]>
Commit | Line | Data |
---|---|---|
3a4d5c94 MT |
1 | /* Copyright (C) 2009 Red Hat, Inc. |
2 | * Author: Michael S. Tsirkin <mst@redhat.com> | |
3 | * | |
4 | * This work is licensed under the terms of the GNU GPL, version 2. | |
5 | * | |
6 | * virtio-net server in host kernel. | |
7 | */ | |
8 | ||
9 | #include <linux/compat.h> | |
10 | #include <linux/eventfd.h> | |
11 | #include <linux/vhost.h> | |
12 | #include <linux/virtio_net.h> | |
3a4d5c94 MT |
13 | #include <linux/miscdevice.h> |
14 | #include <linux/module.h> | |
bab632d6 | 15 | #include <linux/moduleparam.h> |
3a4d5c94 MT |
16 | #include <linux/mutex.h> |
17 | #include <linux/workqueue.h> | |
3a4d5c94 | 18 | #include <linux/file.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
e6017571 | 20 | #include <linux/sched/clock.h> |
174cd4b1 | 21 | #include <linux/sched/signal.h> |
23cc5a99 | 22 | #include <linux/vmalloc.h> |
3a4d5c94 MT |
23 | |
24 | #include <linux/net.h> | |
25 | #include <linux/if_packet.h> | |
26 | #include <linux/if_arp.h> | |
27 | #include <linux/if_tun.h> | |
501c774c | 28 | #include <linux/if_macvlan.h> |
635b8c8e | 29 | #include <linux/if_tap.h> |
c53cff5e | 30 | #include <linux/if_vlan.h> |
c67df11f JW |
31 | #include <linux/skb_array.h> |
32 | #include <linux/skbuff.h> | |
3a4d5c94 MT |
33 | |
34 | #include <net/sock.h> | |
1ffcbc85 | 35 | #include <net/xdp.h> |
3a4d5c94 MT |
36 | |
37 | #include "vhost.h" | |
38 | ||
f9611c43 | 39 | static int experimental_zcopytx = 1; |
bab632d6 | 40 | module_param(experimental_zcopytx, int, 0444); |
f9611c43 MT |
41 | MODULE_PARM_DESC(experimental_zcopytx, "Enable Zero Copy TX;" |
42 | " 1 -Enable; 0 - Disable"); | |
bab632d6 | 43 | |
3a4d5c94 MT |
44 | /* Max number of bytes transferred before requeueing the job. |
45 | * Using this limit prevents one virtqueue from starving others. */ | |
46 | #define VHOST_NET_WEIGHT 0x80000 | |
47 | ||
a2ac9990 | 48 | /* Max number of packets transferred before requeueing the job. |
db688c24 PA |
49 | * Using this limit prevents one virtqueue from starving others with small |
50 | * pkts. | |
51 | */ | |
52 | #define VHOST_NET_PKT_WEIGHT 256 | |
a2ac9990 | 53 | |
bab632d6 MT |
54 | /* MAX number of TX used buffers for outstanding zerocopy */ |
55 | #define VHOST_MAX_PEND 128 | |
56 | #define VHOST_GOODCOPY_LEN 256 | |
57 | ||
eaae8132 MT |
58 | /* |
59 | * For transmit, used buffer len is unused; we override it to track buffer | |
60 | * status internally; used for zerocopy tx only. | |
61 | */ | |
62 | /* Lower device DMA failed */ | |
bf995734 | 63 | #define VHOST_DMA_FAILED_LEN ((__force __virtio32)3) |
eaae8132 | 64 | /* Lower device DMA done */ |
bf995734 | 65 | #define VHOST_DMA_DONE_LEN ((__force __virtio32)2) |
eaae8132 | 66 | /* Lower device DMA in progress */ |
bf995734 | 67 | #define VHOST_DMA_IN_PROGRESS ((__force __virtio32)1) |
eaae8132 | 68 | /* Buffer unused */ |
bf995734 | 69 | #define VHOST_DMA_CLEAR_LEN ((__force __virtio32)0) |
eaae8132 | 70 | |
bf995734 | 71 | #define VHOST_DMA_IS_DONE(len) ((__force u32)(len) >= (__force u32)VHOST_DMA_DONE_LEN) |
eaae8132 | 72 | |
8570a6e7 AH |
73 | enum { |
74 | VHOST_NET_FEATURES = VHOST_FEATURES | | |
75 | (1ULL << VHOST_NET_F_VIRTIO_NET_HDR) | | |
6b1e6cc7 JW |
76 | (1ULL << VIRTIO_NET_F_MRG_RXBUF) | |
77 | (1ULL << VIRTIO_F_IOMMU_PLATFORM) | |
8570a6e7 AH |
78 | }; |
79 | ||
429711ae JW |
80 | enum { |
81 | VHOST_NET_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2) | |
82 | }; | |
83 | ||
3a4d5c94 MT |
84 | enum { |
85 | VHOST_NET_VQ_RX = 0, | |
86 | VHOST_NET_VQ_TX = 1, | |
87 | VHOST_NET_VQ_MAX = 2, | |
88 | }; | |
89 | ||
fe729a57 | 90 | struct vhost_net_ubuf_ref { |
0ad8b480 MT |
91 | /* refcount follows semantics similar to kref: |
92 | * 0: object is released | |
93 | * 1: no outstanding ubufs | |
94 | * >1: outstanding ubufs | |
95 | */ | |
96 | atomic_t refcount; | |
2839400f AH |
97 | wait_queue_head_t wait; |
98 | struct vhost_virtqueue *vq; | |
99 | }; | |
100 | ||
d0d86971 | 101 | #define VHOST_NET_BATCH 64 |
c67df11f | 102 | struct vhost_net_buf { |
5990a305 | 103 | void **queue; |
c67df11f JW |
104 | int tail; |
105 | int head; | |
106 | }; | |
107 | ||
3ab2e420 AH |
108 | struct vhost_net_virtqueue { |
109 | struct vhost_virtqueue vq; | |
81f95a55 MT |
110 | size_t vhost_hlen; |
111 | size_t sock_hlen; | |
2839400f AH |
112 | /* vhost zerocopy support fields below: */ |
113 | /* last used idx for outstanding DMA zerocopy buffers */ | |
114 | int upend_idx; | |
f5a4941a JW |
115 | /* For TX, first used idx for DMA done zerocopy buffers |
116 | * For RX, number of batched heads | |
117 | */ | |
2839400f AH |
118 | int done_idx; |
119 | /* an array of userspace buffers info */ | |
120 | struct ubuf_info *ubuf_info; | |
121 | /* Reference counting for outstanding ubufs. | |
122 | * Protected by vq mutex. Writers must also take device mutex. */ | |
fe729a57 | 123 | struct vhost_net_ubuf_ref *ubufs; |
5990a305 | 124 | struct ptr_ring *rx_ring; |
c67df11f | 125 | struct vhost_net_buf rxq; |
3ab2e420 AH |
126 | }; |
127 | ||
3a4d5c94 MT |
128 | struct vhost_net { |
129 | struct vhost_dev dev; | |
3ab2e420 | 130 | struct vhost_net_virtqueue vqs[VHOST_NET_VQ_MAX]; |
3a4d5c94 | 131 | struct vhost_poll poll[VHOST_NET_VQ_MAX]; |
eaae8132 MT |
132 | /* Number of TX recently submitted. |
133 | * Protected by tx vq lock. */ | |
134 | unsigned tx_packets; | |
135 | /* Number of times zerocopy TX recently failed. | |
136 | * Protected by tx vq lock. */ | |
137 | unsigned tx_zcopy_err; | |
1280c27f MT |
138 | /* Flush in progress. Protected by tx vq lock. */ |
139 | bool tx_flush; | |
3a4d5c94 MT |
140 | }; |
141 | ||
fe729a57 | 142 | static unsigned vhost_net_zcopy_mask __read_mostly; |
2839400f | 143 | |
c67df11f JW |
144 | static void *vhost_net_buf_get_ptr(struct vhost_net_buf *rxq) |
145 | { | |
146 | if (rxq->tail != rxq->head) | |
147 | return rxq->queue[rxq->head]; | |
148 | else | |
149 | return NULL; | |
150 | } | |
151 | ||
152 | static int vhost_net_buf_get_size(struct vhost_net_buf *rxq) | |
153 | { | |
154 | return rxq->tail - rxq->head; | |
155 | } | |
156 | ||
157 | static int vhost_net_buf_is_empty(struct vhost_net_buf *rxq) | |
158 | { | |
159 | return rxq->tail == rxq->head; | |
160 | } | |
161 | ||
162 | static void *vhost_net_buf_consume(struct vhost_net_buf *rxq) | |
163 | { | |
164 | void *ret = vhost_net_buf_get_ptr(rxq); | |
165 | ++rxq->head; | |
166 | return ret; | |
167 | } | |
168 | ||
169 | static int vhost_net_buf_produce(struct vhost_net_virtqueue *nvq) | |
170 | { | |
171 | struct vhost_net_buf *rxq = &nvq->rxq; | |
172 | ||
173 | rxq->head = 0; | |
5990a305 | 174 | rxq->tail = ptr_ring_consume_batched(nvq->rx_ring, rxq->queue, |
d0d86971 | 175 | VHOST_NET_BATCH); |
c67df11f JW |
176 | return rxq->tail; |
177 | } | |
178 | ||
179 | static void vhost_net_buf_unproduce(struct vhost_net_virtqueue *nvq) | |
180 | { | |
181 | struct vhost_net_buf *rxq = &nvq->rxq; | |
182 | ||
5990a305 JW |
183 | if (nvq->rx_ring && !vhost_net_buf_is_empty(rxq)) { |
184 | ptr_ring_unconsume(nvq->rx_ring, rxq->queue + rxq->head, | |
185 | vhost_net_buf_get_size(rxq), | |
3a403076 | 186 | tun_ptr_free); |
c67df11f JW |
187 | rxq->head = rxq->tail = 0; |
188 | } | |
189 | } | |
190 | ||
fc72d1d5 JW |
191 | static int vhost_net_buf_peek_len(void *ptr) |
192 | { | |
1ffcbc85 JDB |
193 | if (tun_is_xdp_frame(ptr)) { |
194 | struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr); | |
fc72d1d5 | 195 | |
1ffcbc85 | 196 | return xdpf->len; |
fc72d1d5 JW |
197 | } |
198 | ||
199 | return __skb_array_len_with_tag(ptr); | |
