blob: 5dc128a8da83e5c724dbd1d8748fa0db2a300d7a [file] [log] [blame]
Michael S. Tsirkin3a4d5c942010-01-14 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. Tsirkin3a4d5c942010-01-14 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. Tsirkin3a4d5c942010-01-14 06:17:27 +000016#include <linux/mutex.h>
17#include <linux/workqueue.h>
Michael S. Tsirkin3a4d5c942010-01-14 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. Tsirkin3a4d5c942010-01-14 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. Tsirkin3a4d5c942010-01-14 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. Tsirkin3a4d5c942010-01-14 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. Tsirkin3a4d5c942010-01-14 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) |
Jason Wang6b1e6cc2016-06-23 02:04:32 -040064 (1ULL << VIRTIO_NET_F_MRG_RXBUF) |
65 (1ULL << VIRTIO_F_IOMMU_PLATFORM)
Asias He8570a6e2013-05-06 16:38:20 +080066};
67
68enum {
Michael S. Tsirkin3a4d5c942010-01-14 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. Tsirkin3a4d5c942010-01-14 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. Tsirkin3a4d5c942010-01-14 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. Tsirkin3a4d5c942010-01-14 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
Jason Wang03088132016-03-04 06:24:53 -0500291static inline unsigned long busy_clock(void)
292{
293 return local_clock() >> 10;
294}
295
296static bool vhost_can_busy_poll(struct vhost_dev *dev,
297 unsigned long endtime)
298{
299 return likely(!need_resched()) &&
300 likely(!time_after(busy_clock(), endtime)) &&
301 likely(!signal_pending(current)) &&
302 !vhost_has_work(dev);
303}
304
Jason Wang8241a1e2016-06-01 01:56:33 -0400305static void vhost_net_disable_vq(struct vhost_net *n,
306 struct vhost_virtqueue *vq)
307{
308 struct vhost_net_virtqueue *nvq =
309 container_of(vq, struct vhost_net_virtqueue, vq);
310 struct vhost_poll *poll = n->poll + (nvq - n->vqs);
311 if (!vq->private_data)
312 return;
313 vhost_poll_stop(poll);
314}
315
316static int vhost_net_enable_vq(struct vhost_net *n,
317 struct vhost_virtqueue *vq)
318{
319 struct vhost_net_virtqueue *nvq =
320 container_of(vq, struct vhost_net_virtqueue, vq);
321 struct vhost_poll *poll = n->poll + (nvq - n->vqs);
322 struct socket *sock;
323
324 sock = vq->private_data;
325 if (!sock)
326 return 0;
327
328 return vhost_poll_start(poll, sock->file);
329}
330
Jason Wang03088132016-03-04 06:24:53 -0500331static int vhost_net_tx_get_vq_desc(struct vhost_net *net,
332 struct vhost_virtqueue *vq,
333 struct iovec iov[], unsigned int iov_size,
334 unsigned int *out_num, unsigned int *in_num)
335{
336 unsigned long uninitialized_var(endtime);
337 int r = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400338 out_num, in_num, NULL, NULL);
Jason Wang03088132016-03-04 06:24:53 -0500339
340 if (r == vq->num && vq->busyloop_timeout) {
341 preempt_disable();
342 endtime = busy_clock() + vq->busyloop_timeout;
343 while (vhost_can_busy_poll(vq->dev, endtime) &&
344 vhost_vq_avail_empty(vq->dev, vq))
345 cpu_relax_lowlatency();
346 preempt_enable();
347 r = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400348 out_num, in_num, NULL, NULL);
Jason Wang03088132016-03-04 06:24:53 -0500349 }
350
351 return r;
352}
353
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000354/* Expects to be always run from workqueue - which acts as
355 * read-size critical section for our kind of RCU. */
356static void handle_tx(struct vhost_net *net)
357{
Asias He28394002013-04-27 15:07:46 +0800358 struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX];
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300359 struct vhost_virtqueue *vq = &nvq->vq;
Al Viro98a527a2014-12-10 15:00:58 -0500360 unsigned out, in;
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +0300361 int head;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000362 struct msghdr msg = {
363 .msg_name = NULL,
364 .msg_namelen = 0,
365 .msg_control = NULL,
366 .msg_controllen = 0,
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000367 .msg_flags = MSG_DONTWAIT,
368 };
369 size_t len, total_len = 0;
Jason Wang70181d512013-04-10 20:50:48 +0000370 int err;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000371 size_t hdr_size;
Arnd Bergmann28457ee2010-03-09 19:24:45 +0100372 struct socket *sock;
Asias Hefe729a52013-05-06 16:38:24 +0800373 struct vhost_net_ubuf_ref *uninitialized_var(ubufs);
Michael S. Tsirkincedb9bdc2012-12-06 17:00:18 +0200374 bool zcopy, zcopy_used;
Arnd Bergmann28457ee2010-03-09 19:24:45 +0100375
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000376 mutex_lock(&vq->mutex);
Asias He2e26af72013-05-07 14:54:33 +0800377 sock = vq->private_data;
378 if (!sock)
379 goto out;
380
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400381 if (!vq_iotlb_prefetch(vq))
382 goto out;
383
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +0300384 vhost_disable_notify(&net->dev, vq);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000385
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300386 hdr_size = nvq->vhost_hlen;
Asias He28394002013-04-27 15:07:46 +0800387 zcopy = nvq->ubufs;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000388
389 for (;;) {
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000390 /* Release DMAs done buffers first */
391 if (zcopy)
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000392 vhost_zerocopy_signal_used(net, vq);
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000393
Jason Wangf7c6be42013-09-02 16:41:01 +0800394 /* If more outstanding DMAs, queue the work.
