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>
27 #include <linux/if_vlan.h>
33 static int experimental_zcopytx;
34 module_param(experimental_zcopytx, int, 0444);
35 MODULE_PARM_DESC(experimental_zcopytx, "Enable Experimental Zero Copy TX");
37 /* Max number of bytes transferred before requeueing the job.
38 * Using this limit prevents one virtqueue from starving others. */
39 #define VHOST_NET_WEIGHT 0x80000
41 /* MAX number of TX used buffers for outstanding zerocopy */
42 #define VHOST_MAX_PEND 128
43 #define VHOST_GOODCOPY_LEN 256
46 * For transmit, used buffer len is unused; we override it to track buffer
47 * status internally; used for zerocopy tx only.
49 /* Lower device DMA failed */
50 #define VHOST_DMA_FAILED_LEN 3
51 /* Lower device DMA done */
52 #define VHOST_DMA_DONE_LEN 2
53 /* Lower device DMA in progress */
54 #define VHOST_DMA_IN_PROGRESS 1
56 #define VHOST_DMA_CLEAR_LEN 0
58 #define VHOST_DMA_IS_DONE(len) ((len) >= VHOST_DMA_DONE_LEN)
66 enum vhost_net_poll_state {
67 VHOST_NET_POLL_DISABLED = 0,
68 VHOST_NET_POLL_STARTED = 1,
69 VHOST_NET_POLL_STOPPED = 2,
74 struct vhost_virtqueue vqs[VHOST_NET_VQ_MAX];
75 struct vhost_poll poll[VHOST_NET_VQ_MAX];
76 /* Tells us whether we are polling a socket for TX.
77 * We only do this when socket buffer fills up.
78 * Protected by tx vq lock. */
79 enum vhost_net_poll_state tx_poll_state;
80 /* Number of TX recently submitted.
81 * Protected by tx vq lock. */
83 /* Number of times zerocopy TX recently failed.
84 * Protected by tx vq lock. */
85 unsigned tx_zcopy_err;
88 static void vhost_net_tx_packet(struct vhost_net *net)
91 if (net->tx_packets < 1024)
94 net->tx_zcopy_err = 0;
97 static void vhost_net_tx_err(struct vhost_net *net)
102 static bool vhost_net_tx_select_zcopy(struct vhost_net *net)
104 return net->tx_packets / 64 >= net->tx_zcopy_err;
107 static bool vhost_sock_zcopy(struct socket *sock)
109 return unlikely(experimental_zcopytx) &&
110 sock_flag(sock->sk, SOCK_ZEROCOPY);
113 /* Pop first len bytes from iovec. Return number of segments used. */
114 static int move_iovec_hdr(struct iovec *from, struct iovec *to,
115 size_t len, int iov_count)
120 while (len && seg < iov_count) {
121 size = min(from->iov_len, len);
122 to->iov_base = from->iov_base;
124 from->iov_len -= size;
125 from->iov_base += size;
133 /* Copy iovec entries for len bytes from iovec. */
134 static void copy_iovec_hdr(const struct iovec *from, struct iovec *to,
135 size_t len, int iovcount)
140 while (len && seg < iovcount) {
141 size = min(from->iov_len, len);
142 to->iov_base = from->iov_base;
151 /* Caller must have TX VQ lock */
152 static void tx_poll_stop(struct vhost_net *net)
154 if (likely(net->tx_poll_state != VHOST_NET_POLL_STARTED))
156 vhost_poll_stop(net->poll + VHOST_NET_VQ_TX);
157 net->tx_poll_state = VHOST_NET_POLL_STOPPED;
160 /* Caller must have TX VQ lock */
161 static void tx_poll_start(struct vhost_net *net, struct socket *sock)
163 if (unlikely(net->tx_poll_state != VHOST_NET_POLL_STOPPED))
165 vhost_poll_start(net->poll + VHOST_NET_VQ_TX, sock->file);
166 net->tx_poll_state = VHOST_NET_POLL_STARTED;
169 /* In case of DMA done not in order in lower device driver for some reason.
