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
32 enum {
33 VHOST_VSOCK_FEATURES = VHOST_FEATURES |
34 (1ULL << VIRTIO_F_ACCESS_PLATFORM) |
35 (1ULL << VIRTIO_VSOCK_F_SEQPACKET)
36 };
37
38 enum {
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. */
43 static DEFINE_MUTEX(vhost_vsock_mutex);
44 static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
45
46 struct 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
vhost_transport_get_local_cid(void)62 static 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 */
vhost_vsock_get(u32 guest_cid)70 static 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
89 static void
vhost_transport_do_send_pkt(struct vhost_vsock * vsock,struct vhost_virtqueue * vq)90 vhost_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 virtio_transport_consume_skb_sent(skb, true);
248 }
249 } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
250 if (added)
251 vhost_signal(&vsock->dev, vq);
252
253 out:
254 mutex_unlock(&vq->mutex);
255
256 if (restart_tx)
257 vhost_poll_queue(&tx_vq->poll);
258 }
259
vhost_transport_send_pkt_work(struct vhost_work * work)260 static 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
271 static int
vhost_transport_send_pkt(struct sk_buff * skb)272 vhost_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
298 static int
vhost_transport_cancel_pkt(struct vsock_sock * vsk)299 vhost_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;
324 out:
325 rcu_read_unlock();
326 return ret;
327 }
328
329 static struct sk_buff *
vhost_vsock_alloc_skb(struct vhost_virtqueue * vq,unsigned int out,unsigned int in)330 vhost_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 if (len < VIRTIO_VSOCK_SKB_HEADROOM ||
348 len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE + VIRTIO_VSOCK_SKB_HEADROOM)
349 return NULL;
350
351 /* len contains both payload and hdr */
352 skb = virtio_vsock_alloc_skb(len, GFP_KERNEL);
353 if (!skb)
354 return NULL;
355
356 iov_iter_init(&iov_iter, ITER_SOURCE, vq->iov, out, len);
357
358 hdr = virtio_vsock_hdr(skb);
359 nbytes = copy_from_iter(hdr, sizeof(*hdr), &iov_iter);
360 if (nbytes != sizeof(*hdr)) {
361 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
362 sizeof(*hdr), nbytes);
363 kfree_skb(skb);
364 return NULL;
365 }
366
367 payload_len = le32_to_cpu(hdr->len);
368
369 /* No payload */
370 if (!payload_len)
371 return skb;
372
373 /* The pkt is too big or the length in the header is invalid */
374 if (payload_len + sizeof(*hdr) > len) {
375 kfree_skb(skb);
376 return NULL;
377 }
378
379 virtio_vsock_skb_put(skb, payload_len);
380
381 if (skb_copy_datagram_from_iter(skb, 0, &iov_iter, payload_len)) {
382 vq_err(vq, "Failed to copy %zu byte payload\n", payload_len);
383 kfree_skb(skb);
384 return NULL;
385 }
386
387 return skb;
388 }
389
390 /* Is there space left for replies to rx packets? */
vhost_vsock_more_replies(struct vhost_vsock * vsock)391 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
392 {
393 struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
394 int val;
395
396 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
397 val = atomic_read(&vsock->queued_replies);
398
399 return val < vq->num;
400 }
401
vhost_transport_msgzerocopy_allow(void)402 static bool vhost_transport_msgzerocopy_allow(void)
403 {
404 return true;
405 }
406
407 static bool vhost_transport_seqpacket_allow(u32 remote_cid);
408
409 static struct virtio_transport vhost_transport = {
410 .transport = {
411 .module = THIS_MODULE,
412
413 .get_local_cid = vhost_transport_get_local_cid,
414
415 .init = virtio_transport_do_socket_init,
416 .destruct = virtio_transport_destruct,
417 .release = virtio_transport_release,
418 .connect = virtio_transport_connect,
419 .shutdown = virtio_transport_shutdown,
420 .cancel_pkt = vhost_transport_cancel_pkt,
421
422 .dgram_enqueue = virtio_transport_dgram_enqueue,
423 .dgram_dequeue = virtio_transport_dgram_dequeue,
424 .dgram_bind = virtio_transport_dgram_bind,
425 .dgram_allow = virtio_transport_dgram_allow,
426
427 .stream_enqueue = virtio_transport_stream_enqueue,
428 .stream_dequeue = virtio_transport_stream_dequeue,
429 .stream_has_data = virtio_transport_stream_has_data,
430 .stream_has_space = virtio_transport_stream_has_space,
431 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
432 .stream_is_active = virtio_transport_stream_is_active,
433 .stream_allow = virtio_transport_stream_allow,
434
435 .seqpacket_dequeue = virtio_transport_seqpacket_dequeue,