395 * Handle upend_idx wrap around
396 */
397 if (unlikely((nvq->upend_idx + vq->num - VHOST_MAX_PEND)
398 % UIO_MAXIOV == nvq->done_idx))
399 break;
400
Jason Wang03088132016-03-04 06:24:53 -0500401 head = vhost_net_tx_get_vq_desc(net, vq, vq->iov,
402 ARRAY_SIZE(vq->iov),
403 &out, &in);
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +0300404 /* On error, stop handling until the next kick. */
Michael S. Tsirkin7b3384f2010-07-01 18:40:12 +0300405 if (unlikely(head < 0))
Michael S. Tsirkind5675bd2010-06-24 16:59:59 +0300406 break;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000407 /* Nothing new? Wait for eventfd to tell us they refilled. */
408 if (head == vq->num) {
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +0300409 if (unlikely(vhost_enable_notify(&net->dev, vq))) {
410 vhost_disable_notify(&net->dev, vq);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000411 continue;
412 }
413 break;
414 }
415 if (in) {
416 vq_err(vq, "Unexpected descriptor format for TX: "
417 "out %d, int %d\n", out, in);
418 break;
419 }
420 /* Skip header. TODO: support TSO. */
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000421 len = iov_length(vq->iov, out);
Al Viroc0371da2014-11-24 10:42:55 -0500422 iov_iter_init(&msg.msg_iter, WRITE, vq->iov, out, len);
Al Viro98a527a2014-12-10 15:00:58 -0500423 iov_iter_advance(&msg.msg_iter, hdr_size);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000424 /* Sanity check */
Al Viro01e97e62014-12-15 21:39:31 -0500425 if (!msg_data_left(&msg)) {
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000426 vq_err(vq, "Unexpected header len for TX: "
427 "%zd expected %zd\n",
Al Viro98a527a2014-12-10 15:00:58 -0500428 len, hdr_size);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000429 break;
430 }
Al Viro01e97e62014-12-15 21:39:31 -0500431 len = msg_data_left(&msg);
Jason Wangce21a022013-09-02 16:40:59 +0800432
433 zcopy_used = zcopy && len >= VHOST_GOODCOPY_LEN
434 && (nvq->upend_idx + 1) % UIO_MAXIOV !=
435 nvq->done_idx
436 && vhost_net_tx_select_zcopy(net);
Michael S. Tsirkincedb9bdc2012-12-06 17:00:18 +0200437
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000438 /* use msg_control to pass vhost zerocopy ubuf info to skb */
Michael S. Tsirkincedb9bdc2012-12-06 17:00:18 +0200439 if (zcopy_used) {
Jason Wangce21a022013-09-02 16:40:59 +0800440 struct ubuf_info *ubuf;
441 ubuf = nvq->ubuf_info + nvq->upend_idx;
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000442
Michael S. Tsirkin8b386942014-10-24 14:19:48 +0300443 vq->heads[nvq->upend_idx].id = cpu_to_vhost32(vq, head);
Jason Wangce21a022013-09-02 16:40:59 +0800444 vq->heads[nvq->upend_idx].len = VHOST_DMA_IN_PROGRESS;
445 ubuf->callback = vhost_zerocopy_callback;
446 ubuf->ctx = nvq->ubufs;
447 ubuf->desc = nvq->upend_idx;
448 msg.msg_control = ubuf;
449 msg.msg_controllen = sizeof(ubuf);
450 ubufs = nvq->ubufs;
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200451 atomic_inc(&ubufs->refcount);
Asias He28394002013-04-27 15:07:46 +0800452 nvq->upend_idx = (nvq->upend_idx + 1) % UIO_MAXIOV;
Jason Wangce21a022013-09-02 16:40:59 +0800453 } else {
Jason Wang4364d5f2013-06-05 15:40:46 +0800454 msg.msg_control = NULL;
Jason Wangce21a022013-09-02 16:40:59 +0800455 ubufs = NULL;
456 }
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000457 /* TODO: Check specific error and bomb out unless ENOBUFS? */
Ying Xue1b784142015-03-02 15:37:48 +0800458 err = sock->ops->sendmsg(sock, &msg, len);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000459 if (unlikely(err < 0)) {
Michael S. Tsirkincedb9bdc2012-12-06 17:00:18 +0200460 if (zcopy_used) {
Jason Wangce21a022013-09-02 16:40:59 +0800461 vhost_net_ubuf_put(ubufs);
Asias He28394002013-04-27 15:07:46 +0800462 nvq->upend_idx = ((unsigned)nvq->upend_idx - 1)
463 % UIO_MAXIOV;
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000464 }
David Stevens8dd014a2010-07-27 18:52:21 +0300465 vhost_discard_vq_desc(vq, 1);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000466 break;
467 }
468 if (err != len)
Michael S. Tsirkin95c0ec62010-06-24 17:10:25 +0300469 pr_debug("Truncated TX packet: "
470 " len %d != %zd\n", err, len);
Michael S. Tsirkincedb9bdc2012-12-06 17:00:18 +0200471 if (!zcopy_used)
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +0000472 vhost_add_used_and_signal(&net->dev, vq, head, 0);
Jason Wangc8fb2172012-05-02 11:42:41 +0800473 else
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000474 vhost_zerocopy_signal_used(net, vq);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000475 total_len += len;
Michael S. Tsirkineaae8132012-11-01 09:16:51 +0000476 vhost_net_tx_packet(net);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000477 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
478 vhost_poll_queue(&vq->poll);
479 break;
480 }
481 }
Asias He2e26af72013-05-07 14:54:33 +0800482out:
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000483 mutex_unlock(&vq->mutex);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000484}
485
David Stevens8dd014a2010-07-27 18:52:21 +0300486static int peek_head_len(struct sock *sk)
487{
Jason Wang1576d982016-06-30 14:45:36 +0800488 struct socket *sock = sk->sk_socket;
David Stevens8dd014a2010-07-27 18:52:21 +0300489 struct sk_buff *head;
490 int len = 0;
Jason Wang783e3982011-01-17 16:11:17 +0800491 unsigned long flags;
David Stevens8dd014a2010-07-27 18:52:21 +0300492
Jason Wang1576d982016-06-30 14:45:36 +0800493 if (sock->ops->peek_len)
494 return sock->ops->peek_len(sock);
495
Jason Wang783e3982011-01-17 16:11:17 +0800496 spin_lock_irqsave(&sk->sk_receive_queue.lock, flags);
David Stevens8dd014a2010-07-27 18:52:21 +0300497 head = skb_peek(&sk->sk_receive_queue);
Basil Gorc53cff5e2012-05-03 22:55:23 +0000498 if (likely(head)) {
David Stevens8dd014a2010-07-27 18:52:21 +0300499 len = head->len;
Jiri Pirkodf8a39de2015-01-13 17:13:44 +0100500 if (skb_vlan_tag_present(head))
Basil Gorc53cff5e2012-05-03 22:55:23 +0000501 len += VLAN_HLEN;
502 }
503
Jason Wang783e3982011-01-17 16:11:17 +0800504 spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags);
David Stevens8dd014a2010-07-27 18:52:21 +0300505 return len;
506}
507
Jason Wang1576d982016-06-30 14:45:36 +0800508static int sk_has_rx_data(struct sock *sk)
509{
510 struct socket *sock = sk->sk_socket;
511
512 if (sock->ops->peek_len)
513 return sock->ops->peek_len(sock);
514
515 return skb_queue_empty(&sk->sk_receive_queue);
516}
517
Jason Wang03088132016-03-04 06:24:53 -0500518static int vhost_net_rx_peek_head_len(struct vhost_net *net, struct sock *sk)
519{
520 struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX];
521 struct vhost_virtqueue *vq = &nvq->vq;
522 unsigned long uninitialized_var(endtime);
523 int len = peek_head_len(sk);
524
525 if (!len && vq->busyloop_timeout) {
526 /* Both tx vq and rx socket were polled here */
527 mutex_lock(&vq->mutex);
528 vhost_disable_notify(&net->dev, vq);
529
530 preempt_disable();
531 endtime = busy_clock() + vq->busyloop_timeout;
532
533 while (vhost_can_busy_poll(&net->dev, endtime) &&
Jason Wang1576d982016-06-30 14:45:36 +0800534 !sk_has_rx_data(sk) &&
Jason Wang03088132016-03-04 06:24:53 -0500535 vhost_vq_avail_empty(&net->dev, vq))
536 cpu_relax_lowlatency();
537
538 preempt_enable();
539
540 if (vhost_enable_notify(&net->dev, vq))
541 vhost_poll_queue(&vq->poll);
542 mutex_unlock(&vq->mutex);
543
544 len = peek_head_len(sk);
545 }
546
547 return len;
548}
549
David Stevens8dd014a2010-07-27 18:52:21 +0300550/* This is a multi-buffer version of vhost_get_desc, that works if
551 * vq has read descriptors only.
