1 /* Copyright (C) 2009 Red Hat, Inc.
2 * Author: Michael S. Tsirkin <mst@redhat.com>
4 * This work is licensed under the terms of the GNU GPL, version 2.
6 * virtio-net server in host kernel.
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/moduleparam.h>
16 #include <linux/mutex.h>
17 #include <linux/workqueue.h>
18 #include <linux/rcupdate.h>
19 #include <linux/file.h>
20 #include <linux/slab.h>
22 #include <linux/net.h>
23 #include <linux/if_packet.h>
24 #include <linux/if_arp.h>
25 #include <linux/if_tun.h>
26 #include <linux/if_macvlan.h>
32 static int experimental_zcopytx
;
33 module_param(experimental_zcopytx
, int, 0444);
34 MODULE_PARM_DESC(experimental_zcopytx
, "Enable Experimental Zero Copy TX");
36 /* Max number of bytes transferred before requeueing the job.
37 * Using this limit prevents one virtqueue from starving others. */
38 #define VHOST_NET_WEIGHT 0x80000
40 /* MAX number of TX used buffers for outstanding zerocopy */
41 #define VHOST_MAX_PEND 128
42 #define VHOST_GOODCOPY_LEN 256
50 enum vhost_net_poll_state
{
51 VHOST_NET_POLL_DISABLED
= 0,
52 VHOST_NET_POLL_STARTED
= 1,
53 VHOST_NET_POLL_STOPPED
= 2,
58 struct vhost_virtqueue vqs
[VHOST_NET_VQ_MAX
];
59 struct vhost_poll poll
[VHOST_NET_VQ_MAX
];
60 /* Tells us whether we are polling a socket for TX.
61 * We only do this when socket buffer fills up.
62 * Protected by tx vq lock. */
63 enum vhost_net_poll_state tx_poll_state
;
66 static bool vhost_sock_zcopy(struct socket
*sock
)
68 return unlikely(experimental_zcopytx
) &&
69 sock_flag(sock
->sk
, SOCK_ZEROCOPY
);
72 /* Pop first len bytes from iovec. Return number of segments used. */
73 static int move_iovec_hdr(struct iovec
*from
, struct iovec
*to
,
74 size_t len
, int iov_count
)
79 while (len
&& seg
< iov_count
) {
80 size
= min(from
->iov_len
, len
);
81 to
->iov_base
= from
->iov_base
;
83 from
->iov_len
-= size
;
84 from
->iov_base
+= size
;
92 /* Copy iovec entries for len bytes from iovec. */
93 static void copy_iovec_hdr(const struct iovec
*from
, struct iovec
*to
,
94 size_t len
, int iovcount
)
99 while (len
&& seg
< iovcount
) {
100 size
= min(from
->iov_len
, len
);
101 to
->iov_base
= from
->iov_base
;
110 /* Caller must have TX VQ lock */
111 static void tx_poll_stop(struct vhost_net
*net
)
113 if (likely(net
->tx_poll_state
!= VHOST_NET_POLL_STARTED
))
115 vhost_poll_stop(net
->poll
+ VHOST_NET_VQ_TX
);
116 net
->tx_poll_state
= VHOST_NET_POLL_STOPPED
;
119 /* Caller must have TX VQ lock */
120 static void tx_poll_start(struct vhost_net
*net
, struct socket
*sock
)
122 if (unlikely(net
->tx_poll_state
!= VHOST_NET_POLL_STOPPED
))
124 vhost_poll_start(net
->poll
+ VHOST_NET_VQ_TX
, sock
->file
);
125 net
->tx_poll_state
= VHOST_NET_POLL_STARTED
;
128 /* Expects to be always run from workqueue - which acts as
129 * read-size critical section for our kind of RCU. */
130 static void handle_tx(struct vhost_net
*net
)
132 struct vhost_virtqueue
*vq
= &net
->dev
.vqs
[VHOST_NET_VQ_TX
];
135 struct msghdr msg
= {
141 .msg_flags
= MSG_DONTWAIT
,
143 size_t len
, total_len
= 0;
147 struct vhost_ubuf_ref
*uninitialized_var(ubufs
);
150 /* TODO: check that we are running from vhost_worker? */
151 sock
= rcu_dereference_check(vq
->private_data
, 1);
155 wmem
= atomic_read(&sock
->sk
->sk_wmem_alloc
);
156 if (wmem
>= sock
->sk
->sk_sndbuf
) {
157 mutex_lock(&vq
->mutex
);
158 tx_poll_start(net
, sock
);
159 mutex_unlock(&vq
->mutex
);
163 mutex_lock(&vq
->mutex
);
164 vhost_disable_notify(&net
->dev
, vq
);
166 if (wmem
< sock
->sk
->sk_sndbuf
/ 2)
168 hdr_size
= vq
->vhost_hlen
;
172 /* Release DMAs done buffers first */
174 vhost_zerocopy_signal_used(vq
);
176 head
= vhost_get_vq_desc(&net
->dev
, vq
, vq
->iov
,
180 /* On error, stop handling until the next kick. */
181 if (unlikely(head
< 0))
183 /* Nothing new? Wait for eventfd to tell us they refilled. */
184 if (head
== vq
->num
) {
187 wmem
= atomic_read(&sock
->sk
->sk_wmem_alloc
);
188 if (wmem
>= sock
->sk
->sk_sndbuf
* 3 / 4) {
189 tx_poll_start(net
, sock
);
190 set_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
);
193 /* If more outstanding DMAs, queue the work.