436 .seqpacket_enqueue = virtio_transport_seqpacket_enqueue,
437 .seqpacket_allow = vhost_transport_seqpacket_allow,
438 .seqpacket_has_data = virtio_transport_seqpacket_has_data,
439
440 .msgzerocopy_allow = vhost_transport_msgzerocopy_allow,
441
442 .notify_poll_in = virtio_transport_notify_poll_in,
443 .notify_poll_out = virtio_transport_notify_poll_out,
444 .notify_recv_init = virtio_transport_notify_recv_init,
445 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
446 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
447 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
448 .notify_send_init = virtio_transport_notify_send_init,
449 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
450 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
451 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
452 .notify_buffer_size = virtio_transport_notify_buffer_size,
453 .notify_set_rcvlowat = virtio_transport_notify_set_rcvlowat,
454
455 .unsent_bytes = virtio_transport_unsent_bytes,
456
457 .read_skb = virtio_transport_read_skb,
458 },
459
460 .send_pkt = vhost_transport_send_pkt,
461 };
462
vhost_transport_seqpacket_allow(u32 remote_cid)463 static bool vhost_transport_seqpacket_allow(u32 remote_cid)
464 {
465 struct vhost_vsock *vsock;
466 bool seqpacket_allow = false;
467
468 rcu_read_lock();
469 vsock = vhost_vsock_get(remote_cid);
470
471 if (vsock)
472 seqpacket_allow = vsock->seqpacket_allow;
473
474 rcu_read_unlock();
475
476 return seqpacket_allow;
477 }
478
vhost_vsock_handle_tx_kick(struct vhost_work * work)479 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
480 {
481 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
482 poll.work);
483 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
484 dev);
485 int head, pkts = 0, total_len = 0;
486 unsigned int out, in;
487 struct sk_buff *skb;
488 bool added = false;
489
490 mutex_lock(&vq->mutex);
491
492 if (!vhost_vq_get_backend(vq))
493 goto out;
494
495 if (!vq_meta_prefetch(vq))
496 goto out;
497
498 vhost_disable_notify(&vsock->dev, vq);
499 do {
500 struct virtio_vsock_hdr *hdr;
501
502 if (!vhost_vsock_more_replies(vsock)) {
503 /* Stop tx until the device processes already
504 * pending replies. Leave tx virtqueue
505 * callbacks disabled.
506 */
507 goto no_more_replies;
508 }
509
510 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
511 &out, &in, NULL, NULL);
512 if (head < 0)
513 break;
514
515 if (head == vq->num) {
516 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
517 vhost_disable_notify(&vsock->dev, vq);
518 continue;
519 }
520 break;
521 }
522
523 skb = vhost_vsock_alloc_skb(vq, out, in);
524 if (!skb) {
525 vq_err(vq, "Faulted on pkt\n");
526 continue;
527 }
528
529 total_len += sizeof(*hdr) + skb->len;
530
531 /* Deliver to monitoring devices all received packets */
532 virtio_transport_deliver_tap_pkt(skb);
533
534 hdr = virtio_vsock_hdr(skb);
535
536 /* Only accept correctly addressed packets */
537 if (le64_to_cpu(hdr->src_cid) == vsock->guest_cid &&
538 le64_to_cpu(hdr->dst_cid) ==
539 vhost_transport_get_local_cid())
540 virtio_transport_recv_pkt(&vhost_transport, skb);
541 else
542 kfree_skb(skb);
543
544 vhost_add_used(vq, head, 0);
545 added = true;
546 } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
547
548 no_more_replies:
549 if (added)
550 vhost_signal(&vsock->dev, vq);
551
552 out:
553 mutex_unlock(&vq->mutex);
554 }
555
vhost_vsock_handle_rx_kick(struct vhost_work * work)556 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
557 {
558 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
559 poll.work);
560 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
561 dev);
562
563 vhost_transport_do_send_pkt(vsock, vq);
564 }
565
vhost_vsock_start(struct vhost_vsock * vsock)566 static int vhost_vsock_start(struct vhost_vsock *vsock)
567 {
568 struct vhost_virtqueue *vq;
569 size_t i;
570 int ret;
571
572 mutex_lock(&vsock->dev.mutex);
573
574 ret = vhost_dev_check_owner(&vsock->dev);
575 if (ret)
576 goto err;
577
578 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
579 vq = &vsock->vqs[i];
580
581 mutex_lock(&vq->mutex);
582
583 if (!vhost_vq_access_ok(vq)) {
584 ret = -EFAULT;
585 goto err_vq;
586 }
587
588 if (!vhost_vq_get_backend(vq)) {
589 vhost_vq_set_backend(vq, vsock);
590 ret = vhost_vq_init_access(vq);
591 if (ret)
592 goto err_vq;
593 }
594
595 mutex_unlock(&vq->mutex);
596 }
597
598 /* Some packets may have been queued before the device was started,
599 * let's kick the send worker to send them.
