sh_eth: fix EESIPR values for SH77{34|63}
[linux/fpc-iii.git] / drivers / vhost / vsock.c
blobbbbf588540ed71d82ed63deb355b77736a2a9628
1 /*
2 * vhost transport for vsock
4 * Copyright (C) 2013-2015 Red Hat, Inc.
5 * Author: Asias He <asias@redhat.com>
6 * Stefan Hajnoczi <stefanha@redhat.com>
8 * This work is licensed under the terms of the GNU GPL, version 2.
9 */
10 #include <linux/miscdevice.h>
11 #include <linux/atomic.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/vmalloc.h>
15 #include <net/sock.h>
16 #include <linux/virtio_vsock.h>
17 #include <linux/vhost.h>
19 #include <net/af_vsock.h>
20 #include "vhost.h"
22 #define VHOST_VSOCK_DEFAULT_HOST_CID 2
24 enum {
25 VHOST_VSOCK_FEATURES = VHOST_FEATURES,
28 /* Used to track all the vhost_vsock instances on the system. */
29 static DEFINE_SPINLOCK(vhost_vsock_lock);
30 static LIST_HEAD(vhost_vsock_list);
32 struct vhost_vsock {
33 struct vhost_dev dev;
34 struct vhost_virtqueue vqs[2];
36 /* Link to global vhost_vsock_list, protected by vhost_vsock_lock */
37 struct list_head list;
39 struct vhost_work send_pkt_work;
40 spinlock_t send_pkt_list_lock;
41 struct list_head send_pkt_list; /* host->guest pending packets */
43 atomic_t queued_replies;
45 u32 guest_cid;
48 static u32 vhost_transport_get_local_cid(void)
50 return VHOST_VSOCK_DEFAULT_HOST_CID;
53 static struct vhost_vsock *__vhost_vsock_get(u32 guest_cid)
55 struct vhost_vsock *vsock;
57 list_for_each_entry(vsock, &vhost_vsock_list, list) {
58 u32 other_cid = vsock->guest_cid;
60 /* Skip instances that have no CID yet */
61 if (other_cid == 0)
62 continue;
64 if (other_cid == guest_cid) {
65 return vsock;
69 return NULL;
72 static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
74 struct vhost_vsock *vsock;
76 spin_lock_bh(&vhost_vsock_lock);
77 vsock = __vhost_vsock_get(guest_cid);
78 spin_unlock_bh(&vhost_vsock_lock);
80 return vsock;
83 static void
84 vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
85 struct vhost_virtqueue *vq)
87 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
88 bool added = false;
89 bool restart_tx = false;
91 mutex_lock(&vq->mutex);
93 if (!vq->private_data)
94 goto out;
96 /* Avoid further vmexits, we're already processing the virtqueue */
97 vhost_disable_notify(&vsock->dev, vq);
99 for (;;) {
100 struct virtio_vsock_pkt *pkt;
101 struct iov_iter iov_iter;
102 unsigned out, in;
103 size_t nbytes;
104 size_t len;
105 int head;
107 spin_lock_bh(&vsock->send_pkt_list_lock);
108 if (list_empty(&vsock->send_pkt_list)) {
109 spin_unlock_bh(&vsock->send_pkt_list_lock);
110 vhost_enable_notify(&vsock->dev, vq);
111 break;
114 pkt = list_first_entry(&vsock->send_pkt_list,
115 struct virtio_vsock_pkt, list);
116 list_del_init(&pkt->list);
117 spin_unlock_bh(&vsock->send_pkt_list_lock);
119 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
120 &out, &in, NULL, NULL);
121 if (head < 0) {
122 spin_lock_bh(&vsock->send_pkt_list_lock);
123 list_add(&pkt->list, &vsock->send_pkt_list);
124 spin_unlock_bh(&vsock->send_pkt_list_lock);
125 break;
128 if (head == vq->num) {
129 spin_lock_bh(&vsock->send_pkt_list_lock);
130 list_add(&pkt->list, &vsock->send_pkt_list);
131 spin_unlock_bh(&vsock->send_pkt_list_lock);
133 /* We cannot finish yet if more buffers snuck in while
134 * re-enabling notify.
