blob: 7d137a43cc86842ed98bc27e63d6dfcb2042f177 [file] [log] [blame]
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001/* 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>
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +000013#include <linux/miscdevice.h>
14#include <linux/module.h>
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +000015#include <linux/moduleparam.h>
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +000016#include <linux/mutex.h>
17#include <linux/workqueue.h>
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +000018#include <linux/file.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/slab.h>
Michael S. Tsirkin23cc5a92013-01-23 21:46:47 +010020#include <linux/vmalloc.h>
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +000021
22#include <linux/net.h>
23#include <linux/if_packet.h>
24#include <linux/if_arp.h>
25#include <linux/if_tun.h>
Arnd Bergmann501c7742010-02-18 05:46:50 +000026#include <linux/if_macvlan.h>
Basil Gorc53cff5e2012-05-03 22:55:23 +000027#include <linux/if_vlan.h>
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +000028
29#include <net/sock.h>
30
31#include "vhost.h"
32
Michael S. Tsirkinf9611c42012-12-06 14:56:00 +020033static int experimental_zcopytx = 1;
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +000034module_param(experimental_zcopytx, int, 0444);
Michael S. Tsirkinf9611c42012-12-06 14:56:00 +020035MODULE_PARM_DESC(experimental_zcopytx, "Enable Zero Copy TX;"
36 " 1 -Enable; 0 - Disable");
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +000037
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +000038/* Max number of bytes transferred before requeueing the job.
39 * Using this limit prevents one virtqueue from starving others. */
40#define VHOST_NET_WEIGHT 0x80000
41
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +000042/* MAX number of TX used buffers for outstanding zerocopy */
43#define VHOST_MAX_PEND 128
44#define VHOST_GOODCOPY_LEN 256
45
Michael S. Tsirkineaae8132012-11-01 09:16:51 +000046/*
47 * For transmit, used buffer len is unused; we override it to track buffer
48 * status internally; used for zerocopy tx only.
49 */
50/* Lower device DMA failed */
Michael S. Tsirkinbf995732014-10-24 11:49:27 +030051#define VHOST_DMA_FAILED_LEN ((__force __virtio32)3)
Michael S. Tsirkineaae8132012-11-01 09:16:51 +000052/* Lower device DMA done */
Michael S. Tsirkinbf995732014-10-24 11:49:27 +030053#define VHOST_DMA_DONE_LEN ((__force __virtio32)2)
Michael S. Tsirkineaae8132012-11-01 09:16:51 +000054/* Lower device DMA in progress */
Michael S. Tsirkinbf995732014-10-24 11:49:27 +030055#define VHOST_DMA_IN_PROGRESS ((__force __virtio32)1)
Michael S. Tsirkineaae8132012-11-01 09:16:51 +000056/* Buffer unused */
Michael S. Tsirkinbf995732014-10-24 11:49:27 +030057#define VHOST_DMA_CLEAR_LEN ((__force __virtio32)0)
Michael S. Tsirkineaae8132012-11-01 09:16:51 +000058
Michael S. Tsirkinbf995732014-10-24 11:49:27 +030059#define VHOST_DMA_IS_DONE(len) ((__force u32)(len) >= (__force u32)VHOST_DMA_DONE_LEN)
Michael S. Tsirkineaae8132012-11-01 09:16:51 +000060
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +000061enum {
Asias He8570a6e2013-05-06 16:38:20 +080062 VHOST_NET_FEATURES = VHOST_FEATURES |
63 (1ULL << VHOST_NET_F_VIRTIO_NET_HDR) |
Michael S. Tsirkin41e3e422014-10-24 14:25:03 +030064 (1ULL << VIRTIO_NET_F_MRG_RXBUF) |
65 (1ULL << VIRTIO_F_VERSION_1),
Asias He8570a6e2013-05-06 16:38:20 +080066};
67
68enum {
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +000069 VHOST_NET_VQ_RX = 0,
70 VHOST_NET_VQ_TX = 1,
71 VHOST_NET_VQ_MAX = 2,
72};
73
Asias Hefe729a52013-05-06 16:38:24 +080074struct vhost_net_ubuf_ref {
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +020075 /* refcount follows semantics similar to kref:
76 * 0: object is released
77 * 1: no outstanding ubufs
78 * >1: outstanding ubufs
79 */
80 atomic_t refcount;
Asias He28394002013-04-27 15:07:46 +080081 wait_queue_head_t wait;
82 struct vhost_virtqueue *vq;
83};
84
Asias He3ab2e422013-04-27 11:16:48 +080085struct vhost_net_virtqueue {
86 struct vhost_virtqueue vq;
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +030087 size_t vhost_hlen;
88 size_t sock_hlen;
Asias He28394002013-04-27 15:07:46 +080089 /* vhost zerocopy support fields below: */
90 /* last used idx for outstanding DMA zerocopy buffers */
91 int upend_idx;
92 /* first used idx for DMA done zerocopy buffers */
93 int done_idx;
94 /* an array of userspace buffers info */
95 struct ubuf_info *ubuf_info;
96 /* Reference counting for outstanding ubufs.
97 * Protected by vq mutex. Writers must also take device mutex. */
Asias Hefe729a52013-05-06 16:38:24 +080098 struct vhost_net_ubuf_ref *ubufs;
Asias He3ab2e422013-04-27 11:16:48 +080099};
100
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000101struct vhost_net {
102 struct vhost_dev dev;
Asias He3ab2e422013-04-27 11:16:48 +0800103 struct vhost_net_virtqueue vqs[VHOST_NET_VQ_MAX];
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000104 struct vhost_poll poll[VHOST_NET_VQ_MAX];
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000105 /* Number of TX recently submitted.
106 * Protected by tx vq lock. */
107 unsigned tx_packets;
108 /* Number of times zerocopy TX recently failed.
