xref: /openbmc/linux/drivers/vhost/net.c (revision 565d76cb)
1 /* Copyright (C) 2009 Red Hat, Inc.
2  * Author: Michael S. Tsirkin <mst@redhat.com>
3  *
4  * This work is licensed under the terms of the GNU GPL, version 2.
5  *
6  * virtio-net server in host kernel.
7  */
8 
9 #include <linux/compat.h>
10 #include <linux/eventfd.h>
11 #include <linux/vhost.h>
12 #include <linux/virtio_net.h>
13 #include <linux/miscdevice.h>
14 #include <linux/module.h>
15 #include <linux/mutex.h>
16 #include <linux/workqueue.h>
17 #include <linux/rcupdate.h>
18 #include <linux/file.h>
19 #include <linux/slab.h>
20 
21 #include <linux/net.h>
22 #include <linux/if_packet.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_tun.h>
25 #include <linux/if_macvlan.h>
26 
27 #include <net/sock.h>
28 
29 #include "vhost.h"
30 
31 /* Max number of bytes transferred before requeueing the job.
32  * Using this limit prevents one virtqueue from starving others. */
33 #define VHOST_NET_WEIGHT 0x80000
34 
35 enum {
36 	VHOST_NET_VQ_RX = 0,
37 	VHOST_NET_VQ_TX = 1,
38 	VHOST_NET_VQ_MAX = 2,
39 };
40 
41 enum vhost_net_poll_state {
42 	VHOST_NET_POLL_DISABLED = 0,
43 	VHOST_NET_POLL_STARTED = 1,
44 	VHOST_NET_POLL_STOPPED = 2,
45 };
46 
47 struct vhost_net {
48 	struct vhost_dev dev;
49 	struct vhost_virtqueue vqs[VHOST_NET_VQ_MAX];
50 	struct vhost_poll poll[VHOST_NET_VQ_MAX];
51 	/* Tells us whether we are polling a socket for TX.
52 	 * We only do this when socket buffer fills up.
53 	 * Protected by tx vq lock. */
54 	enum vhost_net_poll_state tx_poll_state;
55 };
56 
57 /* Pop first len bytes from iovec. Return number of segments used. */
58 static int move_iovec_hdr(struct iovec *from, struct iovec *to,
59 			  size_t len, int iov_count)
60 {
61 	int seg = 0;
62 	size_t size;
63 
64 	while (len && seg < iov_count) {
65 		size = min(from->iov_len, len);
66 		to->iov_base = from->iov_base;
67 		to->iov_len = size;
68 		from->iov_len -= size;
69 		from->iov_base += size;
70 		len -= size;
71 		++from;
72 		++to;
73 		++seg;
74 	}
75 	return seg;
76 }
77 /* Copy iovec entries for len bytes from iovec. */
78 static void copy_iovec_hdr(const struct iovec *from, struct iovec *to,
79 			   size_t len, int iovcount)
80 {
81 	int seg = 0;
82 	size_t size;
83 
84 	while (len && seg < iovcount) {
85 		size = min(from->iov_len, len);
86 		to->iov_base = from->iov_base;
87 		to->iov_len = size;
88 		len -= size;
89 		++from;
90 		++to;
91 		++seg;
92 	}
93 }
94 
95 /* Caller must have TX VQ lock */
96 static void tx_poll_stop(struct vhost_net *net)
97 {
98 	if (likely(net->tx_poll_state != VHOST_NET_POLL_STARTED))
99 		return;
100 	vhost_poll_stop(net->poll + VHOST_NET_VQ_TX);
101 	net->tx_poll_state = VHOST_NET_POLL_STOPPED;
102 }
103 
104 /* Caller must have TX VQ lock */
105 static void tx_poll_start(struct vhost_net *net, struct socket *sock)
106 {
107 	if (unlikely(net->tx_poll_state != VHOST_NET_POLL_STOPPED))
108 		return;
109 	vhost_poll_start(net->poll + VHOST_NET_VQ_TX, sock->file);
110 	net->tx_poll_state = VHOST_NET_POLL_STARTED;
111 }
112 
113 /* Expects to be always run from workqueue - which acts as
114  * read-size critical section for our kind of RCU. */
115 static void handle_tx(struct vhost_net *net)
