblock = file->f_flags & O_NONBLOCK;
911
912 return vhost_chr_read_iter(dev, to, noblock);
913}
914
915static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb,
916 struct iov_iter *from)
917{
918 struct file *file = iocb->ki_filp;
919 struct vhost_vsock *vsock = file->private_data;
920 struct vhost_dev *dev = &vsock->dev;
921
922 return vhost_chr_write_iter(dev, from);
923}
924
925static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait)
926{
927 struct vhost_vsock *vsock = file->private_data;
928 struct vhost_dev *dev = &vsock->dev;
929
930 return vhost_chr_poll(file, dev, wait);
931}
932
933static const struct file_operations vhost_vsock_fops = {
934 .owner = THIS_MODULE,
935 .open = vhost_vsock_dev_open,
936 .release = vhost_vsock_dev_release,
937 .llseek = noop_llseek,
938 .unlocked_ioctl = vhost_vsock_dev_ioctl,
939 .compat_ioctl = compat_ptr_ioctl,
940 .read_iter = vhost_vsock_chr_read_iter,
941 .write_iter = vhost_vsock_chr_write_iter,
942 .poll = vhost_vsock_chr_poll,
943};
944
945static struct miscdevice vhost_vsock_misc = {
946 .minor = VHOST_VSOCK_MINOR,
947 .name = "vhost-vsock",
948 .fops = &vhost_vsock_fops,
949};
950
951static int __init vhost_vsock_init(void)
952{
953 int ret;
954
955 ret = vsock_core_register(&vhost_transport.transport,
956 VSOCK_TRANSPORT_F_H2G);
957 if (ret < 0)
958 return ret;
959 return misc_register(&vhost_vsock_misc);
960};
961
962static void __exit vhost_vsock_exit(void)
963{
964 misc_deregister(&vhost_vsock_misc);
965 vsock_core_unregister(&vhost_transport.transport);
966};
967
968module_init(vhost_vsock_init);
969module_exit(vhost_vsock_exit);
970MODULE_LICENSE("GPL v2");
971MODULE_AUTHOR("Asias He");
972MODULE_DESCRIPTION("vhost transport for vsock ");
973MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
974MODULE_ALIAS("devname:vhost-vsock");