194 * Handle upend_idx wrap around
196 num_pends
= likely(vq
->upend_idx
>= vq
->done_idx
) ?
197 (vq
->upend_idx
- vq
->done_idx
) :
198 (vq
->upend_idx
+ UIO_MAXIOV
- vq
->done_idx
);
199 if (unlikely(num_pends
> VHOST_MAX_PEND
)) {
200 tx_poll_start(net
, sock
);
201 set_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
);
204 if (unlikely(vhost_enable_notify(&net
->dev
, vq
))) {
205 vhost_disable_notify(&net
->dev
, vq
);
211 vq_err(vq
, "Unexpected descriptor format for TX: "
212 "out %d, int %d\n", out
, in
);
215 /* Skip header. TODO: support TSO. */
216 s
= move_iovec_hdr(vq
->iov
, vq
->hdr
, hdr_size
, out
);
217 msg
.msg_iovlen
= out
;
218 len
= iov_length(vq
->iov
, out
);
221 vq_err(vq
, "Unexpected header len for TX: "
222 "%zd expected %zd\n",
223 iov_length(vq
->hdr
, s
), hdr_size
);
226 /* use msg_control to pass vhost zerocopy ubuf info to skb */
228 vq
->heads
[vq
->upend_idx
].id
= head
;
229 if (len
< VHOST_GOODCOPY_LEN
) {
230 /* copy don't need to wait for DMA done */
231 vq
->heads
[vq
->upend_idx
].len
=
233 msg
.msg_control
= NULL
;
234 msg
.msg_controllen
= 0;
237 struct ubuf_info
*ubuf
= &vq
->ubuf_info
[head
];
239 vq
->heads
[vq
->upend_idx
].len
= len
;
240 ubuf
->callback
= vhost_zerocopy_callback
;
241 ubuf
->ctx
= vq
->ubufs
;
242 ubuf
->desc
= vq
->upend_idx
;
243 msg
.msg_control
= ubuf
;
244 msg
.msg_controllen
= sizeof(ubuf
);
246 kref_get(&ubufs
->kref
);
248 vq
->upend_idx
= (vq
->upend_idx
+ 1) % UIO_MAXIOV
;
250 /* TODO: Check specific error and bomb out unless ENOBUFS? */
251 err
= sock
->ops
->sendmsg(NULL
, sock
, &msg
, len
);
252 if (unlikely(err
< 0)) {
255 vhost_ubuf_put(ubufs
);
256 vq
->upend_idx
= ((unsigned)vq
->upend_idx
- 1) %
259 vhost_discard_vq_desc(vq
, 1);
260 if (err
== -EAGAIN
|| err
== -ENOBUFS
)
261 tx_poll_start(net
, sock
);
265 pr_debug("Truncated TX packet: "
266 " len %d != %zd\n", err
, len
);
268 vhost_add_used_and_signal(&net
->dev
, vq
, head
, 0);
270 vhost_zerocopy_signal_used(vq
);
272 if (unlikely(total_len
>= VHOST_NET_WEIGHT
)) {
273 vhost_poll_queue(&vq
->poll
);
278 mutex_unlock(&vq
->mutex
);
281 static int peek_head_len(struct sock
*sk
)
283 struct sk_buff
*head
;
287 spin_lock_irqsave(&sk
->sk_receive_queue
.lock
, flags
);
288 head
= skb_peek(&sk
->sk_receive_queue
);
291 spin_unlock_irqrestore(&sk
->sk_receive_queue
.lock
, flags
);
295 /* This is a multi-buffer version of vhost_get_desc, that works if
296 * vq has read descriptors only.