200 | } | |
201 | ||
c67df11f JW |
202 | static int vhost_net_buf_peek(struct vhost_net_virtqueue *nvq) |
203 | { | |
204 | struct vhost_net_buf *rxq = &nvq->rxq; | |
205 | ||
206 | if (!vhost_net_buf_is_empty(rxq)) | |
207 | goto out; | |
208 | ||
209 | if (!vhost_net_buf_produce(nvq)) | |
210 | return 0; | |
211 | ||
212 | out: | |
fc72d1d5 | 213 | return vhost_net_buf_peek_len(vhost_net_buf_get_ptr(rxq)); |
c67df11f JW |
214 | } |
215 | ||
216 | static void vhost_net_buf_init(struct vhost_net_buf *rxq) | |
217 | { | |
218 | rxq->head = rxq->tail = 0; | |
219 | } | |
220 | ||
fe729a57 | 221 | static void vhost_net_enable_zcopy(int vq) |
2839400f | 222 | { |
fe729a57 | 223 | vhost_net_zcopy_mask |= 0x1 << vq; |
2839400f AH |
224 | } |
225 | ||
fe729a57 AH |
226 | static struct vhost_net_ubuf_ref * |
227 | vhost_net_ubuf_alloc(struct vhost_virtqueue *vq, bool zcopy) | |
2839400f | 228 | { |
fe729a57 | 229 | struct vhost_net_ubuf_ref *ubufs; |
2839400f AH |
230 | /* No zero copy backend? Nothing to count. */ |
231 | if (!zcopy) | |
232 | return NULL; | |
233 | ubufs = kmalloc(sizeof(*ubufs), GFP_KERNEL); | |
234 | if (!ubufs) | |
235 | return ERR_PTR(-ENOMEM); | |
0ad8b480 | 236 | atomic_set(&ubufs->refcount, 1); |
2839400f AH |
237 | init_waitqueue_head(&ubufs->wait); |
238 | ubufs->vq = vq; | |
239 | return ubufs; | |
240 | } | |
241 | ||
0ad8b480 | 242 | static int vhost_net_ubuf_put(struct vhost_net_ubuf_ref *ubufs) |
2839400f | 243 | { |
0ad8b480 MT |
244 | int r = atomic_sub_return(1, &ubufs->refcount); |
245 | if (unlikely(!r)) | |
246 | wake_up(&ubufs->wait); | |
247 | return r; | |
2839400f AH |
248 | } |
249 | ||
fe729a57 | 250 | static void vhost_net_ubuf_put_and_wait(struct vhost_net_ubuf_ref *ubufs) |
2839400f | 251 | { |
0ad8b480 MT |
252 | vhost_net_ubuf_put(ubufs); |
253 | wait_event(ubufs->wait, !atomic_read(&ubufs->refcount)); | |
c38e39c3 MT |
254 | } |
255 | ||
256 | static void vhost_net_ubuf_put_wait_and_free(struct vhost_net_ubuf_ref *ubufs) | |
257 | { | |
258 | vhost_net_ubuf_put_and_wait(ubufs); | |
2839400f AH |
259 | kfree(ubufs); |
260 | } | |
261 | ||
b1ad8496 AH |
262 | static void vhost_net_clear_ubuf_info(struct vhost_net *n) |
263 | { | |
b1ad8496 AH |
264 | int i; |
265 | ||
288cfe78 MT |
266 | for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { |
267 | kfree(n->vqs[i].ubuf_info); | |
268 | n->vqs[i].ubuf_info = NULL; | |
b1ad8496 AH |
269 | } |
270 | } | |
271 | ||
0a1febf7 | 272 | static int vhost_net_set_ubuf_info(struct vhost_net *n) |
2839400f AH |
273 | { |
274 | bool zcopy; | |
275 | int i; | |
276 | ||
288cfe78 | 277 | for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { |
fe729a57 | 278 | zcopy = vhost_net_zcopy_mask & (0x1 << i); |
2839400f AH |
279 | if (!zcopy) |
280 | continue; | |
6da2ec56 KC |
281 | n->vqs[i].ubuf_info = |
282 | kmalloc_array(UIO_MAXIOV, | |
283 | sizeof(*n->vqs[i].ubuf_info), | |
284 | GFP_KERNEL); | |
2839400f AH |
285 | if (!n->vqs[i].ubuf_info) |
286 | goto err; | |
287 | } | |
288 | return 0; | |
289 | ||
290 | err: | |
288cfe78 | 291 | vhost_net_clear_ubuf_info(n); |
2839400f AH |
292 | return -ENOMEM; |
293 | } | |
294 | ||
0a1febf7 | 295 | static void vhost_net_vq_reset(struct vhost_net *n) |
2839400f AH |
296 | { |
297 | int i; | |
298 | ||
288cfe78 MT |
299 | vhost_net_clear_ubuf_info(n); |
300 | ||
2839400f AH |
301 | for (i = 0; i < VHOST_NET_VQ_MAX; i++) { |
302 | n->vqs[i].done_idx = 0; | |
303 | n->vqs[i].upend_idx = 0; | |
304 | n->vqs[i].ubufs = NULL; | |
81f95a55 MT |
305 | n->vqs[i].vhost_hlen = 0; |
306 | n->vqs[i].sock_hlen = 0; | |
c67df11f | 307 | vhost_net_buf_init(&n->vqs[i].rxq); |
2839400f AH |
308 | } |
309 | ||
310 | } | |
311 | ||
eaae8132 MT |
312 | static void vhost_net_tx_packet(struct vhost_net *net) |
313 | { | |
314 | ++net->tx_packets; | |
315 | if (net->tx_packets < 1024) | |
316 | return; | |
317 | net->tx_packets = 0; | |
318 | net->tx_zcopy_err = 0; | |
319 | } | |
320 | ||
321 | static void vhost_net_tx_err(struct vhost_net *net) | |
322 | { | |
323 | ++net->tx_zcopy_err; | |
324 | } | |
325 | ||
326 | static bool vhost_net_tx_select_zcopy(struct vhost_net *net) | |
327 | { | |
1280c27f MT |
328 | /* TX flush waits for outstanding DMAs to be done. |
329 | * Don't start new DMAs. | |
330 | */ | |
331 | return !net->tx_flush && | |
332 | net->tx_packets / 64 >= net->tx_zcopy_err; | |
eaae8132 MT |
333 | } |
334 | ||
bab632d6 MT |
335 | static bool vhost_sock_zcopy(struct socket *sock) |
336 | { | |
337 | return unlikely(experimental_zcopytx) && | |
338 | sock_flag(sock->sk, SOCK_ZEROCOPY); | |
339 | } | |
340 | ||
b211616d MT |
341 | /* In case of DMA done not in order in lower device driver for some reason. |
342 | * upend_idx is used to track end of used idx, done_idx is used to track head | |
343 | * of used idx. Once lower device DMA done contiguously, we will signal KVM | |
344 | * guest used idx. | |
345 | */ | |
094afe7d JW |
346 | static void vhost_zerocopy_signal_used(struct vhost_net *net, |
347 | struct vhost_virtqueue *vq) | |
b211616d | 348 | { |
2839400f AH |
349 | struct vhost_net_virtqueue *nvq = |
350 | container_of(vq, struct vhost_net_virtqueue, vq); | |
c92112ae | 351 | int i, add; |
b211616d MT |
352 | int j = 0; |
353 | ||
2839400f | 354 | for (i = nvq->done_idx; i != nvq->upend_idx; i = (i + 1) % UIO_MAXIOV) { |
eaae8132 MT |
355 | if (vq->heads[i].len == VHOST_DMA_FAILED_LEN) |
356 | vhost_net_tx_err(net); | |
b211616d MT |
357 | if (VHOST_DMA_IS_DONE(vq->heads[i].len)) { |
358 | vq->heads[i].len = VHOST_DMA_CLEAR_LEN; | |
b211616d MT |
359 | ++j; |
360 | } else | |
361 | break; | |
362 | } | |
c92112ae JW |
363 | while (j) { |
364 | add = min(UIO_MAXIOV - nvq->done_idx, j); | |
365 | vhost_add_used_and_signal_n(vq->dev, vq, | |
366 | &vq->heads[nvq->done_idx], add); | |
367 | nvq->done_idx = (nvq->done_idx + add) % UIO_MAXIOV; | |
368 | j -= add; | |
369 | } | |
b211616d MT |
370 | } |
371 | ||
eaae8132 | 372 | static void vhost_zerocopy_callback(struct ubuf_info *ubuf, bool success) |
b211616d | 373 | { |
fe729a57 | 374 | struct vhost_net_ubuf_ref *ubufs = ubuf->ctx; |
b211616d | 375 | struct vhost_virtqueue *vq = ubufs->vq; |
0ad8b480 | 376 | int cnt; |
24eb21a1 | 377 | |
b0c057ca MT |
378 | rcu_read_lock_bh(); |
379 | ||
19c73b3e JW |
380 | /* set len to mark this desc buffers done DMA */ |
381 | vq->heads[ubuf->desc].len = success ? | |
382 | VHOST_DMA_DONE_LEN : VHOST_DMA_FAILED_LEN; | |
0ad8b480 | 383 | cnt = vhost_net_ubuf_put(ubufs); |
19c73b3e | 384 | |
24eb21a1 MT |
385 | /* |
386 | * Trigger polling thread if guest stopped submitting new buffers: | |
0ad8b480 | 387 | * in this case, the refcount after decrement will eventually reach 1. |
24eb21a1 MT |
388 | * We also trigger polling periodically after each 16 packets |
389 | * (the value 16 here is more or less arbitrary, it's tuned to trigger | |
390 | * less than 10% of times). | |
391 | */ | |
0ad8b480 | 392 | if (cnt <= 1 || !(cnt % 16)) |
24eb21a1 | 393 | vhost_poll_queue(&vq->poll); |
b0c057ca MT |
394 | |
395 | rcu_read_unlock_bh(); | |
b211616d MT |
396 | } |
397 | ||
03088137 JW |
398 | static inline unsigned long busy_clock(void) |
399 | { | |
400 | return local_clock() >> 10; | |
401 | } | |
402 | ||
027b1760 | 403 | static bool vhost_can_busy_poll(unsigned long endtime) |
03088137 | 404 | { |
027b1760 TM |
405 | return likely(!need_resched() && !time_after(busy_clock(), endtime) && |
406 | !signal_pending(current)); | |
03088137 JW |
407 | } |
408 | ||
8241a1e4 JW |
409 | static void vhost_net_disable_vq(struct vhost_net *n, |