600 */
601 vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &vsock->send_pkt_work);
602
603 mutex_unlock(&vsock->dev.mutex);
604 return 0;
605
606 err_vq:
607 vhost_vq_set_backend(vq, NULL);
608 mutex_unlock(&vq->mutex);
609
610 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
611 vq = &vsock->vqs[i];
612
613 mutex_lock(&vq->mutex);
614 vhost_vq_set_backend(vq, NULL);
615 mutex_unlock(&vq->mutex);
616 }
617 err:
618 mutex_unlock(&vsock->dev.mutex);
619 return ret;
620 }
621
vhost_vsock_stop(struct vhost_vsock * vsock,bool check_owner)622 static int vhost_vsock_stop(struct vhost_vsock *vsock, bool check_owner)
623 {
624 size_t i;
625 int ret = 0;
626
627 mutex_lock(&vsock->dev.mutex);
628
629 if (check_owner) {
630 ret = vhost_dev_check_owner(&vsock->dev);
631 if (ret)
632 goto err;
633 }
634
635 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
636 struct vhost_virtqueue *vq = &vsock->vqs[i];
637
638 mutex_lock(&vq->mutex);
639 vhost_vq_set_backend(vq, NULL);
640 mutex_unlock(&vq->mutex);
641 }
642
643 err:
644 mutex_unlock(&vsock->dev.mutex);
645 return ret;
646 }
647
vhost_vsock_free(struct vhost_vsock * vsock)648 static void vhost_vsock_free(struct vhost_vsock *vsock)
649 {
650 kvfree(vsock);
651 }
652
vhost_vsock_dev_open(struct inode * inode,struct file * file)653 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
654 {
655 struct vhost_virtqueue **vqs;
656 struct vhost_vsock *vsock;
657 int ret;
658
659 /* This struct is large and allocation could fail, fall back to vmalloc
660 * if there is no other way.
661 */
662 vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
663 if (!vsock)
664 return -ENOMEM;
665
666 vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
667 if (!vqs) {
668 ret = -ENOMEM;
669 goto out;
670 }
671
672 vsock->guest_cid = 0; /* no CID assigned yet */
673 vsock->seqpacket_allow = false;
674
675 atomic_set(&vsock->queued_replies, 0);
676
677 vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
678 vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
679 vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
680 vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
681
682 vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
683 UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
684 VHOST_VSOCK_WEIGHT, true, NULL);
685
686 file->private_data = vsock;
687 skb_queue_head_init(&vsock->send_pkt_queue);
688 vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
689 return 0;
690
691 out:
692 vhost_vsock_free(vsock);
693 return ret;
694 }
695
vhost_vsock_flush(struct vhost_vsock * vsock)696 static void vhost_vsock_flush(struct vhost_vsock *vsock)
697 {
698 vhost_dev_flush(&vsock->dev);
699 }
700
vhost_vsock_reset_orphans(struct sock * sk)701 static void vhost_vsock_reset_orphans(struct sock *sk)
702 {
703 struct vsock_sock *vsk = vsock_sk(sk);
704
705 /* vmci_transport.c doesn't take sk_lock here either. At least we're
706 * under vsock_table_lock so the sock cannot disappear while we're
707 * executing.
708 */
709
710 /* If the peer is still valid, no need to reset connection */
711 if (vhost_vsock_get(vsk->remote_addr.svm_cid))
712 return;
713
714 /* If the close timeout is pending, let it expire. This avoids races
715 * with the timeout callback.