170 * upend_idx is used to track end of used idx, done_idx is used to track head
171 * of used idx. Once lower device DMA done contiguously, we will signal KVM
174 static int vhost_zerocopy_signal_used(struct vhost_net *net,
175 struct vhost_virtqueue *vq)
180 for (i = vq->done_idx; i != vq->upend_idx; i = (i + 1) % UIO_MAXIOV) {
181 if (vq->heads[i].len == VHOST_DMA_FAILED_LEN)
182 vhost_net_tx_err(net);
183 if (VHOST_DMA_IS_DONE(vq->heads[i].len)) {
184 vq->heads[i].len = VHOST_DMA_CLEAR_LEN;
185 vhost_add_used_and_signal(vq->dev, vq,
196 static void vhost_zerocopy_callback(struct ubuf_info *ubuf, bool success)
198 struct vhost_ubuf_ref *ubufs = ubuf->ctx;
199 struct vhost_virtqueue *vq = ubufs->vq;
201 vhost_poll_queue(&vq->poll);
202 /* set len to mark this desc buffers done DMA */
203 vq->heads[ubuf->desc].len = success ?
204 VHOST_DMA_DONE_LEN : VHOST_DMA_FAILED_LEN;
205 vhost_ubuf_put(ubufs);
208 /* Expects to be always run from workqueue - which acts as
209 * read-size critical section for our kind of RCU. */
210 static void handle_tx(struct vhost_net *net)
212 struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_TX];
215 struct msghdr msg = {
221 .msg_flags = MSG_DONTWAIT,
223 size_t len, total_len = 0;
227 struct vhost_ubuf_ref *uninitialized_var(ubufs);
230 /* TODO: check that we are running from vhost_worker? */
231 sock = rcu_dereference_check(vq->private_data, 1);
235 wmem = atomic_read(&sock->sk->sk_wmem_alloc);
236 if (wmem >= sock->sk->sk_sndbuf) {
237 mutex_lock(&vq->mutex);
238 tx_poll_start(net, sock);
239 mutex_unlock(&vq->mutex);
243 mutex_lock(&vq->mutex);
244 vhost_disable_notify(&net->dev, vq);
246 if (wmem < sock->sk->sk_sndbuf / 2)
248 hdr_size = vq->vhost_hlen;
252 /* Release DMAs done buffers first */
254 vhost_zerocopy_signal_used(net, vq);
256 head = vhost_get_vq_desc(&net->dev, vq, vq->iov,
260 /* On error, stop handling until the next kick. */
261 if (unlikely(head < 0))
263 /* Nothing new? Wait for eventfd to tell us they refilled. */
264 if (head == vq->num) {
267 wmem = atomic_read(&sock->sk->sk_wmem_alloc);
268 if (wmem >= sock->sk->sk_sndbuf * 3 / 4) {
269 tx_poll_start(net, sock);
270 set_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
273 /* If more outstanding DMAs, queue the work.
274 * Handle upend_idx wrap around
276 num_pends = likely(vq->upend_idx >= vq->done_idx) ?
277 (vq->upend_idx - vq->done_idx) :
278 (vq->upend_idx + UIO_MAXIOV - vq->done_idx);
279 if (unlikely(num_pends > VHOST_MAX_PEND)) {
280 tx_poll_start(net, sock);
281 set_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
284 if (unlikely(vhost_enable_notify(&net->dev, vq))) {
285 vhost_disable_notify(&net->dev, vq);
291 vq_err(vq, "Unexpected descriptor format for TX: "
292 "out %d, int %d\n", out, in);
295 /* Skip header. TODO: support TSO. */
296 s = move_iovec_hdr(vq->iov, vq->hdr, hdr_size, out);
297 msg.msg_iovlen = out;
298 len = iov_length(vq->iov, out);
301 vq_err(vq, "Unexpected header len for TX: "
302 "%zd expected %zd\n",
303 iov_length(vq->hdr, s), hdr_size);
306 /* use msg_control to pass vhost zerocopy ubuf info to skb */
308 vq->heads[vq->upend_idx].id = head;
309 if (!vhost_net_tx_select_zcopy(net) ||
310 len < VHOST_GOODCOPY_LEN) {
311 /* copy don't need to wait for DMA done */
312 vq->heads[vq->upend_idx].len =
314 msg.msg_control = NULL;