109 * Protected by tx vq lock. */
110 unsigned tx_zcopy_err;
Michael S. Tsirkin1280c272012-12-04 00:17:14 +0200111 /* Flush in progress. Protected by tx vq lock. */
112 bool tx_flush;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000113};
114
Asias Hefe729a52013-05-06 16:38:24 +0800115static unsigned vhost_net_zcopy_mask __read_mostly;
Asias He28394002013-04-27 15:07:46 +0800116
Asias Hefe729a52013-05-06 16:38:24 +0800117static void vhost_net_enable_zcopy(int vq)
Asias He28394002013-04-27 15:07:46 +0800118{
Asias Hefe729a52013-05-06 16:38:24 +0800119 vhost_net_zcopy_mask |= 0x1 << vq;
Asias He28394002013-04-27 15:07:46 +0800120}
121
Asias Hefe729a52013-05-06 16:38:24 +0800122static struct vhost_net_ubuf_ref *
123vhost_net_ubuf_alloc(struct vhost_virtqueue *vq, bool zcopy)
Asias He28394002013-04-27 15:07:46 +0800124{
Asias Hefe729a52013-05-06 16:38:24 +0800125 struct vhost_net_ubuf_ref *ubufs;
Asias He28394002013-04-27 15:07:46 +0800126 /* No zero copy backend? Nothing to count. */
127 if (!zcopy)
128 return NULL;
129 ubufs = kmalloc(sizeof(*ubufs), GFP_KERNEL);
130 if (!ubufs)
131 return ERR_PTR(-ENOMEM);
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200132 atomic_set(&ubufs->refcount, 1);
Asias He28394002013-04-27 15:07:46 +0800133 init_waitqueue_head(&ubufs->wait);
134 ubufs->vq = vq;
135 return ubufs;
136}
137
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200138static int vhost_net_ubuf_put(struct vhost_net_ubuf_ref *ubufs)
Asias He28394002013-04-27 15:07:46 +0800139{
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200140 int r = atomic_sub_return(1, &ubufs->refcount);
141 if (unlikely(!r))
142 wake_up(&ubufs->wait);
143 return r;
Asias He28394002013-04-27 15:07:46 +0800144}
145
Asias Hefe729a52013-05-06 16:38:24 +0800146static void vhost_net_ubuf_put_and_wait(struct vhost_net_ubuf_ref *ubufs)
Asias He28394002013-04-27 15:07:46 +0800147{
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200148 vhost_net_ubuf_put(ubufs);
149 wait_event(ubufs->wait, !atomic_read(&ubufs->refcount));
Michael S. Tsirkinc38e39c2013-06-25 17:29:46 +0300150}
151
152static void vhost_net_ubuf_put_wait_and_free(struct vhost_net_ubuf_ref *ubufs)
153{
154 vhost_net_ubuf_put_and_wait(ubufs);
Asias He28394002013-04-27 15:07:46 +0800155 kfree(ubufs);
156}
157
Asias Heb1ad8492013-05-06 11:16:00 +0800158static void vhost_net_clear_ubuf_info(struct vhost_net *n)
159{
Asias Heb1ad8492013-05-06 11:16:00 +0800160 int i;
161
Michael S. Tsirkin288cfe72013-06-06 15:20:46 +0300162 for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
163 kfree(n->vqs[i].ubuf_info);
164 n->vqs[i].ubuf_info = NULL;
Asias Heb1ad8492013-05-06 11:16:00 +0800165 }
166}
167
Asias He0a1febf2013-06-05 21:17:38 +0800168static int vhost_net_set_ubuf_info(struct vhost_net *n)
Asias He28394002013-04-27 15:07:46 +0800169{
170 bool zcopy;
171 int i;
172
Michael S. Tsirkin288cfe72013-06-06 15:20:46 +0300173 for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
Asias Hefe729a52013-05-06 16:38:24 +0800174 zcopy = vhost_net_zcopy_mask & (0x1 << i);
Asias He28394002013-04-27 15:07:46 +0800175 if (!zcopy)
176 continue;
177 n->vqs[i].ubuf_info = kmalloc(sizeof(*n->vqs[i].ubuf_info) *
178 UIO_MAXIOV, GFP_KERNEL);
179 if (!n->vqs[i].ubuf_info)
180 goto err;
181 }
182 return 0;
183
184err:
Michael S. Tsirkin288cfe72013-06-06 15:20:46 +0300185 vhost_net_clear_ubuf_info(n);
Asias He28394002013-04-27 15:07:46 +0800186 return -ENOMEM;
187}
188
Asias He0a1febf2013-06-05 21:17:38 +0800189static void vhost_net_vq_reset(struct vhost_net *n)
Asias He28394002013-04-27 15:07:46 +0800190{
191 int i;
192
Michael S. Tsirkin288cfe72013-06-06 15:20:46 +0300193 vhost_net_clear_ubuf_info(n);
194
Asias He28394002013-04-27 15:07:46 +0800195 for (i = 0; i < VHOST_NET_VQ_MAX; i++) {
196 n->vqs[i].done_idx = 0;
197 n->vqs[i].upend_idx = 0;
198 n->vqs[i].ubufs = NULL;
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300199 n->vqs[i].vhost_hlen = 0;
200 n->vqs[i].sock_hlen = 0;
Asias He28394002013-04-27 15:07:46 +0800201 }
202
203}
204
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000205static void vhost_net_tx_packet(struct vhost_net *net)
206{
207 ++net->tx_packets;
208 if (net->tx_packets < 1024)
209 return;
210 net->tx_packets = 0;
211 net->tx_zcopy_err = 0;
212}
213
214static void vhost_net_tx_err(struct vhost_net *net)
215{
216 ++net->tx_zcopy_err;
217}
218
219static bool vhost_net_tx_select_zcopy(struct vhost_net *net)
220{
Michael S. Tsirkin1280c272012-12-04 00:17:14 +0200221 /* TX flush waits for outstanding DMAs to be done.
222 * Don't start new DMAs.
223 */
224 return !net->tx_flush &&
225 net->tx_packets / 64 >= net->tx_zcopy_err;
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000226}
227
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000228static bool vhost_sock_zcopy(struct socket *sock)
229{
230 return unlikely(experimental_zcopytx) &&
231 sock_flag(sock->sk, SOCK_ZEROCOPY);
232}
233
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000234/* In case of DMA done not in order in lower device driver for some reason.
235 * upend_idx is used to track end of used idx, done_idx is used to track head
236 * of used idx. Once lower device DMA done contiguously, we will signal KVM
237 * guest used idx.
238 */
Jason Wang094afe72013-09-02 16:40:56 +0800239static void vhost_zerocopy_signal_used(struct vhost_net *net,
240 struct vhost_virtqueue *vq)
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000241{
Asias He28394002013-04-27 15:07:46 +0800242 struct vhost_net_virtqueue *nvq =
243 container_of(vq, struct vhost_net_virtqueue, vq);
Jason Wangc92112a2013-09-02 16:40:57 +0800244 int i, add;
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000245 int j = 0;
246
Asias He28394002013-04-27 15:07:46 +0800247 for (i = nvq->done_idx; i != nvq->upend_idx; i = (i + 1) % UIO_MAXIOV) {
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000248 if (vq->heads[i].len == VHOST_DMA_FAILED_LEN)
249 vhost_net_tx_err(net);
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000250 if (VHOST_DMA_IS_DONE(vq->heads[i].len)) {
251 vq->heads[i].len = VHOST_DMA_CLEAR_LEN;
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000252 ++j;
253 } else
254 break;
255 }
Jason Wangc92112a2013-09-02 16:40:57 +0800256 while (j) {
257 add = min(UIO_MAXIOV - nvq->done_idx, j);
258 vhost_add_used_and_signal_n(vq->dev, vq,
259 &vq->heads[nvq->done_idx], add);
260 nvq->done_idx = (nvq->done_idx + add) % UIO_MAXIOV;
261 j -= add;
262 }
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000263}
264
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000265static void vhost_zerocopy_callback(struct ubuf_info *ubuf, bool success)
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000266{
Asias Hefe729a52013-05-06 16:38:24 +0800267 struct vhost_net_ubuf_ref *ubufs = ubuf->ctx;
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000268 struct vhost_virtqueue *vq = ubufs->vq;
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200269 int cnt;
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000270
Michael S. Tsirkinb0c057c2014-02-13 11:45:11 +0200271 rcu_read_lock_bh();
272
Jason Wang19c73b32013-09-02 16:41:00 +0800273 /* set len to mark this desc buffers done DMA */
274 vq->heads[ubuf->desc].len = success ?
275 VHOST_DMA_DONE_LEN : VHOST_DMA_FAILED_LEN;
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200276 cnt = vhost_net_ubuf_put(ubufs);
Jason Wang19c73b32013-09-02 16:41:00 +0800277
Michael S. Tsirkin24eb21a2012-11-01 09:16:55 +0000278 /*
279 * Trigger polling thread if guest stopped submitting new buffers:
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200280 * in this case, the refcount after decrement will eventually reach 1.
Michael S. Tsirkin24eb21a2012-11-01 09:16:55 +0000281 * We also trigger polling periodically after each 16 packets
282 * (the value 16 here is more or less arbitrary, it's tuned to trigger
283 * less than 10% of times).