552 * @vq - the relevant virtqueue
553 * @datalen - data length we'll be reading
554 * @iovcount - returned count of io vectors we fill
555 * @log - vhost log
556 * @log_num - log offset
Jason Wang94249362011-01-17 16:11:08 +0800557 * @quota - headcount quota, 1 for big buffer
David Stevens8dd014a2010-07-27 18:52:21 +0300558 * returns number of buffer heads allocated, negative on error
559 */
560static int get_rx_bufs(struct vhost_virtqueue *vq,
561 struct vring_used_elem *heads,
562 int datalen,
563 unsigned *iovcount,
564 struct vhost_log *log,
Jason Wang94249362011-01-17 16:11:08 +0800565 unsigned *log_num,
566 unsigned int quota)
David Stevens8dd014a2010-07-27 18:52:21 +0300567{
568 unsigned int out, in;
569 int seg = 0;
570 int headcount = 0;
571 unsigned d;
572 int r, nlogs = 0;
Michael S. Tsirkin8b386942014-10-24 14:19:48 +0300573 /* len is always initialized before use since we are always called with
574 * datalen > 0.
575 */
576 u32 uninitialized_var(len);
David Stevens8dd014a2010-07-27 18:52:21 +0300577
Jason Wang94249362011-01-17 16:11:08 +0800578 while (datalen > 0 && headcount < quota) {
Jason Wange0e9b402010-09-14 23:53:05 +0800579 if (unlikely(seg >= UIO_MAXIOV)) {
David Stevens8dd014a2010-07-27 18:52:21 +0300580 r = -ENOBUFS;
581 goto err;
582 }
Michael S. Tsirkin47283be2014-06-05 15:20:27 +0300583 r = vhost_get_vq_desc(vq, vq->iov + seg,
David Stevens8dd014a2010-07-27 18:52:21 +0300584 ARRAY_SIZE(vq->iov) - seg, &out,
585 &in, log, log_num);
Michael S. Tsirkina39ee442014-03-27 12:53:37 +0200586 if (unlikely(r < 0))
587 goto err;
588
589 d = r;
David Stevens8dd014a2010-07-27 18:52:21 +0300590 if (d == vq->num) {
591 r = 0;
592 goto err;
593 }
594 if (unlikely(out || in <= 0)) {
595 vq_err(vq, "unexpected descriptor format for RX: "
596 "out %d, in %d\n", out, in);
597 r = -EINVAL;
598 goto err;
599 }
600 if (unlikely(log)) {
601 nlogs += *log_num;
602 log += *log_num;
603 }
Michael S. Tsirkin8b386942014-10-24 14:19:48 +0300604 heads[headcount].id = cpu_to_vhost32(vq, d);
605 len = iov_length(vq->iov + seg, in);
606 heads[headcount].len = cpu_to_vhost32(vq, len);
607 datalen -= len;
David Stevens8dd014a2010-07-27 18:52:21 +0300608 ++headcount;
609 seg += in;
610 }
Michael S. Tsirkin99975cc2015-01-07 10:51:00 +0200611 heads[headcount - 1].len = cpu_to_vhost32(vq, len + datalen);
David Stevens8dd014a2010-07-27 18:52:21 +0300612 *iovcount = seg;
613 if (unlikely(log))
614 *log_num = nlogs;
Michael S. Tsirkind8316f32014-03-27 12:00:26 +0200615
616 /* Detect overrun */
617 if (unlikely(datalen > 0)) {
618 r = UIO_MAXIOV + 1;
619 goto err;
620 }
David Stevens8dd014a2010-07-27 18:52:21 +0300621 return headcount;
622err:
623 vhost_discard_vq_desc(vq, headcount);
624 return r;
625}
626
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000627/* Expects to be always run from workqueue - which acts as
628 * read-size critical section for our kind of RCU. */
Jason Wang94249362011-01-17 16:11:08 +0800629static void handle_rx(struct vhost_net *net)
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000630{
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300631 struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_RX];
632 struct vhost_virtqueue *vq = &nvq->vq;
David Stevens8dd014a2010-07-27 18:52:21 +0300633 unsigned uninitialized_var(in), log;
634 struct vhost_log *vq_log;
635 struct msghdr msg = {
636 .msg_name = NULL,
637 .msg_namelen = 0,
638 .msg_control = NULL, /* FIXME: get and handle RX aux data. */
639 .msg_controllen = 0,
David Stevens8dd014a2010-07-27 18:52:21 +0300640 .msg_flags = MSG_DONTWAIT,
641 };
Jason Wang0960b642015-02-15 16:35:17 +0800642 struct virtio_net_hdr hdr = {
643 .flags = 0,
644 .gso_type = VIRTIO_NET_HDR_GSO_NONE
David Stevens8dd014a2010-07-27 18:52:21 +0300645 };
David Stevens8dd014a2010-07-27 18:52:21 +0300646 size_t total_len = 0;
Michael S. Tsirkin910a5782012-10-24 20:37:51 +0200647 int err, mergeable;
648 s16 headcount;
David Stevens8dd014a2010-07-27 18:52:21 +0300649 size_t vhost_hlen, sock_hlen;
650 size_t vhost_len, sock_len;
Asias He2e26af72013-05-07 14:54:33 +0800651 struct socket *sock;
Al Viroba7438ae2014-12-10 15:51:28 -0500652 struct iov_iter fixup;
Jason Wang0960b642015-02-15 16:35:17 +0800653 __virtio16 num_buffers;
David Stevens8dd014a2010-07-27 18:52:21 +0300654
David Stevens8dd014a2010-07-27 18:52:21 +0300655 mutex_lock(&vq->mutex);
Asias He2e26af72013-05-07 14:54:33 +0800656 sock = vq->private_data;
657 if (!sock)
658 goto out;
Jason Wang6b1e6cc2016-06-23 02:04:32 -0400659
660 if (!vq_iotlb_prefetch(vq))
661 goto out;
662
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +0300663 vhost_disable_notify(&net->dev, vq);
Jason Wang8241a1e2016-06-01 01:56:33 -0400664 vhost_net_disable_vq(net, vq);
Asias He2e26af72013-05-07 14:54:33 +0800665
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300666 vhost_hlen = nvq->vhost_hlen;
667 sock_hlen = nvq->sock_hlen;
David Stevens8dd014a2010-07-27 18:52:21 +0300668
Michael S. Tsirkinea16c512014-06-05 15:20:23 +0300669 vq_log = unlikely(vhost_has_feature(vq, VHOST_F_LOG_ALL)) ?