410 | struct vhost_virtqueue *vq) | |
411 | { | |
412 | struct vhost_net_virtqueue *nvq = | |
413 | container_of(vq, struct vhost_net_virtqueue, vq); | |
414 | struct vhost_poll *poll = n->poll + (nvq - n->vqs); | |
415 | if (!vq->private_data) | |
416 | return; | |
417 | vhost_poll_stop(poll); | |
418 | } | |
419 | ||
420 | static int vhost_net_enable_vq(struct vhost_net *n, | |
421 | struct vhost_virtqueue *vq) | |
422 | { | |
423 | struct vhost_net_virtqueue *nvq = | |
424 | container_of(vq, struct vhost_net_virtqueue, vq); | |
425 | struct vhost_poll *poll = n->poll + (nvq - n->vqs); | |
426 | struct socket *sock; | |
427 | ||
428 | sock = vq->private_data; | |
429 | if (!sock) | |
430 | return 0; | |
431 | ||
432 | return vhost_poll_start(poll, sock->file); | |
433 | } | |
434 | ||
4afb52c2 JW |
435 | static void vhost_net_signal_used(struct vhost_net_virtqueue *nvq) |
436 | { | |
437 | struct vhost_virtqueue *vq = &nvq->vq; | |
438 | struct vhost_dev *dev = vq->dev; | |
439 | ||
440 | if (!nvq->done_idx) | |
441 | return; | |
442 | ||
443 | vhost_add_used_and_signal_n(dev, vq, vq->heads, nvq->done_idx); | |
444 | nvq->done_idx = 0; | |
445 | } | |
446 | ||
03088137 | 447 | static int vhost_net_tx_get_vq_desc(struct vhost_net *net, |
4afb52c2 | 448 | struct vhost_net_virtqueue *nvq, |
027b1760 TM |
449 | unsigned int *out_num, unsigned int *in_num, |
450 | bool *busyloop_intr) | |
03088137 | 451 | { |
4afb52c2 | 452 | struct vhost_virtqueue *vq = &nvq->vq; |
03088137 JW |
453 | unsigned long uninitialized_var(endtime); |
454 | int r = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), | |
6b1e6cc7 | 455 | out_num, in_num, NULL, NULL); |
03088137 JW |
456 | |
457 | if (r == vq->num && vq->busyloop_timeout) { | |
4afb52c2 JW |
458 | if (!vhost_sock_zcopy(vq->private_data)) |
459 | vhost_net_signal_used(nvq); | |
03088137 JW |
460 | preempt_disable(); |
461 | endtime = busy_clock() + vq->busyloop_timeout; | |
027b1760 TM |
462 | while (vhost_can_busy_poll(endtime)) { |
463 | if (vhost_has_work(vq->dev)) { | |
464 | *busyloop_intr = true; | |
465 | break; | |
466 | } | |
467 | if (!vhost_vq_avail_empty(vq->dev, vq)) | |
468 | break; | |
f2f09a4c | 469 | cpu_relax(); |
027b1760 | 470 | } |
03088137 JW |
471 | preempt_enable(); |
472 | r = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), | |
6b1e6cc7 | 473 | out_num, in_num, NULL, NULL); |
03088137 JW |
474 | } |
475 | ||
476 | return r; | |
477 | } | |
478 | ||
0ed005ce JW |
479 | static bool vhost_exceeds_maxpend(struct vhost_net *net) |
480 | { | |
481 | struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX]; | |
482 | struct vhost_virtqueue *vq = &nvq->vq; | |
483 | ||
1e6f7453 WB |
484 | return (nvq->upend_idx + UIO_MAXIOV - nvq->done_idx) % UIO_MAXIOV > |
485 | min_t(unsigned int, VHOST_MAX_PEND, vq->num >> 2); | |
0ed005ce JW |
486 | } |
487 | ||
b0d0ea50 JW |
488 | static size_t init_iov_iter(struct vhost_virtqueue *vq, struct iov_iter *iter, |
489 | size_t hdr_size, int out) | |
490 | { | |
491 | /* Skip header. TODO: support TSO. */ | |
492 | size_t len = iov_length(vq->iov, out); | |
493 | ||
494 | iov_iter_init(iter, WRITE, vq->iov, out, len); | |
495 | iov_iter_advance(iter, hdr_size); | |
496 | ||
497 | return iov_iter_count(iter); | |
498 | } | |
499 | ||
272f35cb JW |
500 | static bool vhost_exceeds_weight(int pkts, int total_len) |
501 | { | |
502 | return total_len >= VHOST_NET_WEIGHT || | |
503 | pkts >= VHOST_NET_PKT_WEIGHT; | |
504 | } | |
505 | ||
a2a91a13 JW |
506 | static int get_tx_bufs(struct vhost_net *net, |
507 | struct vhost_net_virtqueue *nvq, | |
508 | struct msghdr *msg, | |
509 | unsigned int *out, unsigned int *in, | |
510 | size_t *len, bool *busyloop_intr) | |
511 | { | |
512 | struct vhost_virtqueue *vq = &nvq->vq; | |
513 | int ret; | |
514 | ||
4afb52c2 JW |
515 | ret = vhost_net_tx_get_vq_desc(net, nvq, out, in, busyloop_intr); |
516 | ||
a2a91a13 JW |
517 | if (ret < 0 || ret == vq->num) |
518 | return ret; | |
519 | ||
520 | if (*in) { | |
521 | vq_err(vq, "Unexpected descriptor format for TX: out %d, int %d\n", | |
522 | *out, *in); | |
523 | return -EFAULT; | |
524 | } | |
525 | ||
526 | /* Sanity check */ | |
527 | *len = init_iov_iter(vq, &msg->msg_iter, nvq->vhost_hlen, *out); | |
528 | if (*len == 0) { | |
529 | vq_err(vq, "Unexpected header len for TX: %zd expected %zd\n", | |
530 | *len, nvq->vhost_hlen); | |
531 | return -EFAULT; | |
532 | } | |
533 | ||
534 | return ret; | |
535 | } | |
536 | ||
c92a8a8c JW |
537 | static bool tx_can_batch(struct vhost_virtqueue *vq, size_t total_len) |
538 | { | |
539 | return total_len < VHOST_NET_WEIGHT && | |
540 | !vhost_vq_avail_empty(vq->dev, vq); | |
541 | } | |
542 | ||
0d20bdf3 | 543 | static void handle_tx_copy(struct vhost_net *net, struct socket *sock) |
3a4d5c94 | 544 | { |
2839400f | 545 | struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX]; |
81f95a55 | 546 | struct vhost_virtqueue *vq = &nvq->vq; |
98a527aa | 547 | unsigned out, in; |
d5675bd2 | 548 | int head; |
3a4d5c94 MT |
549 | struct msghdr msg = { |
550 | .msg_name = NULL, | |
551 | .msg_namelen = 0, | |
552 | .msg_control = NULL, | |
553 | .msg_controllen = 0, | |
3a4d5c94 MT |
554 | .msg_flags = MSG_DONTWAIT, |
555 | }; | |
556 | size_t len, total_len = 0; | |
70181d51 | 557 | int err; |
a2ac9990 | 558 | int sent_pkts = 0; |
28457ee6 | 559 | |
0d20bdf3 JW |
560 | for (;;) { |
561 | bool busyloop_intr = false; | |
3a4d5c94 | 562 | |
0d20bdf3 JW |
563 | head = get_tx_bufs(net, nvq, &msg, &out, &in, &len, |
564 | &busyloop_intr); | |
565 | /* On error, stop handling until the next kick. */ | |
566 | if (unlikely(head < 0)) | |
567 | break; | |
568 | /* Nothing new? Wait for eventfd to tell us they refilled. */ | |
569 | if (head == vq->num) { | |
570 | if (unlikely(busyloop_intr)) { | |
571 | vhost_poll_queue(&vq->poll); | |
572 | } else if (unlikely(vhost_enable_notify(&net->dev, | |
573 | vq))) { | |
574 | vhost_disable_notify(&net->dev, vq); | |
575 | continue; | |
576 | } | |
577 | break; | |
578 | } | |
6b1e6cc7 | 579 | |
4afb52c2 JW |
580 | vq->heads[nvq->done_idx].id = cpu_to_vhost32(vq, head); |
581 | vq->heads[nvq->done_idx].len = 0; | |
582 | ||
0d20bdf3 JW |
583 | total_len += len; |
584 | if (tx_can_batch(vq, total_len)) | |
585 | msg.msg_flags |= MSG_MORE; | |
586 | else | |
587 | msg.msg_flags &= ~MSG_MORE; | |
588 | ||
589 | /* TODO: Check specific error and bomb out unless ENOBUFS? */ | |
590 | err = sock->ops->sendmsg(sock, &msg, len); | |
591 | if (unlikely(err < 0)) { | |
592 | vhost_discard_vq_desc(vq, 1); | |
593 | vhost_net_enable_vq(net, vq); | |
594 | break; | |
595 | } | |
596 | if (err != len) | |
597 | pr_debug("Truncated TX packet: len %d != %zd\n", | |
598 | err, len); | |
4afb52c2 JW |
599 | if (++nvq->done_idx >= VHOST_NET_BATCH) |
600 | vhost_net_signal_used(nvq); | |
0d20bdf3 JW |
601 | if (vhost_exceeds_weight(++sent_pkts, total_len)) { |
602 | vhost_poll_queue(&vq->poll); | |
603 | break; | |
604 | } | |
605 | } | |
4afb52c2 JW |
606 | |
607 | vhost_net_signal_used(nvq); | |
0d20bdf3 | 608 | } |
3a4d5c94 | 609 | |
0d20bdf3 JW |
610 | static void handle_tx_zerocopy(struct vhost_net *net, struct socket *sock) |
611 | { | |
612 | struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX]; | |
613 | struct vhost_virtqueue *vq = &nvq->vq; | |
614 | unsigned out, in; | |
615 | int head; | |
616 | struct msghdr msg = { | |
617 | .msg_name = NULL, | |
618 | .msg_namelen = 0, | |
619 | .msg_control = NULL, | |
620 | .msg_controllen = 0, | |