136 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
137 vhost_disable_notify(&vsock->dev, vq);
138 continue;
140 break;
143 if (out) {
144 virtio_transport_free_pkt(pkt);
145 vq_err(vq, "Expected 0 output buffers, got %u\n", out);
146 break;
149 len = iov_length(&vq->iov[out], in);
150 iov_iter_init(&iov_iter, READ, &vq->iov[out], in, len);
152 nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
153 if (nbytes != sizeof(pkt->hdr)) {
154 virtio_transport_free_pkt(pkt);
155 vq_err(vq, "Faulted on copying pkt hdr\n");
156 break;
159 nbytes = copy_to_iter(pkt->buf, pkt->len, &iov_iter);
160 if (nbytes != pkt->len) {
161 virtio_transport_free_pkt(pkt);
162 vq_err(vq, "Faulted on copying pkt buf\n");
163 break;
166 vhost_add_used(vq, head, sizeof(pkt->hdr) + pkt->len);
167 added = true;
169 if (pkt->reply) {
170 int val;
172 val = atomic_dec_return(&vsock->queued_replies);
174 /* Do we have resources to resume tx processing? */
175 if (val + 1 == tx_vq->num)
176 restart_tx = true;
179 virtio_transport_free_pkt(pkt);
181 if (added)
182 vhost_signal(&vsock->dev, vq);
184 out:
185 mutex_unlock(&vq->mutex);
187 if (restart_tx)
188 vhost_poll_queue(&tx_vq->poll);
191 static void vhost_transport_send_pkt_work(struct vhost_work *work)
193 struct vhost_virtqueue *vq;
194 struct vhost_vsock *vsock;
196 vsock = container_of(work, struct vhost_vsock, send_pkt_work);
197 vq = &vsock->vqs[VSOCK_VQ_RX];
199 vhost_transport_do_send_pkt(vsock, vq);
202 static int
203 vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
205 struct vhost_vsock *vsock;
206 int len = pkt->len;
208 /* Find the vhost_vsock according to guest context id */
209 vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
210 if (!vsock) {
211 virtio_transport_free_pkt(pkt);
212 return -ENODEV;
215 if (pkt->reply)
216 atomic_inc(&vsock->queued_replies);
218 spin_lock_bh(&vsock->send_pkt_list_lock);
219 list_add_tail(&pkt->list, &vsock->send_pkt_list);
220 spin_unlock_bh(&vsock->send_pkt_list_lock);
222 vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
223 return len;
226 static struct virtio_vsock_pkt *
227 vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
228 unsigned int out, unsigned int in)
230 struct virtio_vsock_pkt *pkt;
231 struct iov_iter iov_iter;
232 size_t nbytes;
233 size_t len;
235 if (in != 0) {
236 vq_err(vq, "Expected 0 input buffers, got %u\n", in);
237 return NULL;
240 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
241 if (!pkt)
242 return NULL;
244 len = iov_length(vq->iov, out);
245 iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
247 nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
248 if (nbytes != sizeof(pkt->hdr)) {
249 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
250 sizeof(pkt->hdr), nbytes);
251 kfree(pkt);
252 return NULL;
255 if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM)
256 pkt->len = le32_to_cpu(pkt->hdr.len);
258 /* No payload */
259 if (!pkt->len)
260 return pkt;
262 /* The pkt is too big */
263 if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
264 kfree(pkt);
265 return NULL;
268 pkt->buf = kmalloc(pkt->len, GFP_KERNEL);
269 if (!pkt->buf) {
270 kfree(pkt);
271 return NULL;
274 nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
275 if (nbytes != pkt->len) {
276 vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
277 pkt->len, nbytes);
278 virtio_transport_free_pkt(pkt);
279 return NULL;
282 return pkt;
285 /* Is there space left for replies to rx packets? */
286 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
288 struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
289 int val;
291 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
292 val = atomic_read(&vsock->queued_replies);
294 return val < vq->num;
297 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
299 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
300 poll.work);
301 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
302 dev);
303 struct virtio_vsock_pkt *pkt;
304 int head;
305 unsigned int out, in;
306 bool added = false;
308 mutex_lock(&vq->mutex);
310 if (!vq->private_data)
311 goto out;
313 vhost_disable_notify(&vsock->dev, vq);
314 for (;;) {
315 u32 len;
317 if (!vhost_vsock_more_replies(vsock)) {
318 /* Stop tx until the device processes already
319 * pending replies. Leave tx virtqueue
320 * callbacks disabled.