315 msg.msg_controllen = 0;
318 struct ubuf_info *ubuf = &vq->ubuf_info[head];
320 vq->heads[vq->upend_idx].len =
321 VHOST_DMA_IN_PROGRESS;
322 ubuf->callback = vhost_zerocopy_callback;
323 ubuf->ctx = vq->ubufs;
324 ubuf->desc = vq->upend_idx;
325 msg.msg_control = ubuf;
326 msg.msg_controllen = sizeof(ubuf);
328 kref_get(&ubufs->kref);
330 vq->upend_idx = (vq->upend_idx + 1) % UIO_MAXIOV;
332 /* TODO: Check specific error and bomb out unless ENOBUFS? */
333 err = sock->ops->sendmsg(NULL, sock, &msg, len);
334 if (unlikely(err < 0)) {
337 vhost_ubuf_put(ubufs);
338 vq->upend_idx = ((unsigned)vq->upend_idx - 1) %
341 vhost_discard_vq_desc(vq, 1);
342 if (err == -EAGAIN || err == -ENOBUFS)
343 tx_poll_start(net, sock);
347 pr_debug("Truncated TX packet: "
348 " len %d != %zd\n", err, len);
350 vhost_add_used_and_signal(&net->dev, vq, head, 0);
352 vhost_zerocopy_signal_used(net, vq);
354 vhost_net_tx_packet(net);
355 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
356 vhost_poll_queue(&vq->poll);
361 mutex_unlock(&vq->mutex);
364 static int peek_head_len(struct sock *sk)
366 struct sk_buff *head;
370 spin_lock_irqsave(&sk->sk_receive_queue.lock, flags);
371 head = skb_peek(&sk->sk_receive_queue);
374 if (vlan_tx_tag_present(head))
378 spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags);
382 /* This is a multi-buffer version of vhost_get_desc, that works if
383 * vq has read descriptors only.
384 * @vq - the relevant virtqueue
385 * @datalen - data length we'll be reading
386 * @iovcount - returned count of io vectors we fill
388 * @log_num - log offset
389 * @quota - headcount quota, 1 for big buffer
390 * returns number of buffer heads allocated, negative on error
392 static int get_rx_bufs(struct vhost_virtqueue *vq,
393 struct vring_used_elem *heads,
396 struct vhost_log *log,
400 unsigned int out, in;
406 while (datalen > 0 && headcount < quota) {
407 if (unlikely(seg >= UIO_MAXIOV)) {
411 d = vhost_get_vq_desc(vq->dev, vq, vq->iov + seg,
412 ARRAY_SIZE(vq->iov) - seg, &out,
418 if (unlikely(out || in <= 0)) {
419 vq_err(vq, "unexpected descriptor format for RX: "
420 "out %d, in %d\n", out, in);
428 heads[headcount].id = d;
429 heads[headcount].len = iov_length(vq->iov + seg, in);
430 datalen -= heads[headcount].len;
434 heads[headcount - 1].len += datalen;
440 vhost_discard_vq_desc(vq, headcount);
444 /* Expects to be always run from workqueue - which acts as
445 * read-size critical section for our kind of RCU. */
446 static void handle_rx(struct vhost_net *net)
448 struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_RX];
449 unsigned uninitialized_var(in), log;
450 struct vhost_log *vq_log;
451 struct msghdr msg = {
454 .msg_control = NULL, /* FIXME: get and handle RX aux data. */
457 .msg_flags = MSG_DONTWAIT,
459 struct virtio_net_hdr_mrg_rxbuf hdr = {
461 .hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE
463 size_t total_len = 0;
464 int err, headcount, mergeable;
465 size_t vhost_hlen, sock_hlen;
466 size_t vhost_len, sock_len;
467 /* TODO: check that we are running from vhost_worker? */
468 struct socket *sock = rcu_dereference_check(vq->private_data, 1);
473 mutex_lock(&vq->mutex);
474 vhost_disable_notify(&net->dev, vq);
475 vhost_hlen = vq->vhost_hlen;
476 sock_hlen = vq->sock_hlen;
478 vq_log = unlikely(vhost_has_feature(&net->dev, VHOST_F_LOG_ALL)) ?