116 {
117 	struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_TX];
118 	unsigned out, in, s;
119 	int head;
120 	struct msghdr msg = {
121 		.msg_name = NULL,
122 		.msg_namelen = 0,
123 		.msg_control = NULL,
124 		.msg_controllen = 0,
125 		.msg_iov = vq->iov,
126 		.msg_flags = MSG_DONTWAIT,
127 	};
128 	size_t len, total_len = 0;
129 	int err, wmem;
130 	size_t hdr_size;
131 	struct socket *sock;
132 
133 	/* TODO: check that we are running from vhost_worker? */
134 	sock = rcu_dereference_check(vq->private_data, 1);
135 	if (!sock)
136 		return;
137 
138 	wmem = atomic_read(&sock->sk->sk_wmem_alloc);
139 	if (wmem >= sock->sk->sk_sndbuf) {
140 		mutex_lock(&vq->mutex);
141 		tx_poll_start(net, sock);
142 		mutex_unlock(&vq->mutex);
143 		return;
144 	}
145 
146 	mutex_lock(&vq->mutex);
147 	vhost_disable_notify(vq);
148 
149 	if (wmem < sock->sk->sk_sndbuf / 2)
150 		tx_poll_stop(net);
151 	hdr_size = vq->vhost_hlen;
152 
153 	for (;;) {
154 		head = vhost_get_vq_desc(&net->dev, vq, vq->iov,
155 					 ARRAY_SIZE(vq->iov),
156 					 &out, &in,
157 					 NULL, NULL);
158 		/* On error, stop handling until the next kick. */
159 		if (unlikely(head < 0))
160 			break;
161 		/* Nothing new?  Wait for eventfd to tell us they refilled. */
162 		if (head == vq->num) {
163 			wmem = atomic_read(&sock->sk->sk_wmem_alloc);
164 			if (wmem >= sock->sk->sk_sndbuf * 3 / 4) {
165 				tx_poll_start(net, sock);
166 				set_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
167 				break;
168 			}
169 			if (unlikely(vhost_enable_notify(vq))) {
170 				vhost_disable_notify(vq);
171 				continue;
172 			}
173 			break;
174 		}
175 		if (in) {
176 			vq_err(vq, "Unexpected descriptor format for TX: "
177 			       "out %d, int %d\n", out, in);
178 			break;
179 		}
180 		/* Skip header. TODO: support TSO. */
181 		s = move_iovec_hdr(vq->iov, vq->hdr, hdr_size, out);
182 		msg.msg_iovlen = out;
183 		len = iov_length(vq->iov, out);
184 		/* Sanity check */
185 		if (!len) {
186 			vq_err(vq, "Unexpected header len for TX: "
187 			       "%zd expected %zd\n",
188 			       iov_length(vq->hdr, s), hdr_size);
189 			break;
190 		}
191 		/* TODO: Check specific error and bomb out unless ENOBUFS? */
192 		err = sock->ops->sendmsg(NULL, sock, &msg, len);
193 		if (unlikely(err < 0)) {
194 			vhost_discard_vq_desc(vq, 1);
195 			tx_poll_start(net, sock);
196 			break;
197 		}
198 		if (err != len)
199 			pr_debug("Truncated TX packet: "
200 				 " len %d != %zd\n", err, len);
201 		vhost_add_used_and_signal(&net->dev, vq, head, 0);
202 		total_len += len;
203 		if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
204 			vhost_poll_queue(&vq->poll);
205 			break;
206 		}
207 	}
208 
209 	mutex_unlock(&vq->mutex);
210 }
211 
212 static int peek_head_len(struct sock *sk)
213 {
214 	struct sk_buff *head;
215 	int len = 0;
216 	unsigned long flags;
217 
218 	spin_lock_irqsave(&sk->sk_receive_queue.lock, flags);
219 	head = skb_peek(&sk->sk_receive_queue);
220 	if (likely(head))
221 		len = head->len;
222 	spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags);
223 	return len;
224 }
225 
226 /* This is a multi-buffer version of vhost_get_desc, that works if
227  *	vq has read descriptors only.