322 goto no_more_replies;
325 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
326 &out, &in, NULL, NULL);
327 if (head < 0)
328 break;
330 if (head == vq->num) {
331 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
332 vhost_disable_notify(&vsock->dev, vq);
333 continue;
335 break;
338 pkt = vhost_vsock_alloc_pkt(vq, out, in);
339 if (!pkt) {
340 vq_err(vq, "Faulted on pkt\n");
341 continue;
344 len = pkt->len;
346 /* Only accept correctly addressed packets */
347 if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid)
348 virtio_transport_recv_pkt(pkt);
349 else
350 virtio_transport_free_pkt(pkt);
352 vhost_add_used(vq, head, sizeof(pkt->hdr) + len);
353 added = true;
356 no_more_replies:
357 if (added)
358 vhost_signal(&vsock->dev, vq);
360 out:
361 mutex_unlock(&vq->mutex);
364 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
366 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
367 poll.work);
368 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
369 dev);
371 vhost_transport_do_send_pkt(vsock, vq);
374 static int vhost_vsock_start(struct vhost_vsock *vsock)
376 size_t i;
377 int ret;
379 mutex_lock(&vsock->dev.mutex);
381 ret = vhost_dev_check_owner(&vsock->dev);
382 if (ret)
383 goto err;
385 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
386 struct vhost_virtqueue *vq = &vsock->vqs[i];
388 mutex_lock(&vq->mutex);
390 if (!vhost_vq_access_ok(vq)) {
391 ret = -EFAULT;
392 mutex_unlock(&vq->mutex);
393 goto err_vq;
396 if (!vq->private_data) {
397 vq->private_data = vsock;
398 vhost_vq_init_access(vq);
401 mutex_unlock(&vq->mutex);
404 mutex_unlock(&vsock->dev.mutex);
405 return 0;
407 err_vq:
408 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
409 struct vhost_virtqueue *vq = &vsock->vqs[i];
411 mutex_lock(&vq->mutex);
412 vq->private_data = NULL;
413 mutex_unlock(&vq->mutex);
415 err:
416 mutex_unlock(&vsock->dev.mutex);
417 return ret;
420 static int vhost_vsock_stop(struct vhost_vsock *vsock)
422 size_t i;
423 int ret;
425 mutex_lock(&vsock->dev.mutex);
427 ret = vhost_dev_check_owner(&vsock->dev);
428 if (ret)
429 goto err;
431 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
432 struct vhost_virtqueue *vq = &vsock->vqs[i];
434 mutex_lock(&vq->mutex);
435 vq->private_data = NULL;
436 mutex_unlock(&vq->mutex);
439 err:
440 mutex_unlock(&vsock->dev.mutex);
441 return ret;
444 static void vhost_vsock_free(struct vhost_vsock *vsock)
446 kvfree(vsock);
449 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
451 struct vhost_virtqueue **vqs;
452 struct vhost_vsock *vsock;
453 int ret;
455 /* This struct is large and allocation could fail, fall back to vmalloc
456 * if there is no other way.