621 | .msg_flags = MSG_DONTWAIT, | |
622 | }; | |
fe8dd45b | 623 | struct tun_msg_ctl ctl; |
0d20bdf3 JW |
624 | size_t len, total_len = 0; |
625 | int err; | |
626 | struct vhost_net_ubuf_ref *uninitialized_var(ubufs); | |
627 | bool zcopy_used; | |
628 | int sent_pkts = 0; | |
3a4d5c94 MT |
629 | |
630 | for (;;) { | |
027b1760 TM |
631 | bool busyloop_intr; |
632 | ||
bab632d6 | 633 | /* Release DMAs done buffers first */ |
0d20bdf3 | 634 | vhost_zerocopy_signal_used(net, vq); |
bab632d6 | 635 | |
027b1760 | 636 | busyloop_intr = false; |
a2a91a13 JW |
637 | head = get_tx_bufs(net, nvq, &msg, &out, &in, &len, |
638 | &busyloop_intr); | |
d5675bd2 | 639 | /* On error, stop handling until the next kick. */ |
7b3384fc | 640 | if (unlikely(head < 0)) |
d5675bd2 | 641 | break; |
3a4d5c94 MT |
642 | /* Nothing new? Wait for eventfd to tell us they refilled. */ |
643 | if (head == vq->num) { | |
027b1760 TM |
644 | if (unlikely(busyloop_intr)) { |
645 | vhost_poll_queue(&vq->poll); | |
646 | } else if (unlikely(vhost_enable_notify(&net->dev, vq))) { | |
8ea8cf89 | 647 | vhost_disable_notify(&net->dev, vq); |
3a4d5c94 MT |
648 | continue; |
649 | } | |
650 | break; | |
651 | } | |
ce21a029 | 652 | |
0d20bdf3 JW |
653 | zcopy_used = len >= VHOST_GOODCOPY_LEN |
654 | && !vhost_exceeds_maxpend(net) | |
655 | && vhost_net_tx_select_zcopy(net); | |
cedb9bdc | 656 | |
bab632d6 | 657 | /* use msg_control to pass vhost zerocopy ubuf info to skb */ |
cedb9bdc | 658 | if (zcopy_used) { |
ce21a029 JW |
659 | struct ubuf_info *ubuf; |
660 | ubuf = nvq->ubuf_info + nvq->upend_idx; | |
661 | ||
8b38694a | 662 | vq->heads[nvq->upend_idx].id = cpu_to_vhost32(vq, head); |
ce21a029 JW |
663 | vq->heads[nvq->upend_idx].len = VHOST_DMA_IN_PROGRESS; |
664 | ubuf->callback = vhost_zerocopy_callback; | |
665 | ubuf->ctx = nvq->ubufs; | |
666 | ubuf->desc = nvq->upend_idx; | |
c1d1b437 | 667 | refcount_set(&ubuf->refcnt, 1); |
fe8dd45b JW |
668 | msg.msg_control = &ctl; |
669 | ctl.type = TUN_MSG_UBUF; | |
670 | ctl.ptr = ubuf; | |
671 | msg.msg_controllen = sizeof(ctl); | |
ce21a029 | 672 | ubufs = nvq->ubufs; |
0ad8b480 | 673 | atomic_inc(&ubufs->refcount); |
2839400f | 674 | nvq->upend_idx = (nvq->upend_idx + 1) % UIO_MAXIOV; |
ce21a029 | 675 | } else { |
4364d5f9 | 676 | msg.msg_control = NULL; |
ce21a029 JW |
677 | ubufs = NULL; |
678 | } | |
0ed005ce | 679 | total_len += len; |
c92a8a8c | 680 | if (tx_can_batch(vq, total_len) && |
0ed005ce JW |
681 | likely(!vhost_exceeds_maxpend(net))) { |
682 | msg.msg_flags |= MSG_MORE; | |
683 | } else { | |
684 | msg.msg_flags &= ~MSG_MORE; | |
685 | } | |
686 | ||
3a4d5c94 | 687 | /* TODO: Check specific error and bomb out unless ENOBUFS? */ |
1b784140 | 688 | err = sock->ops->sendmsg(sock, &msg, len); |
3a4d5c94 | 689 | if (unlikely(err < 0)) { |
cedb9bdc | 690 | if (zcopy_used) { |
ce21a029 | 691 | vhost_net_ubuf_put(ubufs); |
2839400f AH |
692 | nvq->upend_idx = ((unsigned)nvq->upend_idx - 1) |
693 | % UIO_MAXIOV; | |
bab632d6 | 694 | } |
8dd014ad | 695 | vhost_discard_vq_desc(vq, 1); |
feb8892c | 696 | vhost_net_enable_vq(net, vq); |
3a4d5c94 MT |
697 | break; |
698 | } | |
699 | if (err != len) | |
95c0ec6a MT |
700 | pr_debug("Truncated TX packet: " |
701 | " len %d != %zd\n", err, len); | |
cedb9bdc | 702 | if (!zcopy_used) |
bab632d6 | 703 | vhost_add_used_and_signal(&net->dev, vq, head, 0); |
c8fb217a | 704 | else |
eaae8132 | 705 | vhost_zerocopy_signal_used(net, vq); |
eaae8132 | 706 | vhost_net_tx_packet(net); |
272f35cb | 707 | if (unlikely(vhost_exceeds_weight(++sent_pkts, total_len))) { |
3a4d5c94 MT |
708 | vhost_poll_queue(&vq->poll); |
709 | break; | |
710 | } | |
711 | } | |
0d20bdf3 JW |
712 | } |
713 | ||
714 | /* Expects to be always run from workqueue - which acts as | |
715 | * read-size critical section for our kind of RCU. */ | |
716 | static void handle_tx(struct vhost_net *net) | |
717 | { | |
718 | struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX]; | |
719 | struct vhost_virtqueue *vq = &nvq->vq; | |
720 | struct socket *sock; | |
721 | ||
722 | mutex_lock(&vq->mutex); | |
723 | sock = vq->private_data; | |
724 | if (!sock) | |
725 | goto out; | |
726 | ||
727 | if (!vq_iotlb_prefetch(vq)) | |
728 | goto out; | |
729 | ||
730 | vhost_disable_notify(&net->dev, vq); | |
731 | vhost_net_disable_vq(net, vq); | |
732 | ||
733 | if (vhost_sock_zcopy(sock)) | |
734 | handle_tx_zerocopy(net, sock); | |
735 | else | |
736 | handle_tx_copy(net, sock); | |
737 | ||
2e26af79 | 738 | out: |
3a4d5c94 | 739 | mutex_unlock(&vq->mutex); |
3a4d5c94 MT |
740 | } |
741 | ||
c67df11f | 742 | static int peek_head_len(struct vhost_net_virtqueue *rvq, struct sock *sk) |
8dd014ad DS |
743 | { |
744 | struct sk_buff *head; | |
745 | int len = 0; | |
783e3988 | 746 | unsigned long flags; |
8dd014ad | 747 | |
5990a305 | 748 | if (rvq->rx_ring) |
c67df11f | 749 | return vhost_net_buf_peek(rvq); |
1576d986 | 750 | |
783e3988 | 751 | spin_lock_irqsave(&sk->sk_receive_queue.lock, flags); |
8dd014ad | 752 | head = skb_peek(&sk->sk_receive_queue); |
c53cff5e | 753 | if (likely(head)) { |
8dd014ad | 754 | len = head->len; |
df8a39de | 755 | if (skb_vlan_tag_present(head)) |
c53cff5e BG |
756 | len += VLAN_HLEN; |
757 | } | |
758 | ||
783e3988 | 759 | spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags); |
8dd014ad DS |
760 | return len; |
761 | } | |
762 | ||
1576d986 JW |
763 | static int sk_has_rx_data(struct sock *sk) |
764 | { | |
765 | struct socket *sock = sk->sk_socket; | |
766 | ||
767 | if (sock->ops->peek_len) | |
768 | return sock->ops->peek_len(sock); | |
769 | ||
770 | return skb_queue_empty(&sk->sk_receive_queue); | |
771 | } | |
772 | ||
be294a51 TM |
773 | static int vhost_net_rx_peek_head_len(struct vhost_net *net, struct sock *sk, |
774 | bool *busyloop_intr) | |
03088137 | 775 | { |
28b9b33b TM |
776 | struct vhost_net_virtqueue *rnvq = &net->vqs[VHOST_NET_VQ_RX]; |
777 | struct vhost_net_virtqueue *tnvq = &net->vqs[VHOST_NET_VQ_TX]; | |
6369fec5 | 778 | struct vhost_virtqueue *rvq = &rnvq->vq; |
28b9b33b | 779 | struct vhost_virtqueue *tvq = &tnvq->vq; |
03088137 | 780 | unsigned long uninitialized_var(endtime); |
28b9b33b | 781 | int len = peek_head_len(rnvq, sk); |
03088137 | 782 | |
28b9b33b | 783 | if (!len && tvq->busyloop_timeout) { |
f5a4941a | 784 | /* Flush batched heads first */ |
09c32489 | 785 | vhost_net_signal_used(rnvq); |
03088137 | 786 | /* Both tx vq and rx socket were polled here */ |
28b9b33b TM |
787 | mutex_lock_nested(&tvq->mutex, 1); |
788 | vhost_disable_notify(&net->dev, tvq); | |
03088137 JW |
789 | |
790 | preempt_disable(); | |
28b9b33b | 791 | endtime = busy_clock() + tvq->busyloop_timeout; |
03088137 | 792 | |
be294a51 TM |
793 | while (vhost_can_busy_poll(endtime)) { |
794 | if (vhost_has_work(&net->dev)) { | |
795 | *busyloop_intr = true; | |
796 | break; | |
797 | } | |
6369fec5 TM |
798 | if ((sk_has_rx_data(sk) && |
799 | !vhost_vq_avail_empty(&net->dev, rvq)) || | |
be294a51 TM |
800 | !vhost_vq_avail_empty(&net->dev, tvq)) |
801 | break; | |
f2f09a4c | 802 | cpu_relax(); |
be294a51 | 803 | } |
03088137 JW |
804 | |
805 | preempt_enable(); | |
806 | ||
28b9b33b TM |
807 | if (!vhost_vq_avail_empty(&net->dev, tvq)) { |
808 | vhost_poll_queue(&tvq->poll); | |
809 | } else if (unlikely(vhost_enable_notify(&net->dev, tvq))) { | |
810 | vhost_disable_notify(&net->dev, tvq); | |
811 | vhost_poll_queue(&tvq->poll); | |
8b949bef JW |
812 | } |
813 | ||
28b9b33b | 814 | mutex_unlock(&tvq->mutex); |
03088137 | 815 | |
28b9b33b | 816 | len = peek_head_len(rnvq, sk); |
03088137 JW |