228  * @vq		- the relevant virtqueue
229  * @datalen	- data length we'll be reading
230  * @iovcount	- returned count of io vectors we fill
231  * @log		- vhost log
232  * @log_num	- log offset
233  * @quota       - headcount quota, 1 for big buffer
234  *	returns number of buffer heads allocated, negative on error
235  */
236 static int get_rx_bufs(struct vhost_virtqueue *vq,
237 		       struct vring_used_elem *heads,
238 		       int datalen,
239 		       unsigned *iovcount,
240 		       struct vhost_log *log,
241 		       unsigned *log_num,
242 		       unsigned int quota)
243 {
244 	unsigned int out, in;
245 	int seg = 0;
246 	int headcount = 0;
247 	unsigned d;
248 	int r, nlogs = 0;
249 
250 	while (datalen > 0 && headcount < quota) {
251 		if (unlikely(seg >= UIO_MAXIOV)) {
252 			r = -ENOBUFS;
253 			goto err;
254 		}
255 		d = vhost_get_vq_desc(vq->dev, vq, vq->iov + seg,
256 				      ARRAY_SIZE(vq->iov) - seg, &out,
257 				      &in, log, log_num);
258 		if (d == vq->num) {
259 			r = 0;
260 			goto err;
261 		}
262 		if (unlikely(out || in <= 0)) {
263 			vq_err(vq, "unexpected descriptor format for RX: "
264 				"out %d, in %d\n", out, in);
265 			r = -EINVAL;
266 			goto err;
267 		}
268 		if (unlikely(log)) {
269 			nlogs += *log_num;
270 			log += *log_num;
271 		}
272 		heads[headcount].id = d;
273 		heads[headcount].len = iov_length(vq->iov + seg, in);
274 		datalen -= heads[headcount].len;
275 		++headcount;
276 		seg += in;
277 	}
278 	heads[headcount - 1].len += datalen;
279 	*iovcount = seg;
280 	if (unlikely(log))
281 		*log_num = nlogs;
282 	return headcount;
283 err:
284 	vhost_discard_vq_desc(vq, headcount);
285 	return r;
286 }
287 
288 /* Expects to be always run from workqueue - which acts as
289  * read-size critical section for our kind of RCU. */
290 static void handle_rx(struct vhost_net *net)
291 {
292 	struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_RX];
293 	unsigned uninitialized_var(in), log;
294 	struct vhost_log *vq_log;
295 	struct msghdr msg = {
296 		.msg_name = NULL,
297 		.msg_namelen = 0,
298 		.msg_control = NULL, /* FIXME: get and handle RX aux data. */
299 		.msg_controllen = 0,
300 		.msg_iov = vq->iov,
301 		.msg_flags = MSG_DONTWAIT,
302 	};
303 	struct virtio_net_hdr_mrg_rxbuf hdr = {
304 		.hdr.flags = 0,
305 		.hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE
306 	};
307 	size_t total_len = 0;
308 	int err, headcount, mergeable;
309 	size_t vhost_hlen, sock_hlen;
310 	size_t vhost_len, sock_len;
311 	/* TODO: check that we are running from vhost_worker? */
312 	struct socket *sock = rcu_dereference_check(vq->private_data, 1);
313 
314 	if (!sock)
315 		return;
316 
317 	mutex_lock(&vq->mutex);
318 	vhost_disable_notify(vq);
319 	vhost_hlen = vq->vhost_hlen;
320 	sock_hlen = vq->sock_hlen;
321 
322 	vq_log = unlikely(vhost_has_feature(&net->dev, VHOST_F_LOG_ALL)) ?