David Stevens8dd014a2010-07-27 18:52:21 +0300670 vq->log : NULL;
Michael S. Tsirkinea16c512014-06-05 15:20:23 +0300671 mergeable = vhost_has_feature(vq, VIRTIO_NET_F_MRG_RXBUF);
David Stevens8dd014a2010-07-27 18:52:21 +0300672
Jason Wang03088132016-03-04 06:24:53 -0500673 while ((sock_len = vhost_net_rx_peek_head_len(net, sock->sk))) {
David Stevens8dd014a2010-07-27 18:52:21 +0300674 sock_len += sock_hlen;
675 vhost_len = sock_len + vhost_hlen;
676 headcount = get_rx_bufs(vq, vq->heads, vhost_len,
Jason Wang94249362011-01-17 16:11:08 +0800677 &in, vq_log, &log,
678 likely(mergeable) ? UIO_MAXIOV : 1);
David Stevens8dd014a2010-07-27 18:52:21 +0300679 /* On error, stop handling until the next kick. */
680 if (unlikely(headcount < 0))
Jason Wang8241a1e2016-06-01 01:56:33 -0400681 goto out;
Michael S. Tsirkind8316f32014-03-27 12:00:26 +0200682 /* On overrun, truncate and discard */
683 if (unlikely(headcount > UIO_MAXIOV)) {
Al Viroc0371da2014-11-24 10:42:55 -0500684 iov_iter_init(&msg.msg_iter, READ, vq->iov, 1, 1);
Ying Xue1b784142015-03-02 15:37:48 +0800685 err = sock->ops->recvmsg(sock, &msg,
Michael S. Tsirkind8316f32014-03-27 12:00:26 +0200686 1, MSG_DONTWAIT | MSG_TRUNC);
687 pr_debug("Discarded rx packet: len %zd\n", sock_len);
688 continue;
689 }
David Stevens8dd014a2010-07-27 18:52:21 +0300690 /* OK, now we need to know about added descriptors. */
691 if (!headcount) {
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +0300692 if (unlikely(vhost_enable_notify(&net->dev, vq))) {
David Stevens8dd014a2010-07-27 18:52:21 +0300693 /* They have slipped one in as we were
694 * doing that: check again. */
Michael S. Tsirkin8ea8cf82011-05-20 02:10:54 +0300695 vhost_disable_notify(&net->dev, vq);
David Stevens8dd014a2010-07-27 18:52:21 +0300696 continue;
697 }
698 /* Nothing new? Wait for eventfd to tell us
699 * they refilled. */
Jason Wang8241a1e2016-06-01 01:56:33 -0400700 goto out;
David Stevens8dd014a2010-07-27 18:52:21 +0300701 }
702 /* We don't need to be notified again. */
Al Viroba7438ae2014-12-10 15:51:28 -0500703 iov_iter_init(&msg.msg_iter, READ, vq->iov, in, vhost_len);
704 fixup = msg.msg_iter;
705 if (unlikely((vhost_hlen))) {
706 /* We will supply the header ourselves
707 * TODO: support TSO.
708 */
709 iov_iter_advance(&msg.msg_iter, vhost_hlen);
Al Viroba7438ae2014-12-10 15:51:28 -0500710 }
Ying Xue1b784142015-03-02 15:37:48 +0800711 err = sock->ops->recvmsg(sock, &msg,
David Stevens8dd014a2010-07-27 18:52:21 +0300712 sock_len, MSG_DONTWAIT | MSG_TRUNC);
713 /* Userspace might have consumed the packet meanwhile:
714 * it's not supposed to do this usually, but might be hard
715 * to prevent. Discard data we got (if any) and keep going. */
716 if (unlikely(err != sock_len)) {
717 pr_debug("Discarded rx packet: "
718 " len %d, expected %zd\n", err, sock_len);
719 vhost_discard_vq_desc(vq, headcount);
720 continue;
721 }
Al Viroba7438ae2014-12-10 15:51:28 -0500722 /* Supply virtio_net_hdr if VHOST_NET_F_VIRTIO_NET_HDR */
Michael S. Tsirkin4c5a8442015-02-25 15:19:28 +0100723 if (unlikely(vhost_hlen)) {
724 if (copy_to_iter(&hdr, sizeof(hdr),
725 &fixup) != sizeof(hdr)) {
726 vq_err(vq, "Unable to write vnet_hdr "
727 "at addr %p\n", vq->iov->iov_base);
Jason Wang8241a1e2016-06-01 01:56:33 -0400728 goto out;
Michael S. Tsirkin4c5a8442015-02-25 15:19:28 +0100729 }
730 } else {
731 /* Header came from socket; we'll need to patch
732 * ->num_buffers over if VIRTIO_NET_F_MRG_RXBUF
733 */
734 iov_iter_advance(&fixup, sizeof(hdr));
David Stevens8dd014a2010-07-27 18:52:21 +0300735 }
736 /* TODO: Should check and handle checksum. */
Michael S. Tsirkin5201aa42015-02-03 11:07:06 +0200737
Jason Wang0960b642015-02-15 16:35:17 +0800738 num_buffers = cpu_to_vhost16(vq, headcount);
Jason Wangcfbdab92011-01-17 16:10:59 +0800739 if (likely(mergeable) &&
Michael S. Tsirkin0d79a492015-02-25 15:20:01 +0100740 copy_to_iter(&num_buffers, sizeof num_buffers,
741 &fixup) != sizeof num_buffers) {
David Stevens8dd014a2010-07-27 18:52:21 +0300742 vq_err(vq, "Failed num_buffers write");
743 vhost_discard_vq_desc(vq, headcount);
Jason Wang8241a1e2016-06-01 01:56:33 -0400744 goto out;
David Stevens8dd014a2010-07-27 18:52:21 +0300745 }
746 vhost_add_used_and_signal_n(&net->dev, vq, vq->heads,
747 headcount);
748 if (unlikely(vq_log))
749 vhost_log_write(vq, vq_log, log, vhost_len);
750 total_len += vhost_len;
751 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
752 vhost_poll_queue(&vq->poll);
Jason Wang8241a1e2016-06-01 01:56:33 -0400753 goto out;
David Stevens8dd014a2010-07-27 18:52:21 +0300754 }
755 }
Jason Wang8241a1e2016-06-01 01:56:33 -0400756 vhost_net_enable_vq(net, vq);
Asias He2e26af72013-05-07 14:54:33 +0800757out:
David Stevens8dd014a2010-07-27 18:52:21 +0300758 mutex_unlock(&vq->mutex);
David Stevens8dd014a2010-07-27 18:52:21 +0300759}
760
Tejun Heoc23f34452010-06-02 20:40:00 +0200761static void handle_tx_kick(struct vhost_work *work)
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000762{