284 */
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200285 if (cnt <= 1 || !(cnt % 16))
Michael S. Tsirkin24eb21a2012-11-01 09:16:55 +0000286 vhost_poll_queue(&vq->poll);
Michael S. Tsirkinb0c057c2014-02-13 11:45:11 +0200287
288 rcu_read_unlock_bh();
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000289}
290
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000291/* Expects to be always run from workqueue - which acts as
292 * read-size critical section for our kind of RCU. */
293static void handle_tx(struct vhost_net *net)
294{
Asias He28394002013-04-27 15:07:46 +0800295 struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX];
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300296 struct vhost_virtqueue *vq = &nvq->vq;
Al Viro98a527a2014-12-10 15:00:58 -0500297 unsigned out, in;
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +0300298 int head;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000299 struct msghdr msg = {
300 .msg_name = NULL,
301 .msg_namelen = 0,
302 .msg_control = NULL,
303 .msg_controllen = 0,
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000304 .msg_flags = MSG_DONTWAIT,
305 };
306 size_t len, total_len = 0;
Jason Wang70181d512013-04-10 20:50:48 +0000307 int err;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000308 size_t hdr_size;
Arnd Bergmann28457ee2010-03-09 19:24:45 +0100309 struct socket *sock;
Asias Hefe729a52013-05-06 16:38:24 +0800310 struct vhost_net_ubuf_ref *uninitialized_var(ubufs);
Michael S. Tsirkincedb9bd2012-12-06 17:00:18 +0200311 bool zcopy, zcopy_used;
Arnd Bergmann28457ee2010-03-09 19:24:45 +0100312
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000313 mutex_lock(&vq->mutex);
Asias He2e26af72013-05-07 14:54:33 +0800314 sock = vq->private_data;
315 if (!sock)
316 goto out;
317
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +0300318 vhost_disable_notify(&net->dev, vq);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000319
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300320 hdr_size = nvq->vhost_hlen;
Asias He28394002013-04-27 15:07:46 +0800321 zcopy = nvq->ubufs;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000322
323 for (;;) {
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000324 /* Release DMAs done buffers first */
325 if (zcopy)
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000326 vhost_zerocopy_signal_used(net, vq);
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000327
Jason Wangf7c6be42013-09-02 16:41:01 +0800328 /* If more outstanding DMAs, queue the work.
329 * Handle upend_idx wrap around
330 */
331 if (unlikely((nvq->upend_idx + vq->num - VHOST_MAX_PEND)
332 % UIO_MAXIOV == nvq->done_idx))
333 break;
334
Michael S. Tsirkin47283be2014-06-05 15:20:27 +0300335 head = vhost_get_vq_desc(vq, vq->iov,
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000336 ARRAY_SIZE(vq->iov),
337 &out, &in,
338 NULL, NULL);
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +0300339 /* On error, stop handling until the next kick. */
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +0300340 if (unlikely(head < 0))
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +0300341 break;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000342 /* Nothing new? Wait for eventfd to tell us they refilled. */
343 if (head == vq->num) {
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +0300344 if (unlikely(vhost_enable_notify(&net->dev, vq))) {
345 vhost_disable_notify(&net->dev, vq);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000346 continue;
347 }
348 break;
349 }
350 if (in) {
351 vq_err(vq, "Unexpected descriptor format for TX: "
352 "out %d, int %d\n", out, in);
353 break;
354 }
355 /* Skip header. TODO: support TSO. */
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000356 len = iov_length(vq->iov, out);
Al Viroc0371da2014-11-24 10:42:55 -0500357 iov_iter_init(&msg.msg_iter, WRITE, vq->iov, out, len);
Al Viro98a527a2014-12-10 15:00:58 -0500358 iov_iter_advance(&msg.msg_iter, hdr_size);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000359 /* Sanity check */
Al Viro01e97e62014-12-15 21:39:31 -0500360 if (!msg_data_left(&msg)) {
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000361 vq_err(vq, "Unexpected header len for TX: "
362 "%zd expected %zd\n",
Al Viro98a527a2014-12-10 15:00:58 -0500363 len, hdr_size);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000364 break;
365 }
Al Viro01e97e62014-12-15 21:39:31 -0500366 len = msg_data_left(&msg);
Jason Wangce21a022013-09-02 16:40:59 +0800367
368 zcopy_used = zcopy && len >= VHOST_GOODCOPY_LEN
369 && (nvq->upend_idx + 1) % UIO_MAXIOV !=
370 nvq->done_idx
371 && vhost_net_tx_select_zcopy(net);
Michael S. Tsirkincedb9bd2012-12-06 17:00:18 +0200372
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000373 /* use msg_control to pass vhost zerocopy ubuf info to skb */
Michael S. Tsirkincedb9bd2012-12-06 17:00:18 +0200374 if (zcopy_used) {
Jason Wangce21a022013-09-02 16:40:59 +0800375 struct ubuf_info *ubuf;
376 ubuf = nvq->ubuf_info + nvq->upend_idx;
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000377
Michael S. Tsirkin8b386942014-10-24 14:19:48 +0300378 vq->heads[nvq->upend_idx].id = cpu_to_vhost32(vq, head);
Jason Wangce21a022013-09-02 16:40:59 +0800379 vq->heads[nvq->upend_idx].len = VHOST_DMA_IN_PROGRESS;
380 ubuf->callback = vhost_zerocopy_callback;
381 ubuf->ctx = nvq->ubufs;
382 ubuf->desc = nvq->upend_idx;
383 msg.msg_control = ubuf;
384 msg.msg_controllen = sizeof(ubuf);
385 ubufs = nvq->ubufs;
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200386 atomic_inc(&ubufs->refcount);
Asias He28394002013-04-27 15:07:46 +0800387 nvq->upend_idx = (nvq->upend_idx + 1) % UIO_MAXIOV;
Jason Wangce21a022013-09-02 16:40:59 +0800388 } else {
Jason Wang4364d5f2013-06-05 15:40:46 +0800389 msg.msg_control = NULL;
Jason Wangce21a022013-09-02 16:40:59 +0800390 ubufs = NULL;
391 }
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000392 /* TODO: Check specific error and bomb out unless ENOBUFS? */
Ying Xue1b784142015-03-02 15:37:48 +0800393 err = sock->ops->sendmsg(sock, &msg, len);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000394 if (unlikely(err < 0)) {
Michael S. Tsirkincedb9bd2012-12-06 17:00:18 +0200395 if (zcopy_used) {
Jason Wangce21a022013-09-02 16:40:59 +0800396 vhost_net_ubuf_put(ubufs);
Asias He28394002013-04-27 15:07:46 +0800397 nvq->upend_idx = ((unsigned)nvq->upend_idx - 1)
398 % UIO_MAXIOV;
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000399 }
David Stevens8dd014a2010-07-27 18:52:21 +0300400 vhost_discard_vq_desc(vq, 1);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000401 break;
402 }
403 if (err != len)
Michael S. Tsirkin95c0ec62010-06-24 17:10:25 +0300404 pr_debug("Truncated TX packet: "
405 " len %d != %zd\n", err, len);
Michael S. Tsirkincedb9bd2012-12-06 17:00:18 +0200406 if (!zcopy_used)
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000407 vhost_add_used_and_signal(&net->dev, vq, head, 0);
Jason Wangc8fb2172012-05-02 11:42:41 +0800408 else
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000409 vhost_zerocopy_signal_used(net, vq);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000410 total_len += len;
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000411 vhost_net_tx_packet(net);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000412 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
413 vhost_poll_queue(&vq->poll);
414 break;
415 }
416 }
Asias He2e26af72013-05-07 14:54:33 +0800417out:
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000418 mutex_unlock(&vq->mutex);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000419}
420
David Stevens8dd014a2010-07-27 18:52:21 +0300421static int peek_head_len(struct sock *sk)
422{
423 struct sk_buff *head;
424 int len = 0;
Jason Wang783e3982011-01-17 16:11:17 +0800425 unsigned long flags;
David Stevens8dd014a2010-07-27 18:52:21 +0300426
Jason Wang783e3982011-01-17 16:11:17 +0800427 spin_lock_irqsave(&sk->sk_receive_queue.lock, flags);
David Stevens8dd014a2010-07-27 18:52:21 +0300428 head = skb_peek(&sk->sk_receive_queue);
Basil Gorc53cff5e2012-05-03 22:55:23 +0000429 if (likely(head)) {
David Stevens8dd014a2010-07-27 18:52:21 +0300430 len = head->len;
Jiri Pirkodf8a39d2015-01-13 17:13:44 +0100431 if (skb_vlan_tag_present(head))
Basil Gorc53cff5e2012-05-03 22:55:23 +0000432 len += VLAN_HLEN;
433 }
434
Jason Wang783e3982011-01-17 16:11:17 +0800435 spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags);
David Stevens8dd014a2010-07-27 18:52:21 +0300436 return len;
437}
438
439/* This is a multi-buffer version of vhost_get_desc, that works if
440 * vq has read descriptors only.