323 		vq->log : NULL;
324 	mergeable = vhost_has_feature(&net->dev, VIRTIO_NET_F_MRG_RXBUF);
325 
326 	while ((sock_len = peek_head_len(sock->sk))) {
327 		sock_len += sock_hlen;
328 		vhost_len = sock_len + vhost_hlen;
329 		headcount = get_rx_bufs(vq, vq->heads, vhost_len,
330 					&in, vq_log, &log,
331 					likely(mergeable) ? UIO_MAXIOV : 1);
332 		/* On error, stop handling until the next kick. */
333 		if (unlikely(headcount < 0))
334 			break;
335 		/* OK, now we need to know about added descriptors. */
336 		if (!headcount) {
337 			if (unlikely(vhost_enable_notify(vq))) {
338 				/* They have slipped one in as we were
339 				 * doing that: check again. */
340 				vhost_disable_notify(vq);
341 				continue;
342 			}
343 			/* Nothing new?  Wait for eventfd to tell us
344 			 * they refilled. */
345 			break;
346 		}
347 		/* We don't need to be notified again. */
348 		if (unlikely((vhost_hlen)))
349 			/* Skip header. TODO: support TSO. */
350 			move_iovec_hdr(vq->iov, vq->hdr, vhost_hlen, in);
351 		else
352 			/* Copy the header for use in VIRTIO_NET_F_MRG_RXBUF:
353 			 * needed because recvmsg can modify msg_iov. */
354 			copy_iovec_hdr(vq->iov, vq->hdr, sock_hlen, in);
355 		msg.msg_iovlen = in;
356 		err = sock->ops->recvmsg(NULL, sock, &msg,
357 					 sock_len, MSG_DONTWAIT | MSG_TRUNC);
358 		/* Userspace might have consumed the packet meanwhile:
359 		 * it's not supposed to do this usually, but might be hard
360 		 * to prevent. Discard data we got (if any) and keep going. */
361 		if (unlikely(err != sock_len)) {
362 			pr_debug("Discarded rx packet: "
363 				 " len %d, expected %zd\n", err, sock_len);
364 			vhost_discard_vq_desc(vq, headcount);
365 			continue;
366 		}
367 		if (unlikely(vhost_hlen) &&
368 		    memcpy_toiovecend(vq->hdr, (unsigned char *)&hdr, 0,
369 				      vhost_hlen)) {
370 			vq_err(vq, "Unable to write vnet_hdr at addr %p\n",
371 			       vq->iov->iov_base);
372 			break;
373 		}
374 		/* TODO: Should check and handle checksum. */
375 		if (likely(mergeable) &&
376 		    memcpy_toiovecend(vq->hdr, (unsigned char *)&headcount,
377 				      offsetof(typeof(hdr), num_buffers),
378 				      sizeof hdr.num_buffers)) {
379 			vq_err(vq, "Failed num_buffers write");
380 			vhost_discard_vq_desc(vq, headcount);
381 			break;
382 		}
383 		vhost_add_used_and_signal_n(&net->dev, vq, vq->heads,
384 					    headcount);
385 		if (unlikely(vq_log))
386 			vhost_log_write(vq, vq_log, log, vhost_len);
387 		total_len += vhost_len;
388 		if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
389 			vhost_poll_queue(&vq->poll);
390 			break;
391 		}
392 	}
393 
394 	mutex_unlock(&vq->mutex);
395 }
396 
397 static void handle_tx_kick(struct vhost_work *work)
398 {
399 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
400 						  poll.work);
401 	struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
402 
403 	handle_tx(net);
404 }
405 
406 static void handle_rx_kick(struct vhost_work *work)
407 {
408 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
409 						  poll.work);
410 	struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
411 
412 	handle_rx(net);
413 }
414 
415 static void handle_tx_net(struct vhost_work *work)
416 {
417 	struct vhost_net *net = container_of(work, struct vhost_net,
418 					     poll[VHOST_NET_VQ_TX].work);