297 * @vq - the relevant virtqueue
298 * @datalen - data length we'll be reading
299 * @iovcount - returned count of io vectors we fill
301 * @log_num - log offset
302 * @quota - headcount quota, 1 for big buffer
303 * returns number of buffer heads allocated, negative on error
305 static int get_rx_bufs(struct vhost_virtqueue
*vq
,
306 struct vring_used_elem
*heads
,
309 struct vhost_log
*log
,
313 unsigned int out
, in
;
319 while (datalen
> 0 && headcount
< quota
) {
320 if (unlikely(seg
>= UIO_MAXIOV
)) {
324 d
= vhost_get_vq_desc(vq
->dev
, vq
, vq
->iov
+ seg
,
325 ARRAY_SIZE(vq
->iov
) - seg
, &out
,
331 if (unlikely(out
|| in
<= 0)) {
332 vq_err(vq
, "unexpected descriptor format for RX: "
333 "out %d, in %d\n", out
, in
);
341 heads
[headcount
].id
= d
;
342 heads
[headcount
].len
= iov_length(vq
->iov
+ seg
, in
);
343 datalen
-= heads
[headcount
].len
;
347 heads
[headcount
- 1].len
+= datalen
;
353 vhost_discard_vq_desc(vq
, headcount
);
357 /* Expects to be always run from workqueue - which acts as
358 * read-size critical section for our kind of RCU. */
359 static void handle_rx(struct vhost_net
*net
)
361 struct vhost_virtqueue
*vq
= &net
->dev
.vqs
[VHOST_NET_VQ_RX
];
362 unsigned uninitialized_var(in
), log
;
363 struct vhost_log
*vq_log
;
364 struct msghdr msg
= {
367 .msg_control
= NULL
, /* FIXME: get and handle RX aux data. */
370 .msg_flags
= MSG_DONTWAIT
,
372 struct virtio_net_hdr_mrg_rxbuf hdr
= {
374 .hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
376 size_t total_len
= 0;
377 int err
, headcount
, mergeable
;
378 size_t vhost_hlen
, sock_hlen
;
379 size_t vhost_len
, sock_len
;
380 /* TODO: check that we are running from vhost_worker? */
381 struct socket
*sock
= rcu_dereference_check(vq
->private_data
, 1);
386 mutex_lock(&vq
->mutex
);
387 vhost_disable_notify(&net
->dev
, vq
);
388 vhost_hlen
= vq
->vhost_hlen
;
389 sock_hlen
= vq
->sock_hlen
;
391 vq_log
= unlikely(vhost_has_feature(&net
->dev
, VHOST_F_LOG_ALL
)) ?
393 mergeable
= vhost_has_feature(&net
->dev
, VIRTIO_NET_F_MRG_RXBUF
);
395 while ((sock_len
= peek_head_len(sock
->sk
))) {
396 sock_len
+= sock_hlen
;
397 vhost_len
= sock_len
+ vhost_hlen
;
398 headcount
= get_rx_bufs(vq
, vq
->heads
, vhost_len
,
400 likely(mergeable
) ? UIO_MAXIOV
: 1);
401 /* On error, stop handling until the next kick. */
402 if (unlikely(headcount
< 0))
404 /* OK, now we need to know about added descriptors. */
406 if (unlikely(vhost_enable_notify(&net
->dev
, vq
))) {
407 /* They have slipped one in as we were
408 * doing that: check again. */
409 vhost_disable_notify(&net
->dev
, vq
);
412 /* Nothing new? Wait for eventfd to tell us
416 /* We don't need to be notified again. */
417 if (unlikely((vhost_hlen
)))
418 /* Skip header. TODO: support TSO. */
419 move_iovec_hdr(vq
->iov
, vq
->hdr
, vhost_hlen
, in
);
421 /* Copy the header for use in VIRTIO_NET_F_MRG_RXBUF:
422 * needed because recvmsg can modify msg_iov. */
423 copy_iovec_hdr(vq
->iov
, vq
->hdr
, sock_hlen
, in
);
425 err
= sock
->ops
->recvmsg(NULL
, sock
, &msg
,
426 sock_len
, MSG_DONTWAIT
| MSG_TRUNC
);
427 /* Userspace might have consumed the packet meanwhile:
428 * it's not supposed to do this usually, but might be hard
429 * to prevent. Discard data we got (if any) and keep going. */
430 if (unlikely(err