716 */
717 if (vsk->close_work_scheduled)
718 return;
719
720 sock_set_flag(sk, SOCK_DONE);
721 vsk->peer_shutdown = SHUTDOWN_MASK;
722 sk->sk_state = SS_UNCONNECTED;
723 sk->sk_err = ECONNRESET;
724 sk_error_report(sk);
725 }
726
vhost_vsock_dev_release(struct inode * inode,struct file * file)727 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
728 {
729 struct vhost_vsock *vsock = file->private_data;
730
731 mutex_lock(&vhost_vsock_mutex);
732 if (vsock->guest_cid)
733 hash_del_rcu(&vsock->hash);
734 mutex_unlock(&vhost_vsock_mutex);
735
736 /* Wait for other CPUs to finish using vsock */
737 synchronize_rcu();
738
739 /* Iterating over all connections for all CIDs to find orphans is
740 * inefficient. Room for improvement here. */
741 vsock_for_each_connected_socket(&vhost_transport.transport,
742 vhost_vsock_reset_orphans);
743
744 /* Don't check the owner, because we are in the release path, so we
745 * need to stop the vsock device in any case.
746 * vhost_vsock_stop() can not fail in this case, so we don't need to
747 * check the return code.
748 */
749 vhost_vsock_stop(vsock, false);
750 vhost_vsock_flush(vsock);
751 vhost_dev_stop(&vsock->dev);
752
753 virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue);
754
755 vhost_dev_cleanup(&vsock->dev);
756 kfree(vsock->dev.vqs);
757 vhost_vsock_free(vsock);
758 return 0;
759 }
760
vhost_vsock_set_cid(struct vhost_vsock * vsock,u64 guest_cid)761 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
762 {
763 struct vhost_vsock *other;
764
765 /* Refuse reserved CIDs */
766 if (guest_cid <= VMADDR_CID_HOST ||
767 guest_cid == U32_MAX)
768 return -EINVAL;
769
770 /* 64-bit CIDs are not yet supported */
771 if (guest_cid > U32_MAX)
772 return -EINVAL;
773
774 /* Refuse if CID is assigned to the guest->host transport (i.e. nested
775 * VM), to make the loopback work.
776 */
777 if (vsock_find_cid(guest_cid))
778 return -EADDRINUSE;
779
780 /* Refuse if CID is already in use */
781 mutex_lock(&vhost_vsock_mutex);
782 other = vhost_vsock_get(guest_cid);
783 if (other && other != vsock) {
784 mutex_unlock(&vhost_vsock_mutex);
785 return -EADDRINUSE;
786 }
787
788 if (vsock->guest_cid)
789 hash_del_rcu(&vsock->hash);
790
791 vsock->guest_cid = guest_cid;
792 hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
793 mutex_unlock(&vhost_vsock_mutex);
794
795 return 0;
796 }
797
vhost_vsock_set_features(struct vhost_vsock * vsock,u64 features)798 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
799 {
800 struct vhost_virtqueue *vq;
801 int i;
802
803 if (features & ~VHOST_VSOCK_FEATURES)
804 return -EOPNOTSUPP;
805
806 mutex_lock(&vsock->dev.mutex);
807 if ((features & (1 << VHOST_F_LOG_ALL)) &&
808 !vhost_log_access_ok(&vsock->dev)) {
809 goto err;
810 }
811
812 if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) {
813 if (vhost_init_device_iotlb(&vsock->dev))
814 goto err;
815 }
816
817 vsock->seqpacket_allow = features & (1ULL << VIRTIO_VSOCK_F_SEQPACKET);
818
819 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
820 vq = &vsock->vqs[i];
821 mutex_lock(&vq->mutex);
822 vq->acked_features = features;
823 mutex_unlock(&vq->mutex);
824 }
825 mutex_unlock(&vsock->dev.mutex);
826 return 0;
827
828 err:
829 mutex_unlock(&vsock->dev.mutex);
830 return -EFAULT;
831 }
832
vhost_vsock_dev_ioctl(struct file * f,unsigned int ioctl,unsigned long arg)833 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
834 unsigned long arg)
835 {
836 struct vhost_vsock *vsock = f->private_data;
837 void __user *argp = (void __user *)arg;
838 u64 guest_cid;
839 u64 features;
840 int start;
841 int r;
842
843 switch (ioctl) {
844 case VHOST_VSOCK_SET_GUEST_CID:
845 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
846 return -EFAULT;