480 mergeable = vhost_has_feature(&net->dev, VIRTIO_NET_F_MRG_RXBUF);
482 while ((sock_len = peek_head_len(sock->sk))) {
483 sock_len += sock_hlen;
484 vhost_len = sock_len + vhost_hlen;
485 headcount = get_rx_bufs(vq, vq->heads, vhost_len,
487 likely(mergeable) ? UIO_MAXIOV : 1);
488 /* On error, stop handling until the next kick. */
489 if (unlikely(headcount < 0))
491 /* OK, now we need to know about added descriptors. */
493 if (unlikely(vhost_enable_notify(&net->dev, vq))) {
494 /* They have slipped one in as we were
495 * doing that: check again. */
496 vhost_disable_notify(&net->dev, vq);
499 /* Nothing new? Wait for eventfd to tell us
503 /* We don't need to be notified again. */
504 if (unlikely((vhost_hlen)))
505 /* Skip header. TODO: support TSO. */
506 move_iovec_hdr(vq->iov, vq->hdr, vhost_hlen, in);
508 /* Copy the header for use in VIRTIO_NET_F_MRG_RXBUF:
509 * needed because recvmsg can modify msg_iov. */
510 copy_iovec_hdr(vq->iov, vq->hdr, sock_hlen, in);
512 err = sock->ops->recvmsg(NULL, sock, &msg,
513 sock_len, MSG_DONTWAIT | MSG_TRUNC);
514 /* Userspace might have consumed the packet meanwhile:
515 * it's not supposed to do this usually, but might be hard
516 * to prevent. Discard data we got (if any) and keep going. */
517 if (unlikely(err != sock_len)) {
518 pr_debug("Discarded rx packet: "
519 " len %d, expected %zd\n", err, sock_len);
520 vhost_discard_vq_desc(vq, headcount);
523 if (unlikely(vhost_hlen) &&
524 memcpy_toiovecend(vq->hdr, (unsigned char *)&hdr, 0,
526 vq_err(vq, "Unable to write vnet_hdr at addr %p\n",
530 /* TODO: Should check and handle checksum. */
531 if (likely(mergeable) &&
532 memcpy_toiovecend(vq->hdr, (unsigned char *)&headcount,
533 offsetof(typeof(hdr), num_buffers),
534 sizeof hdr.num_buffers)) {
535 vq_err(vq, "Failed num_buffers write");
536 vhost_discard_vq_desc(vq, headcount);
539 vhost_add_used_and_signal_n(&net->dev, vq, vq->heads,
541 if (unlikely(vq_log))
542 vhost_log_write(vq, vq_log, log, vhost_len);
543 total_len += vhost_len;
544 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
545 vhost_poll_queue(&vq->poll);
550 mutex_unlock(&vq->mutex);
553 static void handle_tx_kick(struct vhost_work *work)
555 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
557 struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
562 static void handle_rx_kick(struct vhost_work *work)
564 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
566 struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
571 static void handle_tx_net(struct vhost_work *work)
573 struct vhost_net *net = container_of(work, struct vhost_net,
574 poll[VHOST_NET_VQ_TX].work);
578 static void handle_rx_net(struct vhost_work *work)
580 struct vhost_net *net = container_of(work, struct vhost_net,
581 poll[VHOST_NET_VQ_RX].work);
585 static int vhost_net_open(struct inode *inode, struct file *f)
587 struct vhost_net *n = kmalloc(sizeof *n, GFP_KERNEL);
588 struct vhost_dev *dev;
595 n->vqs[VHOST_NET_VQ_TX].handle_kick = handle_tx_kick;
596 n->vqs[VHOST_NET_VQ_RX].handle_kick = handle_rx_kick;
597 r = vhost_dev_init(dev, n->vqs, VHOST_NET_VQ_MAX);
603 vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT, dev);
604 vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN, dev);
605 n->tx_poll_state = VHOST_NET_POLL_DISABLED;
612 static void vhost_net_disable_vq(struct vhost_net *n,
613 struct vhost_virtqueue *vq)
615 if (!vq->private_data)
617 if (vq == n->vqs + VHOST_NET_VQ_TX) {
619 n->tx_poll_state = VHOST_NET_POLL_DISABLED;
621 vhost_poll_stop(n->poll + VHOST_NET_VQ_RX);
624 static void vhost_net_enable_vq(struct vhost_net *n,
625 struct vhost_virtqueue *vq)
629 sock = rcu_dereference_protected(vq->private_data,
630 lockdep_is_held(&vq->mutex));
633 if (vq == n->vqs + VHOST_NET_VQ_TX) {
634 n->tx_poll_state = VHOST_NET_POLL_STOPPED;
635 tx_poll_start(n, sock);