!= sock_len
)) {
431 pr_debug("Discarded rx packet: "
432 " len %d, expected %zd\n", err
, sock_len
);
433 vhost_discard_vq_desc(vq
, headcount
);
436 if (unlikely(vhost_hlen
) &&
437 memcpy_toiovecend(vq
->hdr
, (unsigned char *)&hdr
, 0,
439 vq_err(vq
, "Unable to write vnet_hdr at addr %p\n",
443 /* TODO: Should check and handle checksum. */
444 if (likely(mergeable
) &&
445 memcpy_toiovecend(vq
->hdr
, (unsigned char *)&headcount
,
446 offsetof(typeof(hdr
), num_buffers
),
447 sizeof hdr
.num_buffers
)) {
448 vq_err(vq
, "Failed num_buffers write");
449 vhost_discard_vq_desc(vq
, headcount
);
452 vhost_add_used_and_signal_n(&net
->dev
, vq
, vq
->heads
,
454 if (unlikely(vq_log
))
455 vhost_log_write(vq
, vq_log
, log
, vhost_len
);
456 total_len
+= vhost_len
;
457 if (unlikely(total_len
>= VHOST_NET_WEIGHT
)) {
458 vhost_poll_queue(&vq
->poll
);
463 mutex_unlock(&vq
->mutex
);
466 static void handle_tx_kick(struct vhost_work
*work
)
468 struct vhost_virtqueue
*vq
= container_of(work
, struct vhost_virtqueue
,
470 struct vhost_net
*net
= container_of(vq
->dev
, struct vhost_net
, dev
);
475 static void handle_rx_kick(struct vhost_work
*work
)
477 struct vhost_virtqueue
*vq
= container_of(work
, struct vhost_virtqueue
,
479 struct vhost_net
*net
= container_of(vq
->dev
, struct vhost_net
, dev
);
484 static void handle_tx_net(struct vhost_work
*work
)
486 struct vhost_net
*net
= container_of(work
, struct vhost_net
,
487 poll
[VHOST_NET_VQ_TX
].work
);
491 static void handle_rx_net(struct vhost_work
*work
)
493 struct vhost_net
*net
= container_of(work
, struct vhost_net
,
494 poll
[VHOST_NET_VQ_RX
].work
);
498 static int vhost_net_open(struct inode
*inode
, struct file
*f
)
500 struct vhost_net
*n
= kmalloc(sizeof *n
, GFP_KERNEL
);
501 struct vhost_dev
*dev
;
508 n
->vqs
[VHOST_NET_VQ_TX
].handle_kick
= handle_tx_kick
;
509 n
->vqs
[VHOST_NET_VQ_RX
].handle_kick
= handle_rx_kick
;
510 r
= vhost_dev_init(dev
, n
->vqs
, VHOST_NET_VQ_MAX
);
516 vhost_poll_init(n
->poll
+ VHOST_NET_VQ_TX
, handle_tx_net
, POLLOUT
, dev
);
517 vhost_poll_init(n
->poll
+ VHOST_NET_VQ_RX
, handle_rx_net
, POLLIN
, dev
);
518 n
->tx_poll_state
= VHOST_NET_POLL_DISABLED
;
525 static void vhost_net_disable_vq(struct vhost_net
*n
,
526 struct vhost_virtqueue
*vq
)
528 if (!vq
->private_data
)
530 if (vq
== n
->vqs
+ VHOST_NET_VQ_TX
) {
532 n
->tx_poll_state
= VHOST_NET_POLL_DISABLED
;
534 vhost_poll_stop(n
->poll
+ VHOST_NET_VQ_RX
);
537 static void vhost_net_enable_vq(struct vhost_net
*n
,
538 struct vhost_virtqueue
*vq
)
542 sock
= rcu_dereference_protected(vq
->private_data
,
543 lockdep_is_held(&vq
->mutex
));
546 if (vq
== n
->vqs
+ VHOST_NET_VQ_TX
) {
547 n
->tx_poll_state
= VHOST_NET_POLL_STOPPED
;
548 tx_poll_start(n
, sock
);
550 vhost_poll_start(n
->poll
+ VHOST_NET_VQ_RX
, sock
->file
);
553 static struct socket
*vhost_net_stop_vq(struct vhost_net
*n
,
554 struct vhost_virtqueue
*vq
)
558 mutex_lock(&vq
->mutex
);
559 sock
= rcu_dereference_protected(vq
->private_data
,
560 lockdep_is_held(&vq
->mutex
));
561 vhost_net_disable_vq(n
, vq
);
562 rcu_assign_pointer(vq
->private_data
, NULL
);
563 mutex_unlock(&vq
->mutex
);
567 static void vhost_net_stop(struct vhost_net
*n
, struct socket
**tx_sock
,
568 struct socket
**rx_sock
)
570 *tx_sock