Tejun Heoc23f34452010-06-02 20:40:00 +0200763 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
764 poll.work);
765 struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
766
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000767 handle_tx(net);
768}
769
Tejun Heoc23f34452010-06-02 20:40:00 +0200770static void handle_rx_kick(struct vhost_work *work)
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000771{
Tejun Heoc23f34452010-06-02 20:40:00 +0200772 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
773 poll.work);
774 struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
775
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000776 handle_rx(net);
777}
778
Tejun Heoc23f34452010-06-02 20:40:00 +0200779static void handle_tx_net(struct vhost_work *work)
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000780{
Tejun Heoc23f34452010-06-02 20:40:00 +0200781 struct vhost_net *net = container_of(work, struct vhost_net,
782 poll[VHOST_NET_VQ_TX].work);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000783 handle_tx(net);
784}
785
Tejun Heoc23f34452010-06-02 20:40:00 +0200786static void handle_rx_net(struct vhost_work *work)
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000787{
Tejun Heoc23f34452010-06-02 20:40:00 +0200788 struct vhost_net *net = container_of(work, struct vhost_net,
789 poll[VHOST_NET_VQ_RX].work);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000790 handle_rx(net);
791}
792
793static int vhost_net_open(struct inode *inode, struct file *f)
794{
Michael S. Tsirkin23cc5a92013-01-23 21:46:47 +0100795 struct vhost_net *n;
Tejun Heoc23f34452010-06-02 20:40:00 +0200796 struct vhost_dev *dev;
Asias He3ab2e422013-04-27 11:16:48 +0800797 struct vhost_virtqueue **vqs;
Zhi Yong Wu59566b6e2013-12-07 04:13:03 +0800798 int i;
Tejun Heoc23f34452010-06-02 20:40:00 +0200799
Michael S. Tsirkin23cc5a92013-01-23 21:46:47 +0100800 n = kmalloc(sizeof *n, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
801 if (!n) {
802 n = vmalloc(sizeof *n);
803 if (!n)
804 return -ENOMEM;
805 }
Asias He3ab2e422013-04-27 11:16:48 +0800806 vqs = kmalloc(VHOST_NET_VQ_MAX * sizeof(*vqs), GFP_KERNEL);
807 if (!vqs) {
Romain Francoised04257b2014-06-12 10:42:34 +0200808 kvfree(n);
Asias He3ab2e422013-04-27 11:16:48 +0800809 return -ENOMEM;
810 }
Tejun Heoc23f34452010-06-02 20:40:00 +0200811
812 dev = &n->dev;
Asias He3ab2e422013-04-27 11:16:48 +0800813 vqs[VHOST_NET_VQ_TX] = &n->vqs[VHOST_NET_VQ_TX].vq;
814 vqs[VHOST_NET_VQ_RX] = &n->vqs[VHOST_NET_VQ_RX].vq;
815 n->vqs[VHOST_NET_VQ_TX].vq.handle_kick = handle_tx_kick;
816 n->vqs[VHOST_NET_VQ_RX].vq.handle_kick = handle_rx_kick;
Asias He28394002013-04-27 15:07:46 +0800817 for (i = 0; i < VHOST_NET_VQ_MAX; i++) {
818 n->vqs[i].ubufs = NULL;
819 n->vqs[i].ubuf_info = NULL;
820 n->vqs[i].upend_idx = 0;
821 n->vqs[i].done_idx = 0;
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300822 n->vqs[i].vhost_hlen = 0;
823 n->vqs[i].sock_hlen = 0;
Asias He28394002013-04-27 15:07:46 +0800824 }
Zhi Yong Wu59566b6e2013-12-07 04:13:03 +0800825 vhost_dev_init(dev, vqs, VHOST_NET_VQ_MAX);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000826
Tejun Heoc23f34452010-06-02 20:40:00 +0200827 vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT, dev);
828 vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN, dev);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000829
830 f->private_data = n;
831
832 return 0;
833}
834
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000835static struct socket *vhost_net_stop_vq(struct vhost_net *n,
836 struct vhost_virtqueue *vq)
837{
838 struct socket *sock;
839
840 mutex_lock(&vq->mutex);
Asias He22fa90c2013-05-07 14:54:36 +0800841 sock = vq->private_data;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000842 vhost_net_disable_vq(n, vq);
Asias He22fa90c2013-05-07 14:54:36 +0800843 vq->private_data = NULL;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000844 mutex_unlock(&vq->mutex);
845 return sock;
846}
847
848static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock,
849 struct socket **rx_sock)
850{
Asias He3ab2e422013-04-27 11:16:48 +0800851 *tx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_TX].vq);
852 *rx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_RX].vq);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000853}
854
855static void vhost_net_flush_vq(struct vhost_net *n, int index)
856{
857 vhost_poll_flush(n->poll + index);
Asias He3ab2e422013-04-27 11:16:48 +0800858 vhost_poll_flush(&n->vqs[index].vq.poll);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000859}
860
861static void vhost_net_flush(struct vhost_net *n)
862{
863 vhost_net_flush_vq(n, VHOST_NET_VQ_TX);
864 vhost_net_flush_vq(n, VHOST_NET_VQ_RX);
Asias He28394002013-04-27 15:07:46 +0800865 if (n->vqs[VHOST_NET_VQ_TX].ubufs) {
Asias He3ab2e422013-04-27 11:16:48 +0800866 mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
Michael S. Tsirkin1280c272012-12-04 00:17:14 +0200867 n->tx_flush = true;
Asias He3ab2e422013-04-27 11:16:48 +0800868 mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