817 | } |
818 | ||
819 | return len; | |
820 | } | |
821 | ||
8dd014ad DS |
822 | /* This is a multi-buffer version of vhost_get_desc, that works if |
823 | * vq has read descriptors only. | |
824 | * @vq - the relevant virtqueue | |
825 | * @datalen - data length we'll be reading | |
826 | * @iovcount - returned count of io vectors we fill | |
827 | * @log - vhost log | |
828 | * @log_num - log offset | |
94249369 | 829 | * @quota - headcount quota, 1 for big buffer |
8dd014ad DS |
830 | * returns number of buffer heads allocated, negative on error |
831 | */ | |
832 | static int get_rx_bufs(struct vhost_virtqueue *vq, | |
833 | struct vring_used_elem *heads, | |
834 | int datalen, | |
835 | unsigned *iovcount, | |
836 | struct vhost_log *log, | |
94249369 JW |
837 | unsigned *log_num, |
838 | unsigned int quota) | |
8dd014ad DS |
839 | { |
840 | unsigned int out, in; | |
841 | int seg = 0; | |
842 | int headcount = 0; | |
843 | unsigned d; | |
844 | int r, nlogs = 0; | |
8b38694a MT |
845 | /* len is always initialized before use since we are always called with |
846 | * datalen > 0. | |
847 | */ | |
848 | u32 uninitialized_var(len); | |
8dd014ad | 849 | |
94249369 | 850 | while (datalen > 0 && headcount < quota) { |
e0e9b406 | 851 | if (unlikely(seg >= UIO_MAXIOV)) { |
8dd014ad DS |
852 | r = -ENOBUFS; |
853 | goto err; | |
854 | } | |
47283bef | 855 | r = vhost_get_vq_desc(vq, vq->iov + seg, |
8dd014ad DS |
856 | ARRAY_SIZE(vq->iov) - seg, &out, |
857 | &in, log, log_num); | |
a39ee449 MT |
858 | if (unlikely(r < 0)) |
859 | goto err; | |
860 | ||
861 | d = r; | |
8dd014ad DS |
862 | if (d == vq->num) { |
863 | r = 0; | |
864 | goto err; | |
865 | } | |
866 | if (unlikely(out || in <= 0)) { | |
867 | vq_err(vq, "unexpected descriptor format for RX: " | |
868 | "out %d, in %d\n", out, in); | |
869 | r = -EINVAL; | |
870 | goto err; | |
871 | } | |
872 | if (unlikely(log)) { | |
873 | nlogs += *log_num; | |
874 | log += *log_num; | |
875 | } | |
8b38694a MT |
876 | heads[headcount].id = cpu_to_vhost32(vq, d); |
877 | len = iov_length(vq->iov + seg, in); | |
878 | heads[headcount].len = cpu_to_vhost32(vq, len); | |
879 | datalen -= len; | |
8dd014ad DS |
880 | ++headcount; |
881 | seg += in; | |
882 | } | |
99975cc6 | 883 | heads[headcount - 1].len = cpu_to_vhost32(vq, len + datalen); |
8dd014ad DS |
884 | *iovcount = seg; |
885 | if (unlikely(log)) | |
886 | *log_num = nlogs; | |
d8316f39 MT |
887 | |
888 | /* Detect overrun */ | |
889 | if (unlikely(datalen > 0)) { | |
890 | r = UIO_MAXIOV + 1; | |
891 | goto err; | |
892 | } | |
8dd014ad DS |
893 | return headcount; |
894 | err: | |
895 | vhost_discard_vq_desc(vq, headcount); | |
896 | return r; | |
897 | } | |
898 | ||
3a4d5c94 MT |
899 | /* Expects to be always run from workqueue - which acts as |
900 | * read-size critical section for our kind of RCU. */ | |
94249369 | 901 | static void handle_rx(struct vhost_net *net) |
3a4d5c94 | 902 | { |
81f95a55 MT |
903 | struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_RX]; |
904 | struct vhost_virtqueue *vq = &nvq->vq; | |
8dd014ad DS |
905 | unsigned uninitialized_var(in), log; |
906 | struct vhost_log *vq_log; | |
907 | struct msghdr msg = { | |
908 | .msg_name = NULL, | |
909 | .msg_namelen = 0, | |
910 | .msg_control = NULL, /* FIXME: get and handle RX aux data. */ | |
911 | .msg_controllen = 0, | |
8dd014ad DS |
912 | .msg_flags = MSG_DONTWAIT, |
913 | }; | |
0960b641 JW |
914 | struct virtio_net_hdr hdr = { |
915 | .flags = 0, | |
916 | .gso_type = VIRTIO_NET_HDR_GSO_NONE | |
8dd014ad | 917 | }; |
8dd014ad | 918 | size_t total_len = 0; |
910a578f | 919 | int err, mergeable; |
f5a4941a | 920 | s16 headcount; |
8dd014ad DS |
921 | size_t vhost_hlen, sock_hlen; |
922 | size_t vhost_len, sock_len; | |
be294a51 | 923 | bool busyloop_intr = false; |
2e26af79 | 924 | struct socket *sock; |
ba7438ae | 925 | struct iov_iter fixup; |
0960b641 | 926 | __virtio16 num_buffers; |
db688c24 | 927 | int recv_pkts = 0; |
8dd014ad | 928 | |
aaa3149b | 929 | mutex_lock_nested(&vq->mutex, 0); |
2e26af79 AH |
930 | sock = vq->private_data; |
931 | if (!sock) | |
932 | goto out; | |
6b1e6cc7 JW |
933 | |
934 | if (!vq_iotlb_prefetch(vq)) | |
935 | goto out; | |
936 | ||
8ea8cf89 | 937 | vhost_disable_notify(&net->dev, vq); |
8241a1e4 | 938 | vhost_net_disable_vq(net, vq); |
2e26af79 | 939 | |
81f95a55 MT |
940 | vhost_hlen = nvq->vhost_hlen; |
941 | sock_hlen = nvq->sock_hlen; | |
8dd014ad | 942 | |
ea16c514 | 943 | vq_log = unlikely(vhost_has_feature(vq, VHOST_F_LOG_ALL)) ? |
8dd014ad | 944 | vq->log : NULL; |
ea16c514 | 945 | mergeable = vhost_has_feature(vq, VIRTIO_NET_F_MRG_RXBUF); |
8dd014ad | 946 | |
be294a51 TM |
947 | while ((sock_len = vhost_net_rx_peek_head_len(net, sock->sk, |
948 | &busyloop_intr))) { | |
8dd014ad DS |
949 | sock_len += sock_hlen; |
950 | vhost_len = sock_len + vhost_hlen; | |
f5a4941a JW |
951 | headcount = get_rx_bufs(vq, vq->heads + nvq->done_idx, |
952 | vhost_len, &in, vq_log, &log, | |
94249369 | 953 | likely(mergeable) ? UIO_MAXIOV : 1); |
8dd014ad DS |
954 | /* On error, stop handling until the next kick. */ |
955 | if (unlikely(headcount < 0)) | |
8241a1e4 | 956 | goto out; |
8dd014ad DS |
957 | /* OK, now we need to know about added descriptors. */ |
958 | if (!headcount) { | |
6369fec5 TM |
959 | if (unlikely(busyloop_intr)) { |
960 | vhost_poll_queue(&vq->poll); | |
961 | } else if (unlikely(vhost_enable_notify(&net->dev, vq))) { | |
8dd014ad DS |
962 | /* They have slipped one in as we were |
963 | * doing that: check again. */ | |
8ea8cf89 | 964 | vhost_disable_notify(&net->dev, vq); |
8dd014ad DS |
965 | continue; |
966 | } | |
967 | /* Nothing new? Wait for eventfd to tell us | |
968 | * they refilled. */ | |
8241a1e4 | 969 | goto out; |
8dd014ad | 970 | } |
6369fec5 | 971 | busyloop_intr = false; |
5990a305 | 972 | if (nvq->rx_ring) |
6e474083 WX |
973 | msg.msg_control = vhost_net_buf_consume(&nvq->rxq); |
974 | /* On overrun, truncate and discard */ | |
975 | if (unlikely(headcount > UIO_MAXIOV)) { | |
976 | iov_iter_init(&msg.msg_iter, READ, vq->iov, 1, 1); | |
977 | err = sock->ops->recvmsg(sock, &msg, | |
978 | 1, MSG_DONTWAIT | MSG_TRUNC); | |
979 | pr_debug("Discarded rx packet: len %zd\n", sock_len); | |
980 | continue; | |
981 | } | |
8dd014ad | 982 | /* We don't need to be notified again. */ |
ba7438ae AV |
983 | iov_iter_init(&msg.msg_iter, READ, vq->iov, in, vhost_len); |
984 | fixup = msg.msg_iter; | |
985 | if (unlikely((vhost_hlen))) { | |
986 | /* We will supply the header ourselves | |
987 | * TODO: support TSO. | |
988 | */ | |
989 | iov_iter_advance(&msg.msg_iter, vhost_hlen); | |
ba7438ae | 990 | } |
1b784140 | 991 | err = sock->ops->recvmsg(sock, &msg, |
8dd014ad DS |
992 | sock_len, MSG_DONTWAIT | MSG_TRUNC); |
993 | /* Userspace might have consumed the packet meanwhile: | |
994 | * it's not supposed to do this usually, but might be hard | |
995 | * to prevent. Discard data we got (if any) and keep going. */ | |
996 | if (unlikely(err != sock_len)) { | |
997 | pr_debug("Discarded rx packet: " | |
998 | " len %d, expected %zd\n", err, sock_len); | |
999 | vhost_discard_vq_desc(vq, headcount); | |
1000 | continue; | |
1001 | } | |
ba7438ae | 1002 | /* Supply virtio_net_hdr if VHOST_NET_F_VIRTIO_NET_HDR */ |
4c5a8442 MT |
1003 | if (unlikely(vhost_hlen)) { |
1004 | if (copy_to_iter(&hdr, sizeof(hdr), | |
1005 | &fixup) != sizeof(hdr)) { | |
1006 | vq_err(vq, "Unable to write vnet_hdr " | |
1007 | "at addr %p\n", vq->iov->iov_base); | |
8241a1e4 | 1008 | goto out; |
4c5a8442 MT |