441 * @vq - the relevant virtqueue
442 * @datalen - data length we'll be reading
443 * @iovcount - returned count of io vectors we fill
444 * @log - vhost log
445 * @log_num - log offset
Jason Wang94249362011-01-17 16:11:08 +0800446 * @quota - headcount quota, 1 for big buffer
David Stevens8dd014a2010-07-27 18:52:21 +0300447 * returns number of buffer heads allocated, negative on error
448 */
449static int get_rx_bufs(struct vhost_virtqueue *vq,
450 struct vring_used_elem *heads,
451 int datalen,
452 unsigned *iovcount,
453 struct vhost_log *log,
Jason Wang94249362011-01-17 16:11:08 +0800454 unsigned *log_num,
455 unsigned int quota)
David Stevens8dd014a2010-07-27 18:52:21 +0300456{
457 unsigned int out, in;
458 int seg = 0;
459 int headcount = 0;
460 unsigned d;
461 int r, nlogs = 0;
Michael S. Tsirkin8b386942014-10-24 14:19:48 +0300462 /* len is always initialized before use since we are always called with
463 * datalen > 0.
464 */
465 u32 uninitialized_var(len);
David Stevens8dd014a2010-07-27 18:52:21 +0300466
Jason Wang94249362011-01-17 16:11:08 +0800467 while (datalen > 0 && headcount < quota) {
Jason Wange0e9b402010-09-14 23:53:05 +0800468 if (unlikely(seg >= UIO_MAXIOV)) {
David Stevens8dd014a2010-07-27 18:52:21 +0300469 r = -ENOBUFS;
470 goto err;
471 }
Michael S. Tsirkin47283be2014-06-05 15:20:27 +0300472 r = vhost_get_vq_desc(vq, vq->iov + seg,
David Stevens8dd014a2010-07-27 18:52:21 +0300473 ARRAY_SIZE(vq->iov) - seg, &out,
474 &in, log, log_num);
Michael S. Tsirkina39ee442014-03-27 12:53:37 +0200475 if (unlikely(r < 0))
476 goto err;
477
478 d = r;
David Stevens8dd014a2010-07-27 18:52:21 +0300479 if (d == vq->num) {
480 r = 0;
481 goto err;
482 }
483 if (unlikely(out || in <= 0)) {
484 vq_err(vq, "unexpected descriptor format for RX: "
485 "out %d, in %d\n", out, in);
486 r = -EINVAL;
487 goto err;
488 }
489 if (unlikely(log)) {
490 nlogs += *log_num;
491 log += *log_num;
492 }
Michael S. Tsirkin8b386942014-10-24 14:19:48 +0300493 heads[headcount].id = cpu_to_vhost32(vq, d);
494 len = iov_length(vq->iov + seg, in);
495 heads[headcount].len = cpu_to_vhost32(vq, len);
496 datalen -= len;
David Stevens8dd014a2010-07-27 18:52:21 +0300497 ++headcount;
498 seg += in;
499 }
Michael S. Tsirkin99975cc2015-01-07 10:51:00 +0200500 heads[headcount - 1].len = cpu_to_vhost32(vq, len + datalen);
David Stevens8dd014a2010-07-27 18:52:21 +0300501 *iovcount = seg;
502 if (unlikely(log))
503 *log_num = nlogs;
Michael S. Tsirkind8316f32014-03-27 12:00:26 +0200504
505 /* Detect overrun */
506 if (unlikely(datalen > 0)) {
507 r = UIO_MAXIOV + 1;
508 goto err;
509 }
David Stevens8dd014a2010-07-27 18:52:21 +0300510 return headcount;
511err:
512 vhost_discard_vq_desc(vq, headcount);
513 return r;
514}
515
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000516/* Expects to be always run from workqueue - which acts as
517 * read-size critical section for our kind of RCU. */
Jason Wang94249362011-01-17 16:11:08 +0800518static void handle_rx(struct vhost_net *net)
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000519{
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300520 struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_RX];
521 struct vhost_virtqueue *vq = &nvq->vq;
David Stevens8dd014a2010-07-27 18:52:21 +0300522 unsigned uninitialized_var(in), log;
523 struct vhost_log *vq_log;
524 struct msghdr msg = {
525 .msg_name = NULL,
526 .msg_namelen = 0,
527 .msg_control = NULL, /* FIXME: get and handle RX aux data. */
528 .msg_controllen = 0,
David Stevens8dd014a2010-07-27 18:52:21 +0300529 .msg_flags = MSG_DONTWAIT,
530 };
Jason Wang0960b642015-02-15 16:35:17 +0800531 struct virtio_net_hdr hdr = {
532 .flags = 0,
533 .gso_type = VIRTIO_NET_HDR_GSO_NONE
David Stevens8dd014a2010-07-27 18:52:21 +0300534 };
David Stevens8dd014a2010-07-27 18:52:21 +0300535 size_t total_len = 0;
Michael S. Tsirkin910a5782012-10-24 20:37:51 +0200536 int err, mergeable;
537 s16 headcount;
David Stevens8dd014a2010-07-27 18:52:21 +0300538 size_t vhost_hlen, sock_hlen;
539 size_t vhost_len, sock_len;
Asias He2e26af72013-05-07 14:54:33 +0800540 struct socket *sock;
Al Viroba7438a2014-12-10 15:51:28 -0500541 struct iov_iter fixup;
Jason Wang0960b642015-02-15 16:35:17 +0800542 __virtio16 num_buffers;
David Stevens8dd014a2010-07-27 18:52:21 +0300543
David Stevens8dd014a2010-07-27 18:52:21 +0300544 mutex_lock(&vq->mutex);
Asias He2e26af72013-05-07 14:54:33 +0800545 sock = vq->private_data;
546 if (!sock)
547 goto out;
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +0300548 vhost_disable_notify(&net->dev, vq);
Asias He2e26af72013-05-07 14:54:33 +0800549
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300550 vhost_hlen = nvq->vhost_hlen;
551 sock_hlen = nvq->sock_hlen;
David Stevens8dd014a2010-07-27 18:52:21 +0300552
Michael S. Tsirkinea16c512014-06-05 15:20:23 +0300553 vq_log = unlikely(vhost_has_feature(vq, VHOST_F_LOG_ALL)) ?