419 	handle_tx(net);
420 }
421 
422 static void handle_rx_net(struct vhost_work *work)
423 {
424 	struct vhost_net *net = container_of(work, struct vhost_net,
425 					     poll[VHOST_NET_VQ_RX].work);
426 	handle_rx(net);
427 }
428 
429 static int vhost_net_open(struct inode *inode, struct file *f)
430 {
431 	struct vhost_net *n = kmalloc(sizeof *n, GFP_KERNEL);
432 	struct vhost_dev *dev;
433 	int r;
434 
435 	if (!n)
436 		return -ENOMEM;
437 
438 	dev = &n->dev;
439 	n->vqs[VHOST_NET_VQ_TX].handle_kick = handle_tx_kick;
440 	n->vqs[VHOST_NET_VQ_RX].handle_kick = handle_rx_kick;
441 	r = vhost_dev_init(dev, n->vqs, VHOST_NET_VQ_MAX);
442 	if (r < 0) {
443 		kfree(n);
444 		return r;
445 	}
446 
447 	vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT, dev);
448 	vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN, dev);
449 	n->tx_poll_state = VHOST_NET_POLL_DISABLED;
450 
451 	f->private_data = n;
452 
453 	return 0;
454 }
455 
456 static void vhost_net_disable_vq(struct vhost_net *n,
457 				 struct vhost_virtqueue *vq)
458 {
459 	if (!vq->private_data)
460 		return;
461 	if (vq == n->vqs + VHOST_NET_VQ_TX) {
462 		tx_poll_stop(n);
463 		n->tx_poll_state = VHOST_NET_POLL_DISABLED;
464 	} else
465 		vhost_poll_stop(n->poll + VHOST_NET_VQ_RX);
466 }
467 
468 static void vhost_net_enable_vq(struct vhost_net *n,
469 				struct vhost_virtqueue *vq)
470 {
471 	struct socket *sock;
472 
473 	sock = rcu_dereference_protected(vq->private_data,
474 					 lockdep_is_held(&vq->mutex));
475 	if (!sock)
476 		return;
477 	if (vq == n->vqs + VHOST_NET_VQ_TX) {
478 		n->tx_poll_state = VHOST_NET_POLL_STOPPED;
479 		tx_poll_start(n, sock);
480 	} else
481 		vhost_poll_start(n->poll + VHOST_NET_VQ_RX, sock->file);
482 }
483 
484 static struct socket *vhost_net_stop_vq(struct vhost_net *n,
485 					struct vhost_virtqueue *vq)
486 {
487 	struct socket *sock;
488 
489 	mutex_lock(&vq->mutex);
490 	sock = rcu_dereference_protected(vq->private_data,
491 					 lockdep_is_held(&vq->mutex));
492 	vhost_net_disable_vq(n, vq);
493 	rcu_assign_pointer(vq->private_data, NULL);
494 	mutex_unlock(&vq->mutex);
495 	return sock;
496 }
497 
498 static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock,
499 			   struct socket **rx_sock)
500 {
501 	*tx_sock = vhost_net_stop_vq(n, n->vqs + VHOST_NET_VQ_TX);
502 	*rx_sock = vhost_net_stop_vq(n, n->vqs + VHOST_NET_VQ_RX);
503 }
504 
505 static void vhost_net_flush_vq(struct vhost_net *n, int index)
506 {
507 	vhost_poll_flush(n->poll + index);
508 	vhost_poll_flush(&n->dev.vqs[index].poll);
509 }
510 
511 static void vhost_net_flush(struct vhost_net *n)
512 {
513 	vhost_net_flush_vq(n, VHOST_NET_VQ_TX);
514 	vhost_net_flush_vq(n, VHOST_NET_VQ_RX);
515 }
516 
517 static int vhost_net_release(struct inode *inode, struct file *f)
518 {
519 	struct vhost_net *n = f->private_data;
520 	struct socket *tx_sock;
521 	struct socket *rx_sock;
522 
523 	vhost_net_stop(n, &tx_sock, &rx_sock);
524 	vhost_net_flush(n);
525 	vhost_dev_cleanup(&n->dev);
526 	if (tx_sock)
527 		fput(tx_sock->file);
528 	if (rx_sock)
529 		fput(rx_sock->file);
530 	/* We do an extra flush before freeing memory,
531 	 * since jobs can re-queue themselves. */
532 	vhost_net_flush(n);
533 	kfree(n);
534 	return 0;
535 }
536 
537 static struct socket *get_raw_socket(int fd)
538 {
539 	struct {