1009 | } |
1010 | } else { | |
1011 | /* Header came from socket; we'll need to patch | |
1012 | * ->num_buffers over if VIRTIO_NET_F_MRG_RXBUF | |
1013 | */ | |
1014 | iov_iter_advance(&fixup, sizeof(hdr)); | |
8dd014ad DS |
1015 | } |
1016 | /* TODO: Should check and handle checksum. */ | |
5201aa49 | 1017 | |
0960b641 | 1018 | num_buffers = cpu_to_vhost16(vq, headcount); |
cfbdab95 | 1019 | if (likely(mergeable) && |
0d79a493 MT |
1020 | copy_to_iter(&num_buffers, sizeof num_buffers, |
1021 | &fixup) != sizeof num_buffers) { | |
8dd014ad DS |
1022 | vq_err(vq, "Failed num_buffers write"); |
1023 | vhost_discard_vq_desc(vq, headcount); | |
8241a1e4 | 1024 | goto out; |
8dd014ad | 1025 | } |
f5a4941a | 1026 | nvq->done_idx += headcount; |
d0d86971 | 1027 | if (nvq->done_idx > VHOST_NET_BATCH) |
09c32489 | 1028 | vhost_net_signal_used(nvq); |
8dd014ad DS |
1029 | if (unlikely(vq_log)) |
1030 | vhost_log_write(vq, vq_log, log, vhost_len); | |
1031 | total_len += vhost_len; | |
272f35cb | 1032 | if (unlikely(vhost_exceeds_weight(++recv_pkts, total_len))) { |
8dd014ad | 1033 | vhost_poll_queue(&vq->poll); |
8241a1e4 | 1034 | goto out; |
8dd014ad DS |
1035 | } |
1036 | } | |
be294a51 TM |
1037 | if (unlikely(busyloop_intr)) |
1038 | vhost_poll_queue(&vq->poll); | |
1039 | else | |
1040 | vhost_net_enable_vq(net, vq); | |
2e26af79 | 1041 | out: |
09c32489 | 1042 | vhost_net_signal_used(nvq); |
8dd014ad | 1043 | mutex_unlock(&vq->mutex); |
8dd014ad DS |
1044 | } |
1045 | ||
c23f3445 | 1046 | static void handle_tx_kick(struct vhost_work *work) |
3a4d5c94 | 1047 | { |
c23f3445 TH |
1048 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
1049 | poll.work); | |
1050 | struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev); | |
1051 | ||
3a4d5c94 MT |
1052 | handle_tx(net); |
1053 | } | |
1054 | ||
c23f3445 | 1055 | static void handle_rx_kick(struct vhost_work *work) |
3a4d5c94 | 1056 | { |
c23f3445 TH |
1057 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
1058 | poll.work); | |
1059 | struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev); | |
1060 | ||
3a4d5c94 MT |
1061 | handle_rx(net); |
1062 | } | |
1063 | ||
c23f3445 | 1064 | static void handle_tx_net(struct vhost_work *work) |
3a4d5c94 | 1065 | { |
c23f3445 TH |
1066 | struct vhost_net *net = container_of(work, struct vhost_net, |
1067 | poll[VHOST_NET_VQ_TX].work); | |
3a4d5c94 MT |
1068 | handle_tx(net); |
1069 | } | |
1070 | ||
c23f3445 | 1071 | static void handle_rx_net(struct vhost_work *work) |
3a4d5c94 | 1072 | { |
c23f3445 TH |
1073 | struct vhost_net *net = container_of(work, struct vhost_net, |
1074 | poll[VHOST_NET_VQ_RX].work); | |
3a4d5c94 MT |
1075 | handle_rx(net); |
1076 | } | |
1077 | ||
1078 | static int vhost_net_open(struct inode *inode, struct file *f) | |
1079 | { | |
23cc5a99 | 1080 | struct vhost_net *n; |
c23f3445 | 1081 | struct vhost_dev *dev; |
3ab2e420 | 1082 | struct vhost_virtqueue **vqs; |
5990a305 | 1083 | void **queue; |
59566b6e | 1084 | int i; |
c23f3445 | 1085 | |
dcda9b04 | 1086 | n = kvmalloc(sizeof *n, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
6c5ab651 MH |
1087 | if (!n) |
1088 | return -ENOMEM; | |
6da2ec56 | 1089 | vqs = kmalloc_array(VHOST_NET_VQ_MAX, sizeof(*vqs), GFP_KERNEL); |
3ab2e420 | 1090 | if (!vqs) { |
d04257b0 | 1091 | kvfree(n); |
3ab2e420 AH |
1092 | return -ENOMEM; |
1093 | } | |
c23f3445 | 1094 | |
d0d86971 | 1095 | queue = kmalloc_array(VHOST_NET_BATCH, sizeof(void *), |
c67df11f JW |
1096 | GFP_KERNEL); |
1097 | if (!queue) { | |
1098 | kfree(vqs); | |
1099 | kvfree(n); | |
1100 | return -ENOMEM; | |
1101 | } | |
1102 | n->vqs[VHOST_NET_VQ_RX].rxq.queue = queue; | |
1103 | ||
c23f3445 | 1104 | dev = &n->dev; |
3ab2e420 AH |
1105 | vqs[VHOST_NET_VQ_TX] = &n->vqs[VHOST_NET_VQ_TX].vq; |
1106 | vqs[VHOST_NET_VQ_RX] = &n->vqs[VHOST_NET_VQ_RX].vq; | |
1107 | n->vqs[VHOST_NET_VQ_TX].vq.handle_kick = handle_tx_kick; | |
1108 | n->vqs[VHOST_NET_VQ_RX].vq.handle_kick = handle_rx_kick; | |
2839400f AH |
1109 | for (i = 0; i < VHOST_NET_VQ_MAX; i++) { |
1110 | n->vqs[i].ubufs = NULL; | |
1111 | n->vqs[i].ubuf_info = NULL; | |
1112 | n->vqs[i].upend_idx = 0; | |
1113 | n->vqs[i].done_idx = 0; | |
81f95a55 MT |
1114 | n->vqs[i].vhost_hlen = 0; |
1115 | n->vqs[i].sock_hlen = 0; | |
ab7e34b3 | 1116 | n->vqs[i].rx_ring = NULL; |
c67df11f | 1117 | vhost_net_buf_init(&n->vqs[i].rxq); |
2839400f | 1118 | } |
59566b6e | 1119 | vhost_dev_init(dev, vqs, VHOST_NET_VQ_MAX); |
3a4d5c94 | 1120 | |
a9a08845 LT |
1121 | vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, EPOLLOUT, dev); |
1122 | vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, EPOLLIN, dev); | |
3a4d5c94 MT |
1123 | |
1124 | f->private_data = n; | |
1125 | ||
1126 | return 0; | |
1127 | } | |
1128 | ||
3a4d5c94 MT |
1129 | static struct socket *vhost_net_stop_vq(struct vhost_net *n, |
1130 | struct vhost_virtqueue *vq) | |
1131 | { | |
1132 | struct socket *sock; | |
c67df11f JW |
1133 | struct vhost_net_virtqueue *nvq = |
1134 | container_of(vq, struct vhost_net_virtqueue, vq); | |
3a4d5c94 MT |
1135 | |
1136 | mutex_lock(&vq->mutex); | |
22fa90c7 | 1137 | sock = vq->private_data; |
3a4d5c94 | 1138 | vhost_net_disable_vq(n, vq); |
22fa90c7 | 1139 | vq->private_data = NULL; |
c67df11f | 1140 | vhost_net_buf_unproduce(nvq); |
303fd71b | 1141 | nvq->rx_ring = NULL; |
3a4d5c94 MT |
1142 | mutex_unlock(&vq->mutex); |
1143 | return sock; | |
1144 | } | |
1145 | ||
1146 | static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock, | |
1147 | struct socket **rx_sock) | |
1148 | { | |
3ab2e420 AH |
1149 | *tx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_TX].vq); |
1150 | *rx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_RX].vq); | |
3a4d5c94 MT |
1151 | } |
1152 | ||
1153 | static void vhost_net_flush_vq(struct vhost_net *n, int index) | |
1154 | { | |
1155 | vhost_poll_flush(n->poll + index); | |
3ab2e420 | 1156 | vhost_poll_flush(&n->vqs[index].vq.poll); |
3a4d5c94 MT |
1157 | } |
1158 | ||
1159 | static void vhost_net_flush(struct vhost_net *n) | |
1160 | { | |
1161 | vhost_net_flush_vq(n, VHOST_NET_VQ_TX); | |
1162 | vhost_net_flush_vq(n, VHOST_NET_VQ_RX); | |
2839400f | 1163 | if (n->vqs[VHOST_NET_VQ_TX].ubufs) { |
3ab2e420 | 1164 | mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 1165 | n->tx_flush = true; |
3ab2e420 | 1166 | mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 1167 | /* Wait for all lower device DMAs done. */ |
fe729a57 | 1168 | vhost_net_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].ubufs); |
3ab2e420 | 1169 | mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 1170 | n->tx_flush = false; |
0ad8b480 | 1171 | atomic_set(&n->vqs[VHOST_NET_VQ_TX].ubufs->refcount, 1); |
3ab2e420 | 1172 | mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 1173 | } |
3a4d5c94 MT |
1174 | } |
1175 | ||
1176 | static int vhost_net_release(struct inode *inode, struct file *f) | |
1177 | { | |
1178 | struct vhost_net *n = f->private_data; | |
1179 | struct socket *tx_sock; | |
1180 | struct socket *rx_sock; | |
1181 | ||
1182 | vhost_net_stop(n, &tx_sock, &rx_sock); | |
1183 | vhost_net_flush(n); | |
b211616d | 1184 | vhost_dev_stop(&n->dev); |
f6f93f75 | 1185 | vhost_dev_cleanup(&n->dev); |
81f95a55 | 1186 | vhost_net_vq_reset(n); |
3a4d5c94 | 1187 | if (tx_sock) |
09aaacf0 | 1188 | sockfd_put(tx_sock); |
3a4d5c94 | 1189 | if (rx_sock) |
09aaacf0 | 1190 | sockfd_put(rx_sock); |
b0c057ca MT |
1191 | /* Make sure no callbacks are outstanding */ |
1192 | synchronize_rcu_bh(); | |
3a4d5c94 MT |