David Stevens8dd014a2010-07-27 18:52:21 +0300554 vq->log : NULL;
Michael S. Tsirkinea16c512014-06-05 15:20:23 +0300555 mergeable = vhost_has_feature(vq, VIRTIO_NET_F_MRG_RXBUF);
David Stevens8dd014a2010-07-27 18:52:21 +0300556
557 while ((sock_len = peek_head_len(sock->sk))) {
558 sock_len += sock_hlen;
559 vhost_len = sock_len + vhost_hlen;
560 headcount = get_rx_bufs(vq, vq->heads, vhost_len,
Jason Wang94249362011-01-17 16:11:08 +0800561 &in, vq_log, &log,
562 likely(mergeable) ? UIO_MAXIOV : 1);
David Stevens8dd014a2010-07-27 18:52:21 +0300563 /* On error, stop handling until the next kick. */
564 if (unlikely(headcount < 0))
565 break;
Michael S. Tsirkind8316f32014-03-27 12:00:26 +0200566 /* On overrun, truncate and discard */
567 if (unlikely(headcount > UIO_MAXIOV)) {
Al Viroc0371da2014-11-24 10:42:55 -0500568 iov_iter_init(&msg.msg_iter, READ, vq->iov, 1, 1);
Ying Xue1b784142015-03-02 15:37:48 +0800569 err = sock->ops->recvmsg(sock, &msg,
Michael S. Tsirkind8316f32014-03-27 12:00:26 +0200570 1, MSG_DONTWAIT | MSG_TRUNC);
571 pr_debug("Discarded rx packet: len %zd\n", sock_len);
572 continue;
573 }
David Stevens8dd014a2010-07-27 18:52:21 +0300574 /* OK, now we need to know about added descriptors. */
575 if (!headcount) {
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +0300576 if (unlikely(vhost_enable_notify(&net->dev, vq))) {
David Stevens8dd014a2010-07-27 18:52:21 +0300577 /* They have slipped one in as we were
578 * doing that: check again. */
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +0300579 vhost_disable_notify(&net->dev, vq);
David Stevens8dd014a2010-07-27 18:52:21 +0300580 continue;
581 }
582 /* Nothing new? Wait for eventfd to tell us
583 * they refilled. */
584 break;
585 }
586 /* We don't need to be notified again. */
Al Viroba7438a2014-12-10 15:51:28 -0500587 iov_iter_init(&msg.msg_iter, READ, vq->iov, in, vhost_len);
588 fixup = msg.msg_iter;
589 if (unlikely((vhost_hlen))) {
590 /* We will supply the header ourselves
591 * TODO: support TSO.
592 */
593 iov_iter_advance(&msg.msg_iter, vhost_hlen);
Al Viroba7438a2014-12-10 15:51:28 -0500594 }
Ying Xue1b784142015-03-02 15:37:48 +0800595 err = sock->ops->recvmsg(sock, &msg,
David Stevens8dd014a2010-07-27 18:52:21 +0300596 sock_len, MSG_DONTWAIT | MSG_TRUNC);
597 /* Userspace might have consumed the packet meanwhile:
598 * it's not supposed to do this usually, but might be hard
599 * to prevent. Discard data we got (if any) and keep going. */
600 if (unlikely(err != sock_len)) {
601 pr_debug("Discarded rx packet: "
602 " len %d, expected %zd\n", err, sock_len);
603 vhost_discard_vq_desc(vq, headcount);
604 continue;
605 }
Al Viroba7438a2014-12-10 15:51:28 -0500606 /* Supply virtio_net_hdr if VHOST_NET_F_VIRTIO_NET_HDR */
Michael S. Tsirkin4c5a8442015-02-25 15:19:28 +0100607 if (unlikely(vhost_hlen)) {
608 if (copy_to_iter(&hdr, sizeof(hdr),
609 &fixup) != sizeof(hdr)) {
610 vq_err(vq, "Unable to write vnet_hdr "
611 "at addr %p\n", vq->iov->iov_base);
612 break;
613 }
614 } else {
615 /* Header came from socket; we'll need to patch
616 * ->num_buffers over if VIRTIO_NET_F_MRG_RXBUF
617 */
618 iov_iter_advance(&fixup, sizeof(hdr));
David Stevens8dd014a2010-07-27 18:52:21 +0300619 }
620 /* TODO: Should check and handle checksum. */
Michael S. Tsirkin5201aa42015-02-03 11:07:06 +0200621
Jason Wang0960b642015-02-15 16:35:17 +0800622 num_buffers = cpu_to_vhost16(vq, headcount);
Jason Wangcfbdab92011-01-17 16:10:59 +0800623 if (likely(mergeable) &&
Michael S. Tsirkin0d79a492015-02-25 15:20:01 +0100624 copy_to_iter(&num_buffers, sizeof num_buffers,
625 &fixup) != sizeof num_buffers) {
David Stevens8dd014a2010-07-27 18:52:21 +0300626 vq_err(vq, "Failed num_buffers write");
627 vhost_discard_vq_desc(vq, headcount);
628 break;
629 }
630 vhost_add_used_and_signal_n(&net->dev, vq, vq->heads,
631 headcount);
632 if (unlikely(vq_log))
633 vhost_log_write(vq, vq_log, log, vhost_len);
634 total_len += vhost_len;
635 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
636 vhost_poll_queue(&vq->poll);
637 break;
638 }
639 }
Asias He2e26af72013-05-07 14:54:33 +0800640out:
David Stevens8dd014a2010-07-27 18:52:21 +0300641 mutex_unlock(&vq->mutex);
David Stevens8dd014a2010-07-27 18:52:21 +0300642}
643
Tejun Heoc23f34452010-06-02 20:40:00 +0200644static void handle_tx_kick(struct vhost_work *work)
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000645{
Tejun Heoc23f34452010-06-02 20:40:00 +0200646 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
647 poll.work);
648 struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
649
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000650 handle_tx(net);
651}
652
Tejun Heoc23f34452010-06-02 20:40:00 +0200653static void handle_rx_kick(struct vhost_work *work)
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000654{
Tejun Heoc23f34452010-06-02 20:40:00 +0200655 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
656 poll.work);
657 struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
658
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000659 handle_rx(net);
660}
661
Tejun Heoc23f34452010-06-02 20:40:00 +0200662static void handle_tx_net(struct vhost_work *work)
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000663{
Tejun Heoc23f34452010-06-02 20:40:00 +0200664 struct vhost_net *net = container_of(work, struct vhost_net,
665 poll[VHOST_NET_VQ_TX].work);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000666 handle_tx(net);
667}
668
Tejun Heoc23f34452010-06-02 20:40:00 +0200669static void handle_rx_net(struct vhost_work *work)
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000670{
Tejun Heoc23f34452010-06-02 20:40:00 +0200671 struct vhost_net *net = container_of(work, struct vhost_net,
672 poll[VHOST_NET_VQ_RX].work);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000673 handle_rx(net);
674}
675
676static int vhost_net_open(struct inode *inode, struct file *f)
677{
Michael S. Tsirkin23cc5a92013-01-23 21:46:47 +0100678 struct vhost_net *n;
Tejun Heoc23f34452010-06-02 20:40:00 +0200679 struct vhost_dev *dev;
Asias He3ab2e422013-04-27 11:16:48 +0800680 struct vhost_virtqueue **vqs;
Zhi Yong Wu59566b6e2013-12-07 04:13:03 +0800681 int i;
Tejun Heoc23f34452010-06-02 20:40:00 +0200682
Michael S. Tsirkin23cc5a92013-01-23 21:46:47 +0100683 n = kmalloc(sizeof *n, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
684 if (!n) {
685 n = vmalloc(sizeof *n);
686 if (!n)
687 return -ENOMEM;
688 }
Asias He3ab2e422013-04-27 11:16:48 +0800689 vqs = kmalloc(VHOST_NET_VQ_MAX * sizeof(*vqs), GFP_KERNEL);
690 if (!vqs) {
Romain Francoised04257b2014-06-12 10:42:34 +0200691 kvfree(n);
Asias He3ab2e422013-04-27 11:16:48 +0800692 return -ENOMEM;
693 }
Tejun Heoc23f34452010-06-02 20:40:00 +0200694
695 dev = &n->dev;
Asias He3ab2e422013-04-27 11:16:48 +0800696 vqs[VHOST_NET_VQ_TX] = &n->vqs[VHOST_NET_VQ_TX].vq;