637 vhost_poll_start(n->poll + VHOST_NET_VQ_RX, sock->file);
640 static struct socket *vhost_net_stop_vq(struct vhost_net *n,
641 struct vhost_virtqueue *vq)
645 mutex_lock(&vq->mutex);
646 sock = rcu_dereference_protected(vq->private_data,
647 lockdep_is_held(&vq->mutex));
648 vhost_net_disable_vq(n, vq);
649 rcu_assign_pointer(vq->private_data, NULL);
650 mutex_unlock(&vq->mutex);
654 static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock,
655 struct socket **rx_sock)
657 *tx_sock = vhost_net_stop_vq(n, n->vqs + VHOST_NET_VQ_TX);
658 *rx_sock = vhost_net_stop_vq(n, n->vqs + VHOST_NET_VQ_RX);
661 static void vhost_net_flush_vq(struct vhost_net *n, int index)
663 vhost_poll_flush(n->poll + index);
664 vhost_poll_flush(&n->dev.vqs[index].poll);
667 static void vhost_net_flush(struct vhost_net *n)
669 vhost_net_flush_vq(n, VHOST_NET_VQ_TX);
670 vhost_net_flush_vq(n, VHOST_NET_VQ_RX);
673 static int vhost_net_release(struct inode *inode, struct file *f)
675 struct vhost_net *n = f->private_data;
676 struct socket *tx_sock;
677 struct socket *rx_sock;
680 vhost_net_stop(n, &tx_sock, &rx_sock);
682 vhost_dev_stop(&n->dev);
683 for (i = 0; i < n->dev.nvqs; ++i) {
684 /* Wait for all lower device DMAs done. */
685 if (n->dev.vqs[i].ubufs)
686 vhost_ubuf_put_and_wait(n->dev.vqs[i].ubufs);
688 vhost_zerocopy_signal_used(n, &n->dev.vqs[i]);
690 vhost_dev_cleanup(&n->dev, false);
695 /* We do an extra flush before freeing memory,
696 * since jobs can re-queue themselves. */
702 static struct socket *get_raw_socket(int fd)
705 struct sockaddr_ll sa;
706 char buf[MAX_ADDR_LEN];
708 int uaddr_len = sizeof uaddr, r;
709 struct socket *sock = sockfd_lookup(fd, &r);
712 return ERR_PTR(-ENOTSOCK);
714 /* Parameter checking */
715 if (sock->sk->sk_type != SOCK_RAW) {
716 r = -ESOCKTNOSUPPORT;
720 r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa,
725 if (uaddr.sa.sll_family != AF_PACKET) {
735 static struct socket *get_tap_socket(int fd)
737 struct file *file = fget(fd);
741 return ERR_PTR(-EBADF);
742 sock = tun_get_socket(file);
745 sock = macvtap_get_socket(file);
751 static struct socket *get_socket(int fd)
755 /* special case to disable backend */
758 sock = get_raw_socket(fd);
761 sock = get_tap_socket(fd);
764 return ERR_PTR(-ENOTSOCK);
767 static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
769 struct socket *sock, *oldsock;
770 struct vhost_virtqueue *vq;
771 struct vhost_ubuf_ref *ubufs, *oldubufs = NULL;
774 mutex_lock(&n->dev.mutex);
775 r = vhost_dev_check_owner(&n->dev);
779 if (index >= VHOST_NET_VQ_MAX) {
784 mutex_lock(&vq->mutex);
786 /* Verify that ring has been setup correctly. */
787 if (!vhost_vq_access_ok(vq)) {
791 sock = get_socket(fd);
797 /* start polling new socket */
798 oldsock = rcu_dereference_protected(vq->private_data,
799 lockdep_is_held(&vq->mutex));
800 if (sock != oldsock) {
801 ubufs = vhost_ubuf_alloc(vq, sock && vhost_sock_zcopy(sock));
806 oldubufs = vq->ubufs;
808 vhost_net_disable_vq(n, vq);
809 rcu_assign_pointer(vq->private_data, sock);
810 vhost_net_enable_vq(n, vq);
812 r = vhost_init_used(vq);
817 mutex_unlock(&vq->mutex);
820 vhost_ubuf_put_and_wait(oldubufs);
821 mutex_lock(&vq->mutex);
822 vhost_zerocopy_signal_used(n, vq);
823 mutex_unlock(&vq->mutex);
827 vhost_net_flush_vq(n, index);
831 mutex_unlock(&n->dev.mutex);
837 mutex_unlock(&vq->mutex);
839 mutex_unlock(&n->dev.mutex);
843 static long vhost_net_reset_owner(struct vhost_net *n)
845 struct socket *tx_sock = NULL;
846 struct socket *rx_sock = NULL;
849 mutex_lock(&n->dev.mutex);
850 err = vhost_dev_check_owner(&n->dev);
853 vhost_net_stop(n, &tx_sock, &rx_sock);
855 err = vhost_dev_reset_owner(&n->dev);
857 mutex_unlock(&n->dev.mutex);
865 static int vhost_net_set_features(struct vhost_net *n, u64 features)
867 size_t vhost_hlen, sock_hlen, hdr_len;
870 hdr_len = (features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ?