1193 | /* We do an extra flush before freeing memory, |
1194 | * since jobs can re-queue themselves. */ | |
1195 | vhost_net_flush(n); | |
c67df11f | 1196 | kfree(n->vqs[VHOST_NET_VQ_RX].rxq.queue); |
3ab2e420 | 1197 | kfree(n->dev.vqs); |
d04257b0 | 1198 | kvfree(n); |
3a4d5c94 MT |
1199 | return 0; |
1200 | } | |
1201 | ||
1202 | static struct socket *get_raw_socket(int fd) | |
1203 | { | |
1204 | struct { | |
1205 | struct sockaddr_ll sa; | |
1206 | char buf[MAX_ADDR_LEN]; | |
1207 | } uaddr; | |
9b2c45d4 | 1208 | int r; |
3a4d5c94 | 1209 | struct socket *sock = sockfd_lookup(fd, &r); |
d47effe1 | 1210 | |
3a4d5c94 MT |
1211 | if (!sock) |
1212 | return ERR_PTR(-ENOTSOCK); | |
1213 | ||
1214 | /* Parameter checking */ | |
1215 | if (sock->sk->sk_type != SOCK_RAW) { | |
1216 | r = -ESOCKTNOSUPPORT; | |
1217 | goto err; | |
1218 | } | |
1219 | ||
9b2c45d4 DV |
1220 | r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa, 0); |
1221 | if (r < 0) | |
3a4d5c94 MT |
1222 | goto err; |
1223 | ||
1224 | if (uaddr.sa.sll_family != AF_PACKET) { | |
1225 | r = -EPFNOSUPPORT; | |
1226 | goto err; | |
1227 | } | |
1228 | return sock; | |
1229 | err: | |
09aaacf0 | 1230 | sockfd_put(sock); |
3a4d5c94 MT |
1231 | return ERR_PTR(r); |
1232 | } | |
1233 | ||
5990a305 | 1234 | static struct ptr_ring *get_tap_ptr_ring(int fd) |
c67df11f | 1235 | { |
5990a305 | 1236 | struct ptr_ring *ring; |
c67df11f JW |
1237 | struct file *file = fget(fd); |
1238 | ||
1239 | if (!file) | |
1240 | return NULL; | |
5990a305 JW |
1241 | ring = tun_get_tx_ring(file); |
1242 | if (!IS_ERR(ring)) | |
c67df11f | 1243 | goto out; |
5990a305 JW |
1244 | ring = tap_get_ptr_ring(file); |
1245 | if (!IS_ERR(ring)) | |
c67df11f | 1246 | goto out; |
5990a305 | 1247 | ring = NULL; |
c67df11f JW |
1248 | out: |
1249 | fput(file); | |
5990a305 | 1250 | return ring; |
c67df11f JW |
1251 | } |
1252 | ||
501c774c | 1253 | static struct socket *get_tap_socket(int fd) |
3a4d5c94 MT |
1254 | { |
1255 | struct file *file = fget(fd); | |
1256 | struct socket *sock; | |
d47effe1 | 1257 | |
3a4d5c94 MT |
1258 | if (!file) |
1259 | return ERR_PTR(-EBADF); | |
1260 | sock = tun_get_socket(file); | |
501c774c AB |
1261 | if (!IS_ERR(sock)) |
1262 | return sock; | |
635b8c8e | 1263 | sock = tap_get_socket(file); |
3a4d5c94 MT |
1264 | if (IS_ERR(sock)) |
1265 | fput(file); | |
1266 | return sock; | |
1267 | } | |
1268 | ||
1269 | static struct socket *get_socket(int fd) | |
1270 | { | |
1271 | struct socket *sock; | |
d47effe1 | 1272 | |
3a4d5c94 MT |
1273 | /* special case to disable backend */ |
1274 | if (fd == -1) | |
1275 | return NULL; | |
1276 | sock = get_raw_socket(fd); | |
1277 | if (!IS_ERR(sock)) | |
1278 | return sock; | |
501c774c | 1279 | sock = get_tap_socket(fd); |
3a4d5c94 MT |
1280 | if (!IS_ERR(sock)) |
1281 | return sock; | |
1282 | return ERR_PTR(-ENOTSOCK); | |
1283 | } | |
1284 | ||
1285 | static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) | |
1286 | { | |
1287 | struct socket *sock, *oldsock; | |
1288 | struct vhost_virtqueue *vq; | |
2839400f | 1289 | struct vhost_net_virtqueue *nvq; |
fe729a57 | 1290 | struct vhost_net_ubuf_ref *ubufs, *oldubufs = NULL; |
3a4d5c94 MT |
1291 | int r; |
1292 | ||
1293 | mutex_lock(&n->dev.mutex); | |
1294 | r = vhost_dev_check_owner(&n->dev); | |
1295 | if (r) | |
1296 | goto err; | |
1297 | ||
1298 | if (index >= VHOST_NET_VQ_MAX) { | |
1299 | r = -ENOBUFS; | |
1300 | goto err; | |
1301 | } | |
3ab2e420 | 1302 | vq = &n->vqs[index].vq; |
2839400f | 1303 | nvq = &n->vqs[index]; |
3a4d5c94 MT |
1304 | mutex_lock(&vq->mutex); |
1305 | ||
1306 | /* Verify that ring has been setup correctly. */ | |
1307 | if (!vhost_vq_access_ok(vq)) { | |
1308 | r = -EFAULT; | |
1dace8c8 | 1309 | goto err_vq; |
3a4d5c94 MT |
1310 | } |
1311 | sock = get_socket(fd); | |
1312 | if (IS_ERR(sock)) { | |
1313 | r = PTR_ERR(sock); | |
1dace8c8 | 1314 | goto err_vq; |
3a4d5c94 MT |
1315 | } |
1316 | ||
1317 | /* start polling new socket */ | |
22fa90c7 | 1318 | oldsock = vq->private_data; |
11fe8839 | 1319 | if (sock != oldsock) { |
fe729a57 AH |
1320 | ubufs = vhost_net_ubuf_alloc(vq, |
1321 | sock && vhost_sock_zcopy(sock)); | |
bab632d6 MT |
1322 | if (IS_ERR(ubufs)) { |
1323 | r = PTR_ERR(ubufs); | |
1324 | goto err_ubufs; | |
1325 | } | |
692a998b | 1326 | |
d47effe1 | 1327 | vhost_net_disable_vq(n, vq); |
22fa90c7 | 1328 | vq->private_data = sock; |
c67df11f | 1329 | vhost_net_buf_unproduce(nvq); |
80f7d030 | 1330 | r = vhost_vq_init_access(vq); |
f59281da | 1331 | if (r) |
692a998b | 1332 | goto err_used; |
2b8b328b JW |
1333 | r = vhost_net_enable_vq(n, vq); |
1334 | if (r) | |
1335 | goto err_used; | |
303fd71b JW |
1336 | if (index == VHOST_NET_VQ_RX) |
1337 | nvq->rx_ring = get_tap_ptr_ring(fd); | |
692a998b | 1338 | |
2839400f AH |
1339 | oldubufs = nvq->ubufs; |
1340 | nvq->ubufs = ubufs; | |
64e9a9b8 MT |
1341 | |
1342 | n->tx_packets = 0; | |
1343 | n->tx_zcopy_err = 0; | |
1280c27f | 1344 | n->tx_flush = false; |
dd1f4078 | 1345 | } |
3a4d5c94 | 1346 | |
1680e906 MT |
1347 | mutex_unlock(&vq->mutex); |
1348 | ||
c047e5f3 | 1349 | if (oldubufs) { |
c38e39c3 | 1350 | vhost_net_ubuf_put_wait_and_free(oldubufs); |
c047e5f3 | 1351 | mutex_lock(&vq->mutex); |
eaae8132 | 1352 | vhost_zerocopy_signal_used(n, vq); |
c047e5f3 MT |
1353 | mutex_unlock(&vq->mutex); |
1354 | } | |
bab632d6 | 1355 | |
3a4d5c94 MT |
1356 | if (oldsock) { |
1357 | vhost_net_flush_vq(n, index); | |
09aaacf0 | 1358 | sockfd_put(oldsock); |
3a4d5c94 | 1359 | } |
1dace8c8 | 1360 | |
1680e906 MT |
1361 | mutex_unlock(&n->dev.mutex); |
1362 | return 0; | |
1363 | ||
692a998b | 1364 | err_used: |
22fa90c7 | 1365 | vq->private_data = oldsock; |
692a998b JW |
1366 | vhost_net_enable_vq(n, vq); |
1367 | if (ubufs) | |
c38e39c3 | 1368 | vhost_net_ubuf_put_wait_and_free(ubufs); |
bab632d6 | 1369 | err_ubufs: |
b8f1f658 JW |
1370 | if (sock) |
1371 | sockfd_put(sock); | |
1dace8c8 JD |
1372 | err_vq: |
1373 | mutex_unlock(&vq->mutex); | |
3a4d5c94 MT |
1374 | err: |
1375 | mutex_unlock(&n->dev.mutex); | |
1376 | return r; | |
1377 | } | |
1378 | ||
1379 | static long vhost_net_reset_owner(struct vhost_net *n) | |
1380 | { | |
1381 | struct socket *tx_sock = NULL; | |
1382 | struct socket *rx_sock = NULL; | |
1383 | long err; | |
a9709d68 | 1384 | struct vhost_umem *umem; |
d47effe1 | 1385 | |
3a4d5c94 MT |
1386 | mutex_lock(&n->dev.mutex); |
1387 | err = vhost_dev_check_owner(&n->dev); | |
1388 | if (err) | |
1389 | goto done; | |
a9709d68 JW |
1390 | umem = vhost_dev_reset_owner_prepare(); |
1391 | if (!umem) { | |
150b9e51 MT |
1392 | err = -ENOMEM; |
1393 | goto done; | |
1394 | } | |
3a4d5c94 MT |
1395 | vhost_net_stop(n, &tx_sock, &rx_sock); |
1396 | vhost_net_flush(n); | |
4cd87951 | 1397 | vhost_dev_stop(&n->dev); |
a9709d68 | 1398 | vhost_dev_reset_owner(&n->dev, umem); |
81f95a55 | 1399 | vhost_net_vq_reset(n); |
3a4d5c94 MT |
1400 | done: |
1401 | mutex_unlock(&n->dev.mutex); | |
1402 | if (tx_sock) | |
09aaacf0 | 1403 | sockfd_put(tx_sock); |
3a4d5c94 | 1404 | if (rx_sock) |
09aaacf0 | 1405 | sockfd_put(rx_sock); |
3a4d5c94 MT |
1406 | return err; |
1407 | } | |
1408 | ||
429711ae JW |
1409 | static int vhost_net_set_backend_features(struct vhost_net *n, u64 features) |
1410 | { | |
1411 | int i; | |
1412 | ||
1413 | mutex_lock(&n->dev.mutex); | |
1414 | for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { | |
1415 | mutex_lock(&n->vqs[i].vq.mutex); | |
1416 | n->vqs[i].vq.acked_backend_features = features; | |
1417 | mutex_unlock(&n->vqs[i].vq.mutex); | |
1418 | } | |