697 vqs[VHOST_NET_VQ_RX] = &n->vqs[VHOST_NET_VQ_RX].vq;
698 n->vqs[VHOST_NET_VQ_TX].vq.handle_kick = handle_tx_kick;
699 n->vqs[VHOST_NET_VQ_RX].vq.handle_kick = handle_rx_kick;
Asias He28394002013-04-27 15:07:46 +0800700 for (i = 0; i < VHOST_NET_VQ_MAX; i++) {
701 n->vqs[i].ubufs = NULL;
702 n->vqs[i].ubuf_info = NULL;
703 n->vqs[i].upend_idx = 0;
704 n->vqs[i].done_idx = 0;
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300705 n->vqs[i].vhost_hlen = 0;
706 n->vqs[i].sock_hlen = 0;
Asias He28394002013-04-27 15:07:46 +0800707 }
Zhi Yong Wu59566b6e2013-12-07 04:13:03 +0800708 vhost_dev_init(dev, vqs, VHOST_NET_VQ_MAX);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000709
Tejun Heoc23f34452010-06-02 20:40:00 +0200710 vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT, dev);
711 vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN, dev);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000712
713 f->private_data = n;
714
715 return 0;
716}
717
718static void vhost_net_disable_vq(struct vhost_net *n,
719 struct vhost_virtqueue *vq)
720{
Asias He3ab2e422013-04-27 11:16:48 +0800721 struct vhost_net_virtqueue *nvq =
722 container_of(vq, struct vhost_net_virtqueue, vq);
723 struct vhost_poll *poll = n->poll + (nvq - n->vqs);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000724 if (!vq->private_data)
725 return;
Jason Wang70181d512013-04-10 20:50:48 +0000726 vhost_poll_stop(poll);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000727}
728
Jason Wang2b8b3282013-01-28 01:05:18 +0000729static int vhost_net_enable_vq(struct vhost_net *n,
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000730 struct vhost_virtqueue *vq)
731{
Asias He3ab2e422013-04-27 11:16:48 +0800732 struct vhost_net_virtqueue *nvq =
733 container_of(vq, struct vhost_net_virtqueue, vq);
734 struct vhost_poll *poll = n->poll + (nvq - n->vqs);
Arnd Bergmann28457ee2010-03-09 19:24:45 +0100735 struct socket *sock;
736
Asias He22fa90c2013-05-07 14:54:36 +0800737 sock = vq->private_data;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000738 if (!sock)
Jason Wang2b8b3282013-01-28 01:05:18 +0000739 return 0;
Jason Wang2b8b3282013-01-28 01:05:18 +0000740
Jason Wang70181d512013-04-10 20:50:48 +0000741 return vhost_poll_start(poll, sock->file);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000742}
743
744static struct socket *vhost_net_stop_vq(struct vhost_net *n,
745 struct vhost_virtqueue *vq)
746{
747 struct socket *sock;
748
749 mutex_lock(&vq->mutex);
Asias He22fa90c2013-05-07 14:54:36 +0800750 sock = vq->private_data;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000751 vhost_net_disable_vq(n, vq);
Asias He22fa90c2013-05-07 14:54:36 +0800752 vq->private_data = NULL;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000753 mutex_unlock(&vq->mutex);
754 return sock;
755}
756
757static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock,
758 struct socket **rx_sock)
759{
Asias He3ab2e422013-04-27 11:16:48 +0800760 *tx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_TX].vq);
761 *rx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_RX].vq);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000762}
763
764static void vhost_net_flush_vq(struct vhost_net *n, int index)
765{
766 vhost_poll_flush(n->poll + index);
Asias He3ab2e422013-04-27 11:16:48 +0800767 vhost_poll_flush(&n->vqs[index].vq.poll);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000768}
769
770static void vhost_net_flush(struct vhost_net *n)
771{
772 vhost_net_flush_vq(n, VHOST_NET_VQ_TX);
773 vhost_net_flush_vq(n, VHOST_NET_VQ_RX);
Asias He28394002013-04-27 15:07:46 +0800774 if (n->vqs[VHOST_NET_VQ_TX].ubufs) {
Asias He3ab2e422013-04-27 11:16:48 +0800775 mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
Michael S. Tsirkin1280c272012-12-04 00:17:14 +0200776 n->tx_flush = true;
Asias He3ab2e422013-04-27 11:16:48 +0800777 mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
Michael S. Tsirkin1280c272012-12-04 00:17:14 +0200778 /* Wait for all lower device DMAs done. */
Asias Hefe729a52013-05-06 16:38:24 +0800779 vhost_net_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].ubufs);
Asias He3ab2e422013-04-27 11:16:48 +0800780 mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
Michael S. Tsirkin1280c272012-12-04 00:17:14 +0200781 n->tx_flush = false;
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200782 atomic_set(&n->vqs[VHOST_NET_VQ_TX].ubufs->refcount, 1);
Asias He3ab2e422013-04-27 11:16:48 +0800783 mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
Michael S. Tsirkin1280c272012-12-04 00:17:14 +0200784 }
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000785}
786
787static int vhost_net_release(struct inode *inode, struct file *f)
788{
789 struct vhost_net *n = f->private_data;
790 struct socket *tx_sock;
791 struct socket *rx_sock;
792
793 vhost_net_stop(n, &tx_sock, &rx_sock);
794 vhost_net_flush(n);
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000795 vhost_dev_stop(&n->dev);
Michael S. Tsirkinea5d4042011-11-27 19:05:58 +0200796 vhost_dev_cleanup(&n->dev, false);
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300797 vhost_net_vq_reset(n);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000798 if (tx_sock)
Al Viro09aaacf2014-03-05 20:39:00 -0500799 sockfd_put(tx_sock);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000800 if (rx_sock)
Al Viro09aaacf2014-03-05 20:39:00 -0500801 sockfd_put(rx_sock);
Michael S. Tsirkinb0c057c2014-02-13 11:45:11 +0200802 /* Make sure no callbacks are outstanding */
803 synchronize_rcu_bh();
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000804 /* We do an extra flush before freeing memory,
805 * since jobs can re-queue themselves. */
806 vhost_net_flush(n);
Asias He3ab2e422013-04-27 11:16:48 +0800807 kfree(n->dev.vqs);
Romain Francoised04257b2014-06-12 10:42:34 +0200808 kvfree(n);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000809 return 0;
810}
811
812static struct socket *get_raw_socket(int fd)
813{
814 struct {
815 struct sockaddr_ll sa;
816 char buf[MAX_ADDR_LEN];
817 } uaddr;
818 int uaddr_len = sizeof uaddr, r;
819 struct socket *sock = sockfd_lookup(fd, &r);
Krishna Kumard47effe2011-03-01 17:06:37 +0530820
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000821 if (!sock)
822 return ERR_PTR(-ENOTSOCK);
823
824 /* Parameter checking */
825 if (sock->sk->sk_type != SOCK_RAW) {
826 r = -ESOCKTNOSUPPORT;
827 goto err;
828 }
829
830 r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa,
831 &uaddr_len, 0);
832 if (r)
833 goto err;
834
835 if (uaddr.sa.sll_family != AF_PACKET) {
836 r = -EPFNOSUPPORT;
837 goto err;
838 }
839 return sock;
840err:
Al Viro09aaacf2014-03-05 20:39:00 -0500841 sockfd_put(sock);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000842 return ERR_PTR(r);
843}
844
Arnd Bergmann501c7742010-02-18 05:46:50 +0000845static struct socket *get_tap_socket(int fd)