458 vsock = kzalloc(sizeof(*vsock), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
459 if (!vsock) {
460 vsock = vmalloc(sizeof(*vsock));
461 if (!vsock)
462 return -ENOMEM;
465 vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
466 if (!vqs) {
467 ret = -ENOMEM;
468 goto out;
471 atomic_set(&vsock->queued_replies, 0);
473 vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
474 vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
475 vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
476 vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
478 vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs));
480 file->private_data = vsock;
481 spin_lock_init(&vsock->send_pkt_list_lock);
482 INIT_LIST_HEAD(&vsock->send_pkt_list);
483 vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
485 spin_lock_bh(&vhost_vsock_lock);
486 list_add_tail(&vsock->list, &vhost_vsock_list);
487 spin_unlock_bh(&vhost_vsock_lock);
488 return 0;
490 out:
491 vhost_vsock_free(vsock);
492 return ret;
495 static void vhost_vsock_flush(struct vhost_vsock *vsock)
497 int i;
499 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
500 if (vsock->vqs[i].handle_kick)
501 vhost_poll_flush(&vsock->vqs[i].poll);
502 vhost_work_flush(&vsock->dev, &vsock->send_pkt_work);
505 static void vhost_vsock_reset_orphans(struct sock *sk)
507 struct vsock_sock *vsk = vsock_sk(sk);
509 /* vmci_transport.c doesn't take sk_lock here either. At least we're
510 * under vsock_table_lock so the sock cannot disappear while we're
511 * executing.
514 if (!vhost_vsock_get(vsk->remote_addr.svm_cid)) {
515 sock_set_flag(sk, SOCK_DONE);
516 vsk->peer_shutdown = SHUTDOWN_MASK;
517 sk->sk_state = SS_UNCONNECTED;
518 sk->sk_err = ECONNRESET;
519 sk->sk_error_report(sk);
523 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
525 struct vhost_vsock *vsock = file->private_data;
527 spin_lock_bh(&vhost_vsock_lock);
528 list_del(&vsock->list);
529 spin_unlock_bh(&vhost_vsock_lock);
531 /* Iterating over all connections for all CIDs to find orphans is
532 * inefficient. Room for improvement here. */
533 vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
535 vhost_vsock_stop(vsock);
536 vhost_vsock_flush(vsock);
537 vhost_dev_stop(&vsock->dev);
539 spin_lock_bh(&vsock->send_pkt_list_lock);
540 while (!list_empty(&vsock->send_pkt_list)) {
541 struct virtio_vsock_pkt *pkt;
543 pkt = list_first_entry(&vsock->send_pkt_list,
544 struct virtio_vsock_pkt, list);
545 list_del_init(&pkt->list);
546 virtio_transport_free_pkt(pkt);
548 spin_unlock_bh(&vsock->send_pkt_list_lock);
550 vhost_dev_cleanup(&vsock->dev, false);
551 kfree(vsock->dev.vqs);
552 vhost_vsock_free(vsock);
553 return 0;
556 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
558 struct vhost_vsock *other;
560 /* Refuse reserved CIDs */
561 if (guest_cid <= VMADDR_CID_HOST ||
562 guest_cid == U32_MAX)
563 return -EINVAL;
565 /* 64-bit CIDs are not yet supported */
566 if (guest_cid > U32_MAX)
567 return -EINVAL;
569 /* Refuse if CID is already in use */
570 spin_lock_bh(&vhost_vsock_lock);
571 other = __vhost_vsock_get(guest_cid);
572 if (other && other != vsock) {
573 spin_unlock_bh(&vhost_vsock_lock);
574 return -EADDRINUSE;
576 vsock->guest_cid = guest_cid;
577 spin_unlock_bh(&vhost_vsock_lock);
579 return 0;
582 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
584 struct vhost_virtqueue *vq;
585 int i;
587 if (features & ~VHOST_VSOCK_FEATURES)
588 return -EOPNOTSUPP;
590 mutex_lock(&vsock->dev.mutex);
591 if ((features & (1 << VHOST_F_LOG_ALL)) &&
592 !vhost_log_access_ok(&vsock->dev)) {
593 mutex_unlock(&vsock->dev.mutex);
594 return -EFAULT;
597 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
598 vq = &vsock->vqs[i];
599 mutex_lock(&vq->mutex);
600 vq->acked_features = features;
601 mutex_unlock(&vq->mutex);
603 mutex_unlock(&vsock->dev.mutex);
604 return 0;
607 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
608 unsigned long arg)