540 		struct sockaddr_ll sa;
541 		char  buf[MAX_ADDR_LEN];
542 	} uaddr;
543 	int uaddr_len = sizeof uaddr, r;
544 	struct socket *sock = sockfd_lookup(fd, &r);
545 
546 	if (!sock)
547 		return ERR_PTR(-ENOTSOCK);
548 
549 	/* Parameter checking */
550 	if (sock->sk->sk_type != SOCK_RAW) {
551 		r = -ESOCKTNOSUPPORT;
552 		goto err;
553 	}
554 
555 	r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa,
556 			       &uaddr_len, 0);
557 	if (r)
558 		goto err;
559 
560 	if (uaddr.sa.sll_family != AF_PACKET) {
561 		r = -EPFNOSUPPORT;
562 		goto err;
563 	}
564 	return sock;
565 err:
566 	fput(sock->file);
567 	return ERR_PTR(r);
568 }
569 
570 static struct socket *get_tap_socket(int fd)
571 {
572 	struct file *file = fget(fd);
573 	struct socket *sock;
574 
575 	if (!file)
576 		return ERR_PTR(-EBADF);
577 	sock = tun_get_socket(file);
578 	if (!IS_ERR(sock))
579 		return sock;
580 	sock = macvtap_get_socket(file);
581 	if (IS_ERR(sock))
582 		fput(file);
583 	return sock;
584 }
585 
586 static struct socket *get_socket(int fd)
587 {
588 	struct socket *sock;
589 
590 	/* special case to disable backend */
591 	if (fd == -1)
592 		return NULL;
593 	sock = get_raw_socket(fd);
594 	if (!IS_ERR(sock))
595 		return sock;
596 	sock = get_tap_socket(fd);
597 	if (!IS_ERR(sock))
598 		return sock;
599 	return ERR_PTR(-ENOTSOCK);
600 }
601 
602 static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
603 {
604 	struct socket *sock, *oldsock;
605 	struct vhost_virtqueue *vq;
606 	int r;
607 
608 	mutex_lock(&n->dev.mutex);
609 	r = vhost_dev_check_owner(&n->dev);
610 	if (r)
611 		goto err;
612 
613 	if (index >= VHOST_NET_VQ_MAX) {
614 		r = -ENOBUFS;
615 		goto err;
616 	}
617 	vq = n->vqs + index;
618 	mutex_lock(&vq->mutex);
619 
620 	/* Verify that ring has been setup correctly. */
621 	if (!vhost_vq_access_ok(vq)) {
622 		r = -EFAULT;
623 		goto err_vq;
624 	}
625 	sock = get_socket(fd);
626 	if (IS_ERR(sock)) {
627 		r = PTR_ERR(sock);
628 		goto err_vq;
629 	}
630 
631 	/* start polling new socket */
632 	oldsock = rcu_dereference_protected(vq->private_data,
633 					    lockdep_is_held(&vq->mutex));
634 	if (sock != oldsock) {
635 		vhost_net_disable_vq(n, vq);
636 		rcu_assign_pointer(vq->private_data, sock);
637 		vhost_net_enable_vq(n, vq);
638 	}
639 
640 	mutex_unlock(&vq->mutex);
641 
642 	if (oldsock) {
643 		vhost_net_flush_vq(n, index);
644 		fput(oldsock->file);
645 	}
646 
647 	mutex_unlock(&n->dev.mutex);
648 	return 0;
649 
650 err_vq:
651 	mutex_unlock(&vq->mutex);
652 err:
653 	mutex_unlock(&n->dev.mutex);
654 	return r;
655 }
656 
657 static long vhost_net_reset_owner(struct vhost_net *n)
658 {
659 	struct socket *tx_sock = NULL;
660 	struct socket *rx_sock = NULL;
661 	long err;
662 
663 	mutex_lock(&n->dev.mutex);
664 	err = vhost_dev_check_owner(&n->dev);
665 	if (err)
666 		goto done;
667 	vhost_net_stop(n, &tx_sock, &rx_sock);
668 	vhost_net_flush(n);
669 	err = vhost_dev_reset_owner(&n->dev);
670 done:
671 	mutex_unlock(&n->dev.mutex);
672 	if (tx_sock)
673 		fput(tx_sock->file);
674 	if (rx_sock)
675 		fput(rx_sock->file);
676 	return err;
677 }
678 
679 static int vhost_net_set_features(struct vhost_net *n, u64 features)
680 {
681 	size_t vhost_hlen, sock_hlen, hdr_len;
682 	int i;
683 
684 	hdr_len = (features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ?