Michael S. Tsirkin1280c272012-12-04 00:17:14 +0200869 /* Wait for all lower device DMAs done. */
Asias Hefe729a52013-05-06 16:38:24 +0800870 vhost_net_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].ubufs);
Asias He3ab2e422013-04-27 11:16:48 +0800871 mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
Michael S. Tsirkin1280c272012-12-04 00:17:14 +0200872 n->tx_flush = false;
Michael S. Tsirkin0ad8b482014-02-13 11:42:05 +0200873 atomic_set(&n->vqs[VHOST_NET_VQ_TX].ubufs->refcount, 1);
Asias He3ab2e422013-04-27 11:16:48 +0800874 mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
Michael S. Tsirkin1280c272012-12-04 00:17:14 +0200875 }
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000876}
877
878static int vhost_net_release(struct inode *inode, struct file *f)
879{
880 struct vhost_net *n = f->private_data;
881 struct socket *tx_sock;
882 struct socket *rx_sock;
883
884 vhost_net_stop(n, &tx_sock, &rx_sock);
885 vhost_net_flush(n);
Michael S. Tsirkinb2116162012-11-01 09:16:46 +0000886 vhost_dev_stop(&n->dev);
Michael S. Tsirkinea5d4042011-11-27 19:05:58 +0200887 vhost_dev_cleanup(&n->dev, false);
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +0300888 vhost_net_vq_reset(n);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000889 if (tx_sock)
Al Viro09aaacf2014-03-05 20:39:00 -0500890 sockfd_put(tx_sock);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000891 if (rx_sock)
Al Viro09aaacf2014-03-05 20:39:00 -0500892 sockfd_put(rx_sock);
Michael S. Tsirkinb0c057c2014-02-13 11:45:11 +0200893 /* Make sure no callbacks are outstanding */
894 synchronize_rcu_bh();
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000895 /* We do an extra flush before freeing memory,
896 * since jobs can re-queue themselves. */
897 vhost_net_flush(n);
Asias He3ab2e422013-04-27 11:16:48 +0800898 kfree(n->dev.vqs);
Romain Francoised04257b2014-06-12 10:42:34 +0200899 kvfree(n);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000900 return 0;
901}
902
903static struct socket *get_raw_socket(int fd)
904{
905 struct {
906 struct sockaddr_ll sa;
907 char buf[MAX_ADDR_LEN];
908 } uaddr;
909 int uaddr_len = sizeof uaddr, r;
910 struct socket *sock = sockfd_lookup(fd, &r);
Krishna Kumard47effe2011-03-01 17:06:37 +0530911
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000912 if (!sock)
913 return ERR_PTR(-ENOTSOCK);
914
915 /* Parameter checking */
916 if (sock->sk->sk_type != SOCK_RAW) {
917 r = -ESOCKTNOSUPPORT;
918 goto err;
919 }
920
921 r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa,
922 &uaddr_len, 0);
923 if (r)
924 goto err;
925
926 if (uaddr.sa.sll_family != AF_PACKET) {
927 r = -EPFNOSUPPORT;
928 goto err;
929 }
930 return sock;
931err:
Al Viro09aaacf2014-03-05 20:39:00 -0500932 sockfd_put(sock);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000933 return ERR_PTR(r);
934}
935
Arnd Bergmann501c7742010-02-18 05:46:50 +0000936static struct socket *get_tap_socket(int fd)
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000937{
938 struct file *file = fget(fd);
939 struct socket *sock;
Krishna Kumard47effe2011-03-01 17:06:37 +0530940
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000941 if (!file)
942 return ERR_PTR(-EBADF);
943 sock = tun_get_socket(file);
Arnd Bergmann501c7742010-02-18 05:46:50 +0000944 if (!IS_ERR(sock))
945 return sock;
946 sock = macvtap_get_socket(file);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000947 if (IS_ERR(sock))
948 fput(file);
949 return sock;
950}
951
952static struct socket *get_socket(int fd)
953{
954 struct socket *sock;
Krishna Kumard47effe2011-03-01 17:06:37 +0530955
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000956 /* special case to disable backend */
957 if (fd == -1)
958 return NULL;
959 sock = get_raw_socket(fd);
960 if (!IS_ERR(sock))
961 return sock;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000962 sock = get_tap_socket(fd);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000963 if (!IS_ERR(sock))
964 return sock;
965 return ERR_PTR(-ENOTSOCK);
966}
967
968static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
969{
970 struct socket *sock, *oldsock;
971 struct vhost_virtqueue *vq;
Asias He28394002013-04-27 15:07:46 +0800972 struct vhost_net_virtqueue *nvq;
Asias Hefe729a52013-05-06 16:38:24 +0800973 struct vhost_net_ubuf_ref *ubufs, *oldubufs = NULL;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000974 int r;
975
976 mutex_lock(&n->dev.mutex);
977 r = vhost_dev_check_owner(&n->dev);
978 if (r)
979 goto err;
980
981 if (index >= VHOST_NET_VQ_MAX) {
982 r = -ENOBUFS;
983 goto err;
984 }
Asias He3ab2e422013-04-27 11:16:48 +0800985 vq = &n->vqs[index].vq;
Asias He28394002013-04-27 15:07:46 +0800986 nvq = &n->vqs[index];
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000987 mutex_lock(&vq->mutex);
988
989 /* Verify that ring has been setup correctly. */
990 if (!vhost_vq_access_ok(vq)) {
991 r = -EFAULT;
Jeff Dike1dace8c2010-03-04 16:10:14 -0500992 goto err_vq;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000993 }