= vhost_net_stop_vq(n
, n
->vqs
+ VHOST_NET_VQ_TX
);
571 *rx_sock
= vhost_net_stop_vq(n
, n
->vqs
+ VHOST_NET_VQ_RX
);
574 static void vhost_net_flush_vq(struct vhost_net
*n
, int index
)
576 vhost_poll_flush(n
->poll
+ index
);
577 vhost_poll_flush(&n
->dev
.vqs
[index
].poll
);
580 static void vhost_net_flush(struct vhost_net
*n
)
582 vhost_net_flush_vq(n
, VHOST_NET_VQ_TX
);
583 vhost_net_flush_vq(n
, VHOST_NET_VQ_RX
);
586 static int vhost_net_release(struct inode
*inode
, struct file
*f
)
588 struct vhost_net
*n
= f
->private_data
;
589 struct socket
*tx_sock
;
590 struct socket
*rx_sock
;
592 vhost_net_stop(n
, &tx_sock
, &rx_sock
);
594 vhost_dev_cleanup(&n
->dev
, false);
599 /* We do an extra flush before freeing memory,
600 * since jobs can re-queue themselves. */
606 static struct socket
*get_raw_socket(int fd
)
609 struct sockaddr_ll sa
;
610 char buf
[MAX_ADDR_LEN
];
612 int uaddr_len
= sizeof uaddr
, r
;
613 struct socket
*sock
= sockfd_lookup(fd
, &r
);
616 return ERR_PTR(-ENOTSOCK
);
618 /* Parameter checking */
619 if (sock
->sk
->sk_type
!= SOCK_RAW
) {
620 r
= -ESOCKTNOSUPPORT
;
624 r
= sock
->ops
->getname(sock
, (struct sockaddr
*)&uaddr
.sa
,
629 if (uaddr
.sa
.sll_family
!= AF_PACKET
) {
639 static struct socket
*get_tap_socket(int fd
)
641 struct file
*file
= fget(fd
);
645 return ERR_PTR(-EBADF
);
646 sock
= tun_get_socket(file
);
649 sock
= macvtap_get_socket(file
);
655 static struct socket
*get_socket(int fd
)
659 /* special case to disable backend */
662 sock
= get_raw_socket(fd
);
665 sock
= get_tap_socket(fd
);
668 return ERR_PTR(-ENOTSOCK
);
671 static long vhost_net_set_backend(struct vhost_net
*n
, unsigned index
, int fd
)
673 struct socket
*sock
, *oldsock
;
674 struct vhost_virtqueue
*vq
;
675 struct vhost_ubuf_ref
*ubufs
, *oldubufs
= NULL
;
678 mutex_lock(&n
->dev
.mutex
);
679 r
= vhost_dev_check_owner(&n
->dev
);
683 if (index
>= VHOST_NET_VQ_MAX
) {
688 mutex_lock(&vq
->mutex
);
690 /* Verify that ring has been setup correctly. */
691 if (!vhost_vq_access_ok(vq
)) {
695 sock
= get_socket(fd
);
701 /* start polling new socket */
702 oldsock
= rcu_dereference_protected(vq
->private_data
,
703 lockdep_is_held(&vq
->mutex
));
704 if (sock
!= oldsock
) {
705 ubufs
= vhost_ubuf_alloc(vq
, sock
&& vhost_sock_zcopy(sock
));
710 oldubufs
= vq
->ubufs
;
712 vhost_net_disable_vq(n
, vq
);
713 rcu_assign_pointer(vq
->private_data
, sock
);
714 vhost_net_enable_vq(n
, vq
);
716 r
= vhost_init_used(vq
);
721 mutex_unlock(&vq
->mutex
);
724 vhost_ubuf_put_and_wait(oldubufs
);
725 mutex_lock(&vq
->mutex
);
726 vhost_zerocopy_signal_used(vq
);
727 mutex_unlock(&vq
->mutex
);
731 vhost_net_flush_vq(n
, index
);
735 mutex_unlock(&n
->dev
.mutex
);
741 mutex_unlock(&vq
->mutex
);
743 mutex_unlock(&n
->dev
.mutex
);
747 static long vhost_net_reset_owner(struct vhost_net
*n
)
749 struct socket
*tx_sock
= NULL
;
750 struct socket
*rx_sock
= NULL
;
753 mutex_lock(&n
->dev
.mutex
);
754 err
= vhost_dev_check_owner(&n
->dev
);
757 vhost_net_stop(n
, &tx_sock
, &rx_sock
);
759 err
= vhost_dev_reset_owner(&n
->dev
);
761 mutex_unlock(&n
->dev
.mutex
);
769 static int vhost_net_set_features(struct vhost_net
*n
, u64 features
)
771 size_t vhost_hlen
, sock_hlen
, hdr_len
;
774 hdr_len
= (features
& (1 << VIRTIO_NET_F_MRG_RXBUF
)) ?