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000846{
847 struct file *file = fget(fd);
848 struct socket *sock;
Krishna Kumard47effe2011-03-01 17:06:37 +0530849
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000850 if (!file)
851 return ERR_PTR(-EBADF);
852 sock = tun_get_socket(file);
Arnd Bergmann501c7742010-02-18 05:46:50 +0000853 if (!IS_ERR(sock))
854 return sock;
855 sock = macvtap_get_socket(file);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000856 if (IS_ERR(sock))
857 fput(file);
858 return sock;
859}
860
861static struct socket *get_socket(int fd)
862{
863 struct socket *sock;
Krishna Kumard47effe2011-03-01 17:06:37 +0530864
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000865 /* special case to disable backend */
866 if (fd == -1)
867 return NULL;
868 sock = get_raw_socket(fd);
869 if (!IS_ERR(sock))
870 return sock;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000871 sock = get_tap_socket(fd);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000872 if (!IS_ERR(sock))
873 return sock;
874 return ERR_PTR(-ENOTSOCK);
875}
876
877static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
878{
879 struct socket *sock, *oldsock;
880 struct vhost_virtqueue *vq;
Asias He28394002013-04-27 15:07:46 +0800881 struct vhost_net_virtqueue *nvq;
Asias Hefe729a52013-05-06 16:38:24 +0800882 struct vhost_net_ubuf_ref *ubufs, *oldubufs = NULL;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000883 int r;
884
885 mutex_lock(&n->dev.mutex);
886 r = vhost_dev_check_owner(&n->dev);
887 if (r)
888 goto err;
889
890 if (index >= VHOST_NET_VQ_MAX) {
891 r = -ENOBUFS;
892 goto err;
893 }
Asias He3ab2e422013-04-27 11:16:48 +0800894 vq = &n->vqs[index].vq;
Asias He28394002013-04-27 15:07:46 +0800895 nvq = &n->vqs[index];
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000896 mutex_lock(&vq->mutex);
897
898 /* Verify that ring has been setup correctly. */
899 if (!vhost_vq_access_ok(vq)) {
900 r = -EFAULT;
Jeff Dike1dace8c2010-03-04 16:10:14 -0500901 goto err_vq;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000902 }
903 sock = get_socket(fd);
904 if (IS_ERR(sock)) {
905 r = PTR_ERR(sock);
Jeff Dike1dace8c2010-03-04 16:10:14 -0500906 goto err_vq;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000907 }
908
909 /* start polling new socket */
Asias He22fa90c2013-05-07 14:54:36 +0800910 oldsock = vq->private_data;
David S. Miller11fe8832010-07-20 18:25:24 -0700911 if (sock != oldsock) {
Asias Hefe729a52013-05-06 16:38:24 +0800912 ubufs = vhost_net_ubuf_alloc(vq,
913 sock && vhost_sock_zcopy(sock));
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000914 if (IS_ERR(ubufs)) {
915 r = PTR_ERR(ubufs);
916 goto err_ubufs;
917 }
Jason Wang692a9982013-01-28 01:05:17 +0000918
Krishna Kumard47effe2011-03-01 17:06:37 +0530919 vhost_net_disable_vq(n, vq);
Asias He22fa90c2013-05-07 14:54:36 +0800920 vq->private_data = sock;
Jason Wangf59281d2011-06-21 18:04:27 +0800921 r = vhost_init_used(vq);
922 if (r)
Jason Wang692a9982013-01-28 01:05:17 +0000923 goto err_used;
Jason Wang2b8b3282013-01-28 01:05:18 +0000924 r = vhost_net_enable_vq(n, vq);
925 if (r)
926 goto err_used;
Jason Wang692a9982013-01-28 01:05:17 +0000927
Asias He28394002013-04-27 15:07:46 +0800928 oldubufs = nvq->ubufs;
929 nvq->ubufs = ubufs;
Michael S. Tsirkin64e9a9b2012-12-03 07:31:51 +0000930
931 n->tx_packets = 0;
932 n->tx_zcopy_err = 0;
Michael S. Tsirkin1280c272012-12-04 00:17:14 +0200933 n->tx_flush = false;
Jeff Dikedd1f4072010-03-04 16:10:14 -0500934 }
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000935
Michael S. Tsirkin1680e902010-07-15 15:19:12 +0300936 mutex_unlock(&vq->mutex);
937
Michael S. Tsirkinc047e5f2011-07-20 13:41:31 +0300938 if (oldubufs) {
Michael S. Tsirkinc38e39c2013-06-25 17:29:46 +0300939 vhost_net_ubuf_put_wait_and_free(oldubufs);
Michael S. Tsirkinc047e5f2011-07-20 13:41:31 +0300940 mutex_lock(&vq->mutex);
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000941 vhost_zerocopy_signal_used(n, vq);
Michael S. Tsirkinc047e5f2011-07-20 13:41:31 +0300942 mutex_unlock(&vq->mutex);
943 }
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000944
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000945 if (oldsock) {
946 vhost_net_flush_vq(n, index);
Al Viro09aaacf2014-03-05 20:39:00 -0500947 sockfd_put(oldsock);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000948 }
Jeff Dike1dace8c2010-03-04 16:10:14 -0500949
Michael S. Tsirkin1680e902010-07-15 15:19:12 +0300950 mutex_unlock(&n->dev.mutex);
951 return 0;
952
Jason Wang692a9982013-01-28 01:05:17 +0000953err_used:
Asias He22fa90c2013-05-07 14:54:36 +0800954 vq->private_data = oldsock;
Jason Wang692a9982013-01-28 01:05:17 +0000955 vhost_net_enable_vq(n, vq);
956 if (ubufs)
Michael S. Tsirkinc38e39c2013-06-25 17:29:46 +0300957 vhost_net_ubuf_put_wait_and_free(ubufs);
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000958err_ubufs:
Al Viro09aaacf2014-03-05 20:39:00 -0500959 sockfd_put(sock);
Jeff Dike1dace8c2010-03-04 16:10:14 -0500960err_vq:
961 mutex_unlock(&vq->mutex);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000962err:
963 mutex_unlock(&n->dev.mutex);
964 return r;
965}
966
967static long vhost_net_reset_owner(struct vhost_net *n)
968{
969 struct socket *tx_sock = NULL;
970 struct socket *rx_sock = NULL;
971 long err;
Michael S. Tsirkin150b9e52013-04-28 17:12:08 +0300972 struct vhost_memory *memory;
Krishna Kumard47effe2011-03-01 17:06:37 +0530973
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000974 mutex_lock(&n->dev.mutex);
975 err = vhost_dev_check_owner(&n->dev);
976 if (err)
977 goto done;
Michael S. Tsirkin150b9e52013-04-28 17:12:08 +0300978 memory = vhost_dev_reset_owner_prepare();
979 if (!memory) {
980 err = -ENOMEM;
981 goto done;
982 }
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000983 vhost_net_stop(n, &tx_sock, &rx_sock);
984 vhost_net_flush(n);
Michael S. Tsirkin150b9e52013-04-28 17:12:08 +0300985 vhost_dev_reset_owner(&n->dev, memory);
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300986 vhost_net_vq_reset(n);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000987done:
988 mutex_unlock(&n->dev.mutex);
989 if (tx_sock)
Al Viro09aaacf2014-03-05 20:39:00 -0500990 sockfd_put(tx_sock);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000991 if (rx_sock)
Al Viro09aaacf2014-03-05 20:39:00 -0500992 sockfd_put(rx_sock);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000993 return err;
994}
995
996static int vhost_net_set_features(struct vhost_net *n, u64 features)
997{
David Stevens8dd014a2010-07-27 18:52:21 +0300998 size_t vhost_hlen, sock_hlen, hdr_len;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +0000999 int i;
David Stevens8dd014a2010-07-27 18:52:21 +03001000
Michael S. Tsirkine4fca7d2014-10-24 14:23:52 +03001001 hdr_len = (features & ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
1002 (1ULL << VIRTIO_F_VERSION_1))) ?