994 sock = get_socket(fd);
995 if (IS_ERR(sock)) {
996 r = PTR_ERR(sock);
Jeff Dike1dace8c2010-03-04 16:10:14 -0500997 goto err_vq;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +0000998 }
999
1000 /* start polling new socket */
Asias He22fa90c2013-05-07 14:54:36 +08001001 oldsock = vq->private_data;
David S. Miller11fe8832010-07-20 18:25:24 -07001002 if (sock != oldsock) {
Asias Hefe729a52013-05-06 16:38:24 +08001003 ubufs = vhost_net_ubuf_alloc(vq,
1004 sock && vhost_sock_zcopy(sock));
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +00001005 if (IS_ERR(ubufs)) {
1006 r = PTR_ERR(ubufs);
1007 goto err_ubufs;
1008 }
Jason Wang692a9982013-01-28 01:05:17 +00001009
Krishna Kumard47effe2011-03-01 17:06:37 +05301010 vhost_net_disable_vq(n, vq);
Asias He22fa90c2013-05-07 14:54:36 +08001011 vq->private_data = sock;
Greg Kurz80f7d032016-02-16 15:59:44 +01001012 r = vhost_vq_init_access(vq);
Jason Wangf59281d2011-06-21 18:04:27 +08001013 if (r)
Jason Wang692a9982013-01-28 01:05:17 +00001014 goto err_used;
Jason Wang2b8b3282013-01-28 01:05:18 +00001015 r = vhost_net_enable_vq(n, vq);
1016 if (r)
1017 goto err_used;
Jason Wang692a9982013-01-28 01:05:17 +00001018
Asias He28394002013-04-27 15:07:46 +08001019 oldubufs = nvq->ubufs;
1020 nvq->ubufs = ubufs;
Michael S. Tsirkin64e9a9b2012-12-03 07:31:51 +00001021
1022 n->tx_packets = 0;
1023 n->tx_zcopy_err = 0;
Michael S. Tsirkin1280c272012-12-04 00:17:14 +02001024 n->tx_flush = false;
Jeff Dikedd1f4072010-03-04 16:10:14 -05001025 }
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001026
Michael S. Tsirkin1680e902010-07-15 15:19:12 +03001027 mutex_unlock(&vq->mutex);
1028
Michael S. Tsirkinc047e5f2011-07-20 13:41:31 +03001029 if (oldubufs) {
Michael S. Tsirkinc38e39c2013-06-25 17:29:46 +03001030 vhost_net_ubuf_put_wait_and_free(oldubufs);
Michael S. Tsirkinc047e5f2011-07-20 13:41:31 +03001031 mutex_lock(&vq->mutex);
Michael S. Tsirkineaae8132012-11-01 09:16:51 +00001032 vhost_zerocopy_signal_used(n, vq);
Michael S. Tsirkinc047e5f2011-07-20 13:41:31 +03001033 mutex_unlock(&vq->mutex);
1034 }
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +00001035
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001036 if (oldsock) {
1037 vhost_net_flush_vq(n, index);
Al Viro09aaacf2014-03-05 20:39:00 -05001038 sockfd_put(oldsock);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001039 }
Jeff Dike1dace8c2010-03-04 16:10:14 -05001040
Michael S. Tsirkin1680e902010-07-15 15:19:12 +03001041 mutex_unlock(&n->dev.mutex);
1042 return 0;
1043
Jason Wang692a9982013-01-28 01:05:17 +00001044err_used:
Asias He22fa90c2013-05-07 14:54:36 +08001045 vq->private_data = oldsock;
Jason Wang692a9982013-01-28 01:05:17 +00001046 vhost_net_enable_vq(n, vq);
1047 if (ubufs)
Michael S. Tsirkinc38e39c2013-06-25 17:29:46 +03001048 vhost_net_ubuf_put_wait_and_free(ubufs);
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +00001049err_ubufs:
Al Viro09aaacf2014-03-05 20:39:00 -05001050 sockfd_put(sock);
Jeff Dike1dace8c2010-03-04 16:10:14 -05001051err_vq:
1052 mutex_unlock(&vq->mutex);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001053err:
1054 mutex_unlock(&n->dev.mutex);
1055 return r;
1056}
1057
1058static long vhost_net_reset_owner(struct vhost_net *n)
1059{
1060 struct socket *tx_sock = NULL;
1061 struct socket *rx_sock = NULL;
1062 long err;
Jason Wanga9709d62016-06-23 02:04:31 -04001063 struct vhost_umem *umem;
Krishna Kumard47effe2011-03-01 17:06:37 +05301064
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001065 mutex_lock(&n->dev.mutex);
1066 err = vhost_dev_check_owner(&n->dev);
1067 if (err)
1068 goto done;
Jason Wanga9709d62016-06-23 02:04:31 -04001069 umem = vhost_dev_reset_owner_prepare();
1070 if (!umem) {
Michael S. Tsirkin150b9e52013-04-28 17:12:08 +03001071 err = -ENOMEM;
1072 goto done;
1073 }
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001074 vhost_net_stop(n, &tx_sock, &rx_sock);
1075 vhost_net_flush(n);
Jason Wanga9709d62016-06-23 02:04:31 -04001076 vhost_dev_reset_owner(&n->dev, umem);
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +03001077 vhost_net_vq_reset(n);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001078done:
1079 mutex_unlock(&n->dev.mutex);
1080 if (tx_sock)
Al Viro09aaacf2014-03-05 20:39:00 -05001081 sockfd_put(tx_sock);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001082 if (rx_sock)
Al Viro09aaacf2014-03-05 20:39:00 -05001083 sockfd_put(rx_sock);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001084 return err;
1085}
1086
1087static int vhost_net_set_features(struct vhost_net *n, u64 features)
1088{
David Stevens8dd014a2010-07-27 18:52:21 +03001089 size_t vhost_hlen, sock_hlen, hdr_len;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001090 int i;
David Stevens8dd014a2010-07-27 18:52:21 +03001091
Michael S. Tsirkine4fca7d2014-10-24 14:23:52 +03001092 hdr_len = (features & ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
1093 (1ULL << VIRTIO_F_VERSION_1))) ?