1419 | mutex_unlock(&n->dev.mutex); | |
1420 | ||
1421 | return 0; | |
1422 | } | |
1423 | ||
3a4d5c94 MT |
1424 | static int vhost_net_set_features(struct vhost_net *n, u64 features) |
1425 | { | |
8dd014ad | 1426 | size_t vhost_hlen, sock_hlen, hdr_len; |
3a4d5c94 | 1427 | int i; |
8dd014ad | 1428 | |
e4fca7d6 MT |
1429 | hdr_len = (features & ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | |
1430 | (1ULL << VIRTIO_F_VERSION_1))) ? | |
8dd014ad DS |
1431 | sizeof(struct virtio_net_hdr_mrg_rxbuf) : |
1432 | sizeof(struct virtio_net_hdr); | |
1433 | if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) { | |
1434 | /* vhost provides vnet_hdr */ | |
1435 | vhost_hlen = hdr_len; | |
1436 | sock_hlen = 0; | |
1437 | } else { | |
1438 | /* socket provides vnet_hdr */ | |
1439 | vhost_hlen = 0; | |
1440 | sock_hlen = hdr_len; | |
1441 | } | |
3a4d5c94 MT |
1442 | mutex_lock(&n->dev.mutex); |
1443 | if ((features & (1 << VHOST_F_LOG_ALL)) && | |
6b1e6cc7 JW |
1444 | !vhost_log_access_ok(&n->dev)) |
1445 | goto out_unlock; | |
1446 | ||
1447 | if ((features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))) { | |
1448 | if (vhost_init_device_iotlb(&n->dev, true)) | |
1449 | goto out_unlock; | |
3a4d5c94 | 1450 | } |
6b1e6cc7 | 1451 | |
3a4d5c94 | 1452 | for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { |
3ab2e420 | 1453 | mutex_lock(&n->vqs[i].vq.mutex); |
ea16c514 | 1454 | n->vqs[i].vq.acked_features = features; |
81f95a55 MT |
1455 | n->vqs[i].vhost_hlen = vhost_hlen; |
1456 | n->vqs[i].sock_hlen = sock_hlen; | |
3ab2e420 | 1457 | mutex_unlock(&n->vqs[i].vq.mutex); |
3a4d5c94 | 1458 | } |
3a4d5c94 MT |
1459 | mutex_unlock(&n->dev.mutex); |
1460 | return 0; | |
6b1e6cc7 JW |
1461 | |
1462 | out_unlock: | |
1463 | mutex_unlock(&n->dev.mutex); | |
1464 | return -EFAULT; | |
3a4d5c94 MT |
1465 | } |
1466 | ||
b1ad8496 AH |
1467 | static long vhost_net_set_owner(struct vhost_net *n) |
1468 | { | |
1469 | int r; | |
1470 | ||
1471 | mutex_lock(&n->dev.mutex); | |
05c05351 MT |
1472 | if (vhost_dev_has_owner(&n->dev)) { |
1473 | r = -EBUSY; | |
1474 | goto out; | |
1475 | } | |
b1ad8496 AH |
1476 | r = vhost_net_set_ubuf_info(n); |
1477 | if (r) | |
1478 | goto out; | |
1479 | r = vhost_dev_set_owner(&n->dev); | |
1480 | if (r) | |
1481 | vhost_net_clear_ubuf_info(n); | |
1482 | vhost_net_flush(n); | |
1483 | out: | |
1484 | mutex_unlock(&n->dev.mutex); | |
1485 | return r; | |
1486 | } | |
1487 | ||
3a4d5c94 MT |
1488 | static long vhost_net_ioctl(struct file *f, unsigned int ioctl, |
1489 | unsigned long arg) | |
1490 | { | |
1491 | struct vhost_net *n = f->private_data; | |
1492 | void __user *argp = (void __user *)arg; | |
1493 | u64 __user *featurep = argp; | |
1494 | struct vhost_vring_file backend; | |
1495 | u64 features; | |
1496 | int r; | |
d47effe1 | 1497 | |
3a4d5c94 MT |
1498 | switch (ioctl) { |
1499 | case VHOST_NET_SET_BACKEND: | |
d3553a52 TY |
1500 | if (copy_from_user(&backend, argp, sizeof backend)) |
1501 | return -EFAULT; | |
3a4d5c94 MT |
1502 | return vhost_net_set_backend(n, backend.index, backend.fd); |
1503 | case VHOST_GET_FEATURES: | |
0dd05a3b | 1504 | features = VHOST_NET_FEATURES; |
d3553a52 TY |
1505 | if (copy_to_user(featurep, &features, sizeof features)) |
1506 | return -EFAULT; | |
1507 | return 0; | |
3a4d5c94 | 1508 | case VHOST_SET_FEATURES: |
d3553a52 TY |
1509 | if (copy_from_user(&features, featurep, sizeof features)) |
1510 | return -EFAULT; | |
0dd05a3b | 1511 | if (features & ~VHOST_NET_FEATURES) |
3a4d5c94 MT |
1512 | return -EOPNOTSUPP; |
1513 | return vhost_net_set_features(n, features); | |
429711ae JW |
1514 | case VHOST_GET_BACKEND_FEATURES: |
1515 | features = VHOST_NET_BACKEND_FEATURES; | |
1516 | if (copy_to_user(featurep, &features, sizeof(features))) | |
1517 | return -EFAULT; | |
1518 | return 0; | |
1519 | case VHOST_SET_BACKEND_FEATURES: | |
1520 | if (copy_from_user(&features, featurep, sizeof(features))) | |
1521 | return -EFAULT; | |
1522 | if (features & ~VHOST_NET_BACKEND_FEATURES) | |
1523 | return -EOPNOTSUPP; | |
1524 | return vhost_net_set_backend_features(n, features); | |
3a4d5c94 MT |
1525 | case VHOST_RESET_OWNER: |
1526 | return vhost_net_reset_owner(n); | |
b1ad8496 AH |
1527 | case VHOST_SET_OWNER: |
1528 | return vhost_net_set_owner(n); | |
3a4d5c94 MT |
1529 | default: |
1530 | mutex_lock(&n->dev.mutex); | |
935cdee7 MT |
1531 | r = vhost_dev_ioctl(&n->dev, ioctl, argp); |
1532 | if (r == -ENOIOCTLCMD) | |
1533 | r = vhost_vring_ioctl(&n->dev, ioctl, argp); | |
1534 | else | |
1535 | vhost_net_flush(n); | |
3a4d5c94 MT |
1536 | mutex_unlock(&n->dev.mutex); |
1537 | return r; | |
1538 | } | |
1539 | } | |
1540 | ||
1541 | #ifdef CONFIG_COMPAT | |
1542 | static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl, | |
1543 | unsigned long arg) | |
1544 | { | |
1545 | return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg)); | |
1546 | } | |
1547 | #endif | |
1548 | ||
6b1e6cc7 JW |
1549 | static ssize_t vhost_net_chr_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1550 | { | |
1551 | struct file *file = iocb->ki_filp; | |
1552 | struct vhost_net *n = file->private_data; | |
1553 | struct vhost_dev *dev = &n->dev; | |
1554 | int noblock = file->f_flags & O_NONBLOCK; | |
1555 | ||
1556 | return vhost_chr_read_iter(dev, to, noblock); | |
1557 | } | |
1558 | ||
1559 | static ssize_t vhost_net_chr_write_iter(struct kiocb *iocb, | |
1560 | struct iov_iter *from) | |
1561 | { | |
1562 | struct file *file = iocb->ki_filp; | |
1563 | struct vhost_net *n = file->private_data; | |
1564 | struct vhost_dev *dev = &n->dev; | |
1565 | ||
1566 | return vhost_chr_write_iter(dev, from); | |
1567 | } | |
1568 | ||
afc9a42b | 1569 | static __poll_t vhost_net_chr_poll(struct file *file, poll_table *wait) |
6b1e6cc7 JW |
1570 | { |
1571 | struct vhost_net *n = file->private_data; | |
1572 | struct vhost_dev *dev = &n->dev; | |
1573 | ||
1574 | return vhost_chr_poll(file, dev, wait); | |
1575 | } | |
1576 | ||
373a83a6 | 1577 | static const struct file_operations vhost_net_fops = { |
3a4d5c94 MT |
1578 | .owner = THIS_MODULE, |
1579 | .release = vhost_net_release, | |
6b1e6cc7 JW |
1580 | .read_iter = vhost_net_chr_read_iter, |
1581 | .write_iter = vhost_net_chr_write_iter, | |
1582 | .poll = vhost_net_chr_poll, | |
3a4d5c94 MT |
1583 | .unlocked_ioctl = vhost_net_ioctl, |
1584 | #ifdef CONFIG_COMPAT | |
1585 | .compat_ioctl = vhost_net_compat_ioctl, | |
1586 | #endif | |
1587 | .open = vhost_net_open, | |
6038f373 | 1588 | .llseek = noop_llseek, |
3a4d5c94 MT |
1589 | }; |
1590 | ||
1591 | static struct miscdevice vhost_net_misc = { | |
7c7c7f01 | 1592 | .minor = VHOST_NET_MINOR, |
1593 | .name = "vhost-net", | |
1594 | .fops = &vhost_net_fops, | |
3a4d5c94 MT |
1595 | }; |
1596 | ||
a8d3782f | 1597 | static int vhost_net_init(void) |
3a4d5c94 | 1598 | { |
bab632d6 | 1599 | if (experimental_zcopytx) |
fe729a57 | 1600 | vhost_net_enable_zcopy(VHOST_NET_VQ_TX); |
c23f3445 | 1601 | return misc_register(&vhost_net_misc); |
3a4d5c94 MT |
1602 | } |
1603 | module_init(vhost_net_init); | |
1604 | ||
a8d3782f | 1605 | static void vhost_net_exit(void) |
3a4d5c94 MT |
1606 | { |
1607 | misc_deregister(&vhost_net_misc); | |
3a4d5c94 MT |
1608 | } |
1609 | module_exit(vhost_net_exit); | |
1610 | ||
1611 | MODULE_VERSION("0.0.1"); | |
1612 | MODULE_LICENSE("GPL v2"); | |
1613 | MODULE_AUTHOR("Michael S. Tsirkin"); | |
1614 | MODULE_DESCRIPTION("Host kernel accelerator for virtio net"); | |
7c7c7f01 | 1615 | MODULE_ALIAS_MISCDEV(VHOST_NET_MINOR); |
1616 | MODULE_ALIAS("devname:vhost-net"); |