610 struct vhost_vsock *vsock = f->private_data;
611 void __user *argp = (void __user *)arg;
612 u64 guest_cid;
613 u64 features;
614 int start;
615 int r;
617 switch (ioctl) {
618 case VHOST_VSOCK_SET_GUEST_CID:
619 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
620 return -EFAULT;
621 return vhost_vsock_set_cid(vsock, guest_cid);
622 case VHOST_VSOCK_SET_RUNNING:
623 if (copy_from_user(&start, argp, sizeof(start)))
624 return -EFAULT;
625 if (start)
626 return vhost_vsock_start(vsock);
627 else
628 return vhost_vsock_stop(vsock);
629 case VHOST_GET_FEATURES:
630 features = VHOST_VSOCK_FEATURES;
631 if (copy_to_user(argp, &features, sizeof(features)))
632 return -EFAULT;
633 return 0;
634 case VHOST_SET_FEATURES:
635 if (copy_from_user(&features, argp, sizeof(features)))
636 return -EFAULT;
637 return vhost_vsock_set_features(vsock, features);
638 default:
639 mutex_lock(&vsock->dev.mutex);
640 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
641 if (r == -ENOIOCTLCMD)
642 r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
643 else
644 vhost_vsock_flush(vsock);
645 mutex_unlock(&vsock->dev.mutex);
646 return r;
650 static const struct file_operations vhost_vsock_fops = {
651 .owner = THIS_MODULE,
652 .open = vhost_vsock_dev_open,
653 .release = vhost_vsock_dev_release,
654 .llseek = noop_llseek,
655 .unlocked_ioctl = vhost_vsock_dev_ioctl,
658 static struct miscdevice vhost_vsock_misc = {
659 .minor = MISC_DYNAMIC_MINOR,
660 .name = "vhost-vsock",
661 .fops = &vhost_vsock_fops,
664 static struct virtio_transport vhost_transport = {
665 .transport = {
666 .get_local_cid = vhost_transport_get_local_cid,
668 .init = virtio_transport_do_socket_init,
669 .destruct = virtio_transport_destruct,
670 .release = virtio_transport_release,
671 .connect = virtio_transport_connect,
672 .shutdown = virtio_transport_shutdown,
674 .dgram_enqueue = virtio_transport_dgram_enqueue,
675 .dgram_dequeue = virtio_transport_dgram_dequeue,
676 .dgram_bind = virtio_transport_dgram_bind,
677 .dgram_allow = virtio_transport_dgram_allow,
679 .stream_enqueue = virtio_transport_stream_enqueue,
680 .stream_dequeue = virtio_transport_stream_dequeue,
681 .stream_has_data = virtio_transport_stream_has_data,
682 .stream_has_space = virtio_transport_stream_has_space,
683 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
684 .stream_is_active = virtio_transport_stream_is_active,
685 .stream_allow = virtio_transport_stream_allow,
687 .notify_poll_in = virtio_transport_notify_poll_in,
688 .notify_poll_out = virtio_transport_notify_poll_out,
689 .notify_recv_init = virtio_transport_notify_recv_init,
690 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
691 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
692 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
693 .notify_send_init = virtio_transport_notify_send_init,
694 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
695 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
696 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
698 .set_buffer_size = virtio_transport_set_buffer_size,
699 .set_min_buffer_size = virtio_transport_set_min_buffer_size,
700 .set_max_buffer_size = virtio_transport_set_max_buffer_size,
701 .get_buffer_size = virtio_transport_get_buffer_size,
702 .get_min_buffer_size = virtio_transport_get_min_buffer_size,
703 .get_max_buffer_size = virtio_transport_get_max_buffer_size,
706 .send_pkt = vhost_transport_send_pkt,
709 static int __init vhost_vsock_init(void)
711 int ret;
713 ret = vsock_core_init(&vhost_transport.transport);
714 if (ret < 0)
715 return ret;
716 return misc_register(&vhost_vsock_misc);
719 static void __exit vhost_vsock_exit(void)
721 misc_deregister(&vhost_vsock_misc);
722 vsock_core_exit();
725 module_init(vhost_vsock_init);
726 module_exit(vhost_vsock_exit);
727 MODULE_LICENSE("GPL v2");
728 MODULE_AUTHOR("Asias He");
729 MODULE_DESCRIPTION("vhost transport for vsock ");