871 sizeof(struct virtio_net_hdr_mrg_rxbuf) :
872 sizeof(struct virtio_net_hdr);
873 if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) {
874 /* vhost provides vnet_hdr */
875 vhost_hlen = hdr_len;
878 /* socket provides vnet_hdr */
882 mutex_lock(&n->dev.mutex);
883 if ((features & (1 << VHOST_F_LOG_ALL)) &&
884 !vhost_log_access_ok(&n->dev)) {
885 mutex_unlock(&n->dev.mutex);
888 n->dev.acked_features = features;
890 for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
891 mutex_lock(&n->vqs[i].mutex);
892 n->vqs[i].vhost_hlen = vhost_hlen;
893 n->vqs[i].sock_hlen = sock_hlen;
894 mutex_unlock(&n->vqs[i].mutex);
897 mutex_unlock(&n->dev.mutex);
901 static long vhost_net_ioctl(struct file *f, unsigned int ioctl,
904 struct vhost_net *n = f->private_data;
905 void __user *argp = (void __user *)arg;
906 u64 __user *featurep = argp;
907 struct vhost_vring_file backend;
912 case VHOST_NET_SET_BACKEND:
913 if (copy_from_user(&backend, argp, sizeof backend))
915 return vhost_net_set_backend(n, backend.index, backend.fd);
916 case VHOST_GET_FEATURES:
917 features = VHOST_NET_FEATURES;
918 if (copy_to_user(featurep, &features, sizeof features))
921 case VHOST_SET_FEATURES:
922 if (copy_from_user(&features, featurep, sizeof features))
924 if (features & ~VHOST_NET_FEATURES)
926 return vhost_net_set_features(n, features);
927 case VHOST_RESET_OWNER:
928 return vhost_net_reset_owner(n);
930 mutex_lock(&n->dev.mutex);
931 r = vhost_dev_ioctl(&n->dev, ioctl, arg);
933 mutex_unlock(&n->dev.mutex);
939 static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl,
942 return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
946 static const struct file_operations vhost_net_fops = {
947 .owner = THIS_MODULE,
948 .release = vhost_net_release,
949 .unlocked_ioctl = vhost_net_ioctl,
951 .compat_ioctl = vhost_net_compat_ioctl,
953 .open = vhost_net_open,
954 .llseek = noop_llseek,
957 static struct miscdevice vhost_net_misc = {
958 .minor = VHOST_NET_MINOR,
960 .fops = &vhost_net_fops,
963 static int vhost_net_init(void)
965 if (experimental_zcopytx)
966 vhost_enable_zcopy(VHOST_NET_VQ_TX);
967 return misc_register(&vhost_net_misc);
969 module_init(vhost_net_init);
971 static void vhost_net_exit(void)
973 misc_deregister(&vhost_net_misc);
975 module_exit(vhost_net_exit);
977 MODULE_VERSION("0.0.1");
978 MODULE_LICENSE("GPL v2");
979 MODULE_AUTHOR("Michael S. Tsirkin");
980 MODULE_DESCRIPTION("Host kernel accelerator for virtio net");
981 MODULE_ALIAS_MISCDEV(VHOST_NET_MINOR);
982 MODULE_ALIAS("devname:vhost-net");