685 			sizeof(struct virtio_net_hdr_mrg_rxbuf) :
686 			sizeof(struct virtio_net_hdr);
687 	if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) {
688 		/* vhost provides vnet_hdr */
689 		vhost_hlen = hdr_len;
690 		sock_hlen = 0;
691 	} else {
692 		/* socket provides vnet_hdr */
693 		vhost_hlen = 0;
694 		sock_hlen = hdr_len;
695 	}
696 	mutex_lock(&n->dev.mutex);
697 	if ((features & (1 << VHOST_F_LOG_ALL)) &&
698 	    !vhost_log_access_ok(&n->dev)) {
699 		mutex_unlock(&n->dev.mutex);
700 		return -EFAULT;
701 	}
702 	n->dev.acked_features = features;
703 	smp_wmb();
704 	for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
705 		mutex_lock(&n->vqs[i].mutex);
706 		n->vqs[i].vhost_hlen = vhost_hlen;
707 		n->vqs[i].sock_hlen = sock_hlen;
708 		mutex_unlock(&n->vqs[i].mutex);
709 	}
710 	vhost_net_flush(n);
711 	mutex_unlock(&n->dev.mutex);
712 	return 0;
713 }
714 
715 static long vhost_net_ioctl(struct file *f, unsigned int ioctl,
716 			    unsigned long arg)
717 {
718 	struct vhost_net *n = f->private_data;
719 	void __user *argp = (void __user *)arg;
720 	u64 __user *featurep = argp;
721 	struct vhost_vring_file backend;
722 	u64 features;
723 	int r;
724 
725 	switch (ioctl) {
726 	case VHOST_NET_SET_BACKEND:
727 		if (copy_from_user(&backend, argp, sizeof backend))
728 			return -EFAULT;
729 		return vhost_net_set_backend(n, backend.index, backend.fd);
730 	case VHOST_GET_FEATURES:
731 		features = VHOST_FEATURES;
732 		if (copy_to_user(featurep, &features, sizeof features))
733 			return -EFAULT;
734 		return 0;
735 	case VHOST_SET_FEATURES:
736 		if (copy_from_user(&features, featurep, sizeof features))
737 			return -EFAULT;
738 		if (features & ~VHOST_FEATURES)
739 			return -EOPNOTSUPP;
740 		return vhost_net_set_features(n, features);
741 	case VHOST_RESET_OWNER:
742 		return vhost_net_reset_owner(n);
743 	default:
744 		mutex_lock(&n->dev.mutex);
745 		r = vhost_dev_ioctl(&n->dev, ioctl, arg);
746 		vhost_net_flush(n);
747 		mutex_unlock(&n->dev.mutex);
748 		return r;
749 	}
750 }
751 
752 #ifdef CONFIG_COMPAT
753 static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl,
754 				   unsigned long arg)
755 {
756 	return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
757 }
758 #endif
759 
760 static const struct file_operations vhost_net_fops = {
761 	.owner          = THIS_MODULE,
762 	.release        = vhost_net_release,
763 	.unlocked_ioctl = vhost_net_ioctl,
764 #ifdef CONFIG_COMPAT
765 	.compat_ioctl   = vhost_net_compat_ioctl,
766 #endif
767 	.open           = vhost_net_open,
768 	.llseek		= noop_llseek,
769 };
770 
771 static struct miscdevice vhost_net_misc = {
772 	MISC_DYNAMIC_MINOR,
773 	"vhost-net",
774 	&vhost_net_fops,
775 };
776 
777 static int vhost_net_init(void)
778 {
779 	return misc_register(&vhost_net_misc);
780 }
781 module_init(vhost_net_init);
782 
783 static void vhost_net_exit(void)
784 {
785 	misc_deregister(&vhost_net_misc);
786 }
787 module_exit(vhost_net_exit);
788 
789 MODULE_VERSION("0.0.1");
790 MODULE_LICENSE("GPL v2");
791 MODULE_AUTHOR("Michael S. Tsirkin");
792 MODULE_DESCRIPTION("Host kernel accelerator for virtio net");
793