David Stevens8dd014a2010-07-27 18:52:21 +03001003 sizeof(struct virtio_net_hdr_mrg_rxbuf) :
1004 sizeof(struct virtio_net_hdr);
1005 if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) {
1006 /* vhost provides vnet_hdr */
1007 vhost_hlen = hdr_len;
1008 sock_hlen = 0;
1009 } else {
1010 /* socket provides vnet_hdr */
1011 vhost_hlen = 0;
1012 sock_hlen = hdr_len;
1013 }
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001014 mutex_lock(&n->dev.mutex);
1015 if ((features & (1 << VHOST_F_LOG_ALL)) &&
1016 !vhost_log_access_ok(&n->dev)) {
1017 mutex_unlock(&n->dev.mutex);
1018 return -EFAULT;
1019 }
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001020 for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
Asias He3ab2e422013-04-27 11:16:48 +08001021 mutex_lock(&n->vqs[i].vq.mutex);
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03001022 n->vqs[i].vq.acked_features = features;
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +03001023 n->vqs[i].vhost_hlen = vhost_hlen;
1024 n->vqs[i].sock_hlen = sock_hlen;
Asias He3ab2e422013-04-27 11:16:48 +08001025 mutex_unlock(&n->vqs[i].vq.mutex);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001026 }
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001027 mutex_unlock(&n->dev.mutex);
1028 return 0;
1029}
1030
Asias Heb1ad8492013-05-06 11:16:00 +08001031static long vhost_net_set_owner(struct vhost_net *n)
1032{
1033 int r;
1034
1035 mutex_lock(&n->dev.mutex);
Michael S. Tsirkin05c05352013-06-06 15:20:39 +03001036 if (vhost_dev_has_owner(&n->dev)) {
1037 r = -EBUSY;
1038 goto out;
1039 }
Asias Heb1ad8492013-05-06 11:16:00 +08001040 r = vhost_net_set_ubuf_info(n);
1041 if (r)
1042 goto out;
1043 r = vhost_dev_set_owner(&n->dev);
1044 if (r)
1045 vhost_net_clear_ubuf_info(n);
1046 vhost_net_flush(n);
1047out:
1048 mutex_unlock(&n->dev.mutex);
1049 return r;
1050}
1051
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001052static long vhost_net_ioctl(struct file *f, unsigned int ioctl,
1053 unsigned long arg)
1054{
1055 struct vhost_net *n = f->private_data;
1056 void __user *argp = (void __user *)arg;
1057 u64 __user *featurep = argp;
1058 struct vhost_vring_file backend;
1059 u64 features;
1060 int r;
Krishna Kumard47effe2011-03-01 17:06:37 +05301061
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001062 switch (ioctl) {
1063 case VHOST_NET_SET_BACKEND:
Takuya Yoshikawad3553a52010-05-27 19:01:58 +09001064 if (copy_from_user(&backend, argp, sizeof backend))
1065 return -EFAULT;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001066 return vhost_net_set_backend(n, backend.index, backend.fd);
1067 case VHOST_GET_FEATURES:
Stefan Hajnoczi0dd05a32012-07-21 06:55:36 +00001068 features = VHOST_NET_FEATURES;
Takuya Yoshikawad3553a52010-05-27 19:01:58 +09001069 if (copy_to_user(featurep, &features, sizeof features))
1070 return -EFAULT;
1071 return 0;
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001072 case VHOST_SET_FEATURES:
Takuya Yoshikawad3553a52010-05-27 19:01:58 +09001073 if (copy_from_user(&features, featurep, sizeof features))
1074 return -EFAULT;
Stefan Hajnoczi0dd05a32012-07-21 06:55:36 +00001075 if (features & ~VHOST_NET_FEATURES)
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001076 return -EOPNOTSUPP;
1077 return vhost_net_set_features(n, features);
1078 case VHOST_RESET_OWNER:
1079 return vhost_net_reset_owner(n);
Asias Heb1ad8492013-05-06 11:16:00 +08001080 case VHOST_SET_OWNER:
1081 return vhost_net_set_owner(n);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001082 default:
1083 mutex_lock(&n->dev.mutex);
Michael S. Tsirkin935cdee2012-12-06 14:03:34 +02001084 r = vhost_dev_ioctl(&n->dev, ioctl, argp);
1085 if (r == -ENOIOCTLCMD)
1086 r = vhost_vring_ioctl(&n->dev, ioctl, argp);
1087 else
1088 vhost_net_flush(n);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001089 mutex_unlock(&n->dev.mutex);
1090 return r;
1091 }
1092}
1093
1094#ifdef CONFIG_COMPAT
1095static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl,
1096 unsigned long arg)
1097{
1098 return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1099}
1100#endif
1101
Tobias Klauser373a83a2010-05-17 15:12:49 +02001102static const struct file_operations vhost_net_fops = {
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001103 .owner = THIS_MODULE,
1104 .release = vhost_net_release,
1105 .unlocked_ioctl = vhost_net_ioctl,
1106#ifdef CONFIG_COMPAT
1107 .compat_ioctl = vhost_net_compat_ioctl,
1108#endif
1109 .open = vhost_net_open,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001110 .llseek = noop_llseek,
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001111};
1112
1113static struct miscdevice vhost_net_misc = {
stephen hemminger7c7c7f02012-01-11 19:30:38 +00001114 .minor = VHOST_NET_MINOR,
1115 .name = "vhost-net",
1116 .fops = &vhost_net_fops,
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001117};
1118
Christoph Hellwiga8d37822010-04-13 14:11:25 -04001119static int vhost_net_init(void)
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001120{
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +00001121 if (experimental_zcopytx)
Asias Hefe729a52013-05-06 16:38:24 +08001122 vhost_net_enable_zcopy(VHOST_NET_VQ_TX);
Tejun Heoc23f34452010-06-02 20:40:00 +02001123 return misc_register(&vhost_net_misc);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001124}
1125module_init(vhost_net_init);
1126
Christoph Hellwiga8d37822010-04-13 14:11:25 -04001127static void vhost_net_exit(void)
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001128{
1129 misc_deregister(&vhost_net_misc);
Michael S. Tsirkin3a4d5c94e2010-01-14 06:17:27 +00001130}
1131module_exit(vhost_net_exit);
1132
1133MODULE_VERSION("0.0.1");
1134MODULE_LICENSE("GPL v2");
1135MODULE_AUTHOR("Michael S. Tsirkin");
1136MODULE_DESCRIPTION("Host kernel accelerator for virtio net");
stephen hemminger7c7c7f02012-01-11 19:30:38 +00001137MODULE_ALIAS_MISCDEV(VHOST_NET_MINOR);
1138MODULE_ALIAS("devname:vhost-net");