847 return vhost_vsock_set_cid(vsock, guest_cid);
848 case VHOST_VSOCK_SET_RUNNING:
849 if (copy_from_user(&start, argp, sizeof(start)))
850 return -EFAULT;
851 if (start)
852 return vhost_vsock_start(vsock);
853 else
854 return vhost_vsock_stop(vsock, true);
855 case VHOST_GET_FEATURES:
856 features = VHOST_VSOCK_FEATURES;
857 if (copy_to_user(argp, &features, sizeof(features)))
858 return -EFAULT;
859 return 0;
860 case VHOST_SET_FEATURES:
861 if (copy_from_user(&features, argp, sizeof(features)))
862 return -EFAULT;
863 return vhost_vsock_set_features(vsock, features);
864 case VHOST_GET_BACKEND_FEATURES:
865 features = VHOST_VSOCK_BACKEND_FEATURES;
866 if (copy_to_user(argp, &features, sizeof(features)))
867 return -EFAULT;
868 return 0;
869 case VHOST_SET_BACKEND_FEATURES:
870 if (copy_from_user(&features, argp, sizeof(features)))
871 return -EFAULT;
872 if (features & ~VHOST_VSOCK_BACKEND_FEATURES)
873 return -EOPNOTSUPP;
874 vhost_set_backend_features(&vsock->dev, features);
875 return 0;
876 default:
877 mutex_lock(&vsock->dev.mutex);
878 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
879 if (r == -ENOIOCTLCMD)
880 r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
881 else
882 vhost_vsock_flush(vsock);
883 mutex_unlock(&vsock->dev.mutex);
884 return r;
885 }
886 }
887
vhost_vsock_chr_read_iter(struct kiocb * iocb,struct iov_iter * to)888 static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
889 {
890 struct file *file = iocb->ki_filp;
891 struct vhost_vsock *vsock = file->private_data;
892 struct vhost_dev *dev = &vsock->dev;
893 int noblock = file->f_flags & O_NONBLOCK;
894
895 return vhost_chr_read_iter(dev, to, noblock);
896 }
897
vhost_vsock_chr_write_iter(struct kiocb * iocb,struct iov_iter * from)898 static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb,
899 struct iov_iter *from)
900 {
901 struct file *file = iocb->ki_filp;
902 struct vhost_vsock *vsock = file->private_data;
903 struct vhost_dev *dev = &vsock->dev;
904
905 return vhost_chr_write_iter(dev, from);
906 }
907
vhost_vsock_chr_poll(struct file * file,poll_table * wait)908 static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait)
909 {
910 struct vhost_vsock *vsock = file->private_data;
911 struct vhost_dev *dev = &vsock->dev;
912
913 return vhost_chr_poll(file, dev, wait);
914 }
915
916 static const struct file_operations vhost_vsock_fops = {
917 .owner = THIS_MODULE,
918 .open = vhost_vsock_dev_open,
919 .release = vhost_vsock_dev_release,
920 .llseek = noop_llseek,
921 .unlocked_ioctl = vhost_vsock_dev_ioctl,
922 .compat_ioctl = compat_ptr_ioctl,
923 .read_iter = vhost_vsock_chr_read_iter,
924 .write_iter = vhost_vsock_chr_write_iter,
925 .poll = vhost_vsock_chr_poll,
926 };
927
928 static struct miscdevice vhost_vsock_misc = {
929 .minor = VHOST_VSOCK_MINOR,
930 .name = "vhost-vsock",
931 .fops = &vhost_vsock_fops,
932 };
933
vhost_vsock_init(void)934 static int __init vhost_vsock_init(void)
935 {
936 int ret;
937
938 ret = vsock_core_register(&vhost_transport.transport,
939 VSOCK_TRANSPORT_F_H2G);
940 if (ret < 0)
941 return ret;
942
943 ret = misc_register(&vhost_vsock_misc);
944 if (ret) {
945 vsock_core_unregister(&vhost_transport.transport);
946 return ret;
947 }
948
949 return 0;
950 };
951
vhost_vsock_exit(void)952 static void __exit vhost_vsock_exit(void)
953 {
954 misc_deregister(&vhost_vsock_misc);
955 vsock_core_unregister(&vhost_transport.transport);
956 };
957
958 module_init(vhost_vsock_init);
959 module_exit(vhost_vsock_exit);
960 MODULE_LICENSE("GPL v2");
961 MODULE_AUTHOR("Asias He");
962 MODULE_DESCRIPTION("vhost transport for vsock ");
963 MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
964 MODULE_ALIAS("devname:vhost-vsock");
965