775 sizeof(struct virtio_net_hdr_mrg_rxbuf
) :
776 sizeof(struct virtio_net_hdr
);
777 if (features
& (1 << VHOST_NET_F_VIRTIO_NET_HDR
)) {
778 /* vhost provides vnet_hdr */
779 vhost_hlen
= hdr_len
;
782 /* socket provides vnet_hdr */
786 mutex_lock(&n
->dev
.mutex
);
787 if ((features
& (1 << VHOST_F_LOG_ALL
)) &&
788 !vhost_log_access_ok(&n
->dev
)) {
789 mutex_unlock(&n
->dev
.mutex
);
792 n
->dev
.acked_features
= features
;
794 for (i
= 0; i
< VHOST_NET_VQ_MAX
; ++i
) {
795 mutex_lock(&n
->vqs
[i
].mutex
);
796 n
->vqs
[i
].vhost_hlen
= vhost_hlen
;
797 n
->vqs
[i
].sock_hlen
= sock_hlen
;
798 mutex_unlock(&n
->vqs
[i
].mutex
);
801 mutex_unlock(&n
->dev
.mutex
);
805 static long vhost_net_ioctl(struct file
*f
, unsigned int ioctl
,
808 struct vhost_net
*n
= f
->private_data
;
809 void __user
*argp
= (void __user
*)arg
;
810 u64 __user
*featurep
= argp
;
811 struct vhost_vring_file backend
;
816 case VHOST_NET_SET_BACKEND
:
817 if (copy_from_user(&backend
, argp
, sizeof backend
))
819 return vhost_net_set_backend(n
, backend
.index
, backend
.fd
);
820 case VHOST_GET_FEATURES
:
821 features
= VHOST_FEATURES
;
822 if (copy_to_user(featurep
, &features
, sizeof features
))
825 case VHOST_SET_FEATURES
:
826 if (copy_from_user(&features
, featurep
, sizeof features
))
828 if (features
& ~VHOST_FEATURES
)
830 return vhost_net_set_features(n
, features
);
831 case VHOST_RESET_OWNER
:
832 return vhost_net_reset_owner(n
);
834 mutex_lock(&n
->dev
.mutex
);
835 r
= vhost_dev_ioctl(&n
->dev
, ioctl
, arg
);
837 mutex_unlock(&n
->dev
.mutex
);
843 static long vhost_net_compat_ioctl(struct file
*f
, unsigned int ioctl
,
846 return vhost_net_ioctl(f
, ioctl
, (unsigned long)compat_ptr(arg
));
850 static const struct file_operations vhost_net_fops
= {
851 .owner
= THIS_MODULE
,
852 .release
= vhost_net_release
,
853 .unlocked_ioctl
= vhost_net_ioctl
,
855 .compat_ioctl
= vhost_net_compat_ioctl
,
857 .open
= vhost_net_open
,
858 .llseek
= noop_llseek
,
861 static struct miscdevice vhost_net_misc
= {
862 .minor
= VHOST_NET_MINOR
,
864 .fops
= &vhost_net_fops
,
867 static int vhost_net_init(void)
869 if (experimental_zcopytx
)
870 vhost_enable_zcopy(VHOST_NET_VQ_TX
);
871 return misc_register(&vhost_net_misc
);
873 module_init(vhost_net_init
);
875 static void vhost_net_exit(void)
877 misc_deregister(&vhost_net_misc
);
879 module_exit(vhost_net_exit
);
881 MODULE_VERSION("0.0.1");
882 MODULE_LICENSE("GPL v2");
883 MODULE_AUTHOR("Michael S. Tsirkin");
884 MODULE_DESCRIPTION("Host kernel accelerator for virtio net");
885 MODULE_ALIAS_MISCDEV(VHOST_NET_MINOR
);
886 MODULE_ALIAS("devname:vhost-net");