David Stevens8dd014a2010-07-27 18:52:21 +03001094 sizeof(struct virtio_net_hdr_mrg_rxbuf) :
1095 sizeof(struct virtio_net_hdr);
1096 if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) {
1097 /* vhost provides vnet_hdr */
1098 vhost_hlen = hdr_len;
1099 sock_hlen = 0;
1100 } else {
1101 /* socket provides vnet_hdr */
1102 vhost_hlen = 0;
1103 sock_hlen = hdr_len;
1104 }
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001105 mutex_lock(&n->dev.mutex);
1106 if ((features & (1 << VHOST_F_LOG_ALL)) &&
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001107 !vhost_log_access_ok(&n->dev))
1108 goto out_unlock;
1109
1110 if ((features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))) {
1111 if (vhost_init_device_iotlb(&n->dev, true))
1112 goto out_unlock;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001113 }
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001114
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001115 for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
Asias He3ab2e422013-04-27 11:16:48 +08001116 mutex_lock(&n->vqs[i].vq.mutex);
Michael S. Tsirkinea16c512014-06-05 15:20:23 +03001117 n->vqs[i].vq.acked_features = features;
Michael S. Tsirkin81f95a52013-04-28 15:51:40 +03001118 n->vqs[i].vhost_hlen = vhost_hlen;
1119 n->vqs[i].sock_hlen = sock_hlen;
Asias He3ab2e422013-04-27 11:16:48 +08001120 mutex_unlock(&n->vqs[i].vq.mutex);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001121 }
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001122 mutex_unlock(&n->dev.mutex);
1123 return 0;
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001124
1125out_unlock:
1126 mutex_unlock(&n->dev.mutex);
1127 return -EFAULT;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001128}
1129
Asias Heb1ad8492013-05-06 11:16:00 +08001130static long vhost_net_set_owner(struct vhost_net *n)
1131{
1132 int r;
1133
1134 mutex_lock(&n->dev.mutex);
Michael S. Tsirkin05c05352013-06-06 15:20:39 +03001135 if (vhost_dev_has_owner(&n->dev)) {
1136 r = -EBUSY;
1137 goto out;
1138 }
Asias Heb1ad8492013-05-06 11:16:00 +08001139 r = vhost_net_set_ubuf_info(n);
1140 if (r)
1141 goto out;
1142 r = vhost_dev_set_owner(&n->dev);
1143 if (r)
1144 vhost_net_clear_ubuf_info(n);
1145 vhost_net_flush(n);
1146out:
1147 mutex_unlock(&n->dev.mutex);
1148 return r;
1149}
1150
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001151static long vhost_net_ioctl(struct file *f, unsigned int ioctl,
1152 unsigned long arg)
1153{
1154 struct vhost_net *n = f->private_data;
1155 void __user *argp = (void __user *)arg;
1156 u64 __user *featurep = argp;
1157 struct vhost_vring_file backend;
1158 u64 features;
1159 int r;
Krishna Kumard47effe2011-03-01 17:06:37 +05301160
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001161 switch (ioctl) {
1162 case VHOST_NET_SET_BACKEND:
Takuya Yoshikawad3553a52010-05-27 19:01:58 +09001163 if (copy_from_user(&backend, argp, sizeof backend))
1164 return -EFAULT;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001165 return vhost_net_set_backend(n, backend.index, backend.fd);
1166 case VHOST_GET_FEATURES:
Stefan Hajnoczi0dd05a32012-07-21 06:55:36 +00001167 features = VHOST_NET_FEATURES;
Takuya Yoshikawad3553a52010-05-27 19:01:58 +09001168 if (copy_to_user(featurep, &features, sizeof features))
1169 return -EFAULT;
1170 return 0;
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001171 case VHOST_SET_FEATURES:
Takuya Yoshikawad3553a52010-05-27 19:01:58 +09001172 if (copy_from_user(&features, featurep, sizeof features))
1173 return -EFAULT;
Stefan Hajnoczi0dd05a32012-07-21 06:55:36 +00001174 if (features & ~VHOST_NET_FEATURES)
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001175 return -EOPNOTSUPP;
1176 return vhost_net_set_features(n, features);
1177 case VHOST_RESET_OWNER:
1178 return vhost_net_reset_owner(n);
Asias Heb1ad8492013-05-06 11:16:00 +08001179 case VHOST_SET_OWNER:
1180 return vhost_net_set_owner(n);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001181 default:
1182 mutex_lock(&n->dev.mutex);
Michael S. Tsirkin935cdee2012-12-06 14:03:34 +02001183 r = vhost_dev_ioctl(&n->dev, ioctl, argp);
1184 if (r == -ENOIOCTLCMD)
1185 r = vhost_vring_ioctl(&n->dev, ioctl, argp);
1186 else
1187 vhost_net_flush(n);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001188 mutex_unlock(&n->dev.mutex);
1189 return r;
1190 }
1191}
1192
1193#ifdef CONFIG_COMPAT
1194static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl,
1195 unsigned long arg)
1196{
1197 return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1198}
1199#endif
1200
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001201static ssize_t vhost_net_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1202{
1203 struct file *file = iocb->ki_filp;
1204 struct vhost_net *n = file->private_data;
1205 struct vhost_dev *dev = &n->dev;
1206 int noblock = file->f_flags & O_NONBLOCK;
1207
1208 return vhost_chr_read_iter(dev, to, noblock);
1209}
1210
1211static ssize_t vhost_net_chr_write_iter(struct kiocb *iocb,
1212 struct iov_iter *from)
1213{
1214 struct file *file = iocb->ki_filp;
1215 struct vhost_net *n = file->private_data;
1216 struct vhost_dev *dev = &n->dev;
1217
1218 return vhost_chr_write_iter(dev, from);
1219}
1220
1221static unsigned int vhost_net_chr_poll(struct file *file, poll_table *wait)
1222{
1223 struct vhost_net *n = file->private_data;
1224 struct vhost_dev *dev = &n->dev;
1225
1226 return vhost_chr_poll(file, dev, wait);
1227}
1228
Tobias Klauser373a83a2010-05-17 15:12:49 +02001229static const struct file_operations vhost_net_fops = {
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001230 .owner = THIS_MODULE,
1231 .release = vhost_net_release,
Jason Wang6b1e6cc2016-06-23 02:04:32 -04001232 .read_iter = vhost_net_chr_read_iter,
1233 .write_iter = vhost_net_chr_write_iter,
1234 .poll = vhost_net_chr_poll,
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001235 .unlocked_ioctl = vhost_net_ioctl,
1236#ifdef CONFIG_COMPAT
1237 .compat_ioctl = vhost_net_compat_ioctl,
1238#endif
1239 .open = vhost_net_open,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001240 .llseek = noop_llseek,
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001241};
1242
1243static struct miscdevice vhost_net_misc = {
stephen hemminger7c7c7f02012-01-11 19:30:38 +00001244 .minor = VHOST_NET_MINOR,
1245 .name = "vhost-net",
1246 .fops = &vhost_net_fops,
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001247};
1248
Christoph Hellwiga8d37822010-04-13 14:11:25 -04001249static int vhost_net_init(void)
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001250{
Michael S. Tsirkinbab632d2011-07-18 03:48:46 +00001251 if (experimental_zcopytx)
Asias Hefe729a52013-05-06 16:38:24 +08001252 vhost_net_enable_zcopy(VHOST_NET_VQ_TX);
Tejun Heoc23f34452010-06-02 20:40:00 +02001253 return misc_register(&vhost_net_misc);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001254}
1255module_init(vhost_net_init);
1256
Christoph Hellwiga8d37822010-04-13 14:11:25 -04001257static void vhost_net_exit(void)
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001258{
1259 misc_deregister(&vhost_net_misc);
Michael S. Tsirkin3a4d5c942010-01-14 06:17:27 +00001260}
1261module_exit(vhost_net_exit);
1262
1263MODULE_VERSION("0.0.1");
1264MODULE_LICENSE("GPL v2");
1265MODULE_AUTHOR("Michael S. Tsirkin");
1266MODULE_DESCRIPTION("Host kernel accelerator for virtio net");
stephen hemminger7c7c7f02012-01-11 19:30:38 +00001267MODULE_ALIAS_MISCDEV(VHOST_NET_MINOR);
1268MODULE_ALIAS("devname:vhost-net");