diff options
Diffstat (limited to 'drivers')
-rw-r--r-- | drivers/vhost/net.c | 128 |
1 files changed, 7 insertions, 121 deletions
diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c index 9f57cd45fe8..0329c411bbf 100644 --- a/drivers/vhost/net.c +++ b/drivers/vhost/net.c | |||
@@ -229,6 +229,7 @@ static int peek_head_len(struct sock *sk) | |||
229 | * @iovcount - returned count of io vectors we fill | 229 | * @iovcount - returned count of io vectors we fill |
230 | * @log - vhost log | 230 | * @log - vhost log |
231 | * @log_num - log offset | 231 | * @log_num - log offset |
232 | * @quota - headcount quota, 1 for big buffer | ||
232 | * returns number of buffer heads allocated, negative on error | 233 | * returns number of buffer heads allocated, negative on error |
233 | */ | 234 | */ |
234 | static int get_rx_bufs(struct vhost_virtqueue *vq, | 235 | static int get_rx_bufs(struct vhost_virtqueue *vq, |
@@ -236,7 +237,8 @@ static int get_rx_bufs(struct vhost_virtqueue *vq, | |||
236 | int datalen, | 237 | int datalen, |
237 | unsigned *iovcount, | 238 | unsigned *iovcount, |
238 | struct vhost_log *log, | 239 | struct vhost_log *log, |
239 | unsigned *log_num) | 240 | unsigned *log_num, |
241 | unsigned int quota) | ||
240 | { | 242 | { |
241 | unsigned int out, in; | 243 | unsigned int out, in; |
242 | int seg = 0; | 244 | int seg = 0; |
@@ -244,7 +246,7 @@ static int get_rx_bufs(struct vhost_virtqueue *vq, | |||
244 | unsigned d; | 246 | unsigned d; |
245 | int r, nlogs = 0; | 247 | int r, nlogs = 0; |
246 | 248 | ||
247 | while (datalen > 0) { | 249 | while (datalen > 0 && headcount < quota) { |
248 | if (unlikely(seg >= UIO_MAXIOV)) { | 250 | if (unlikely(seg >= UIO_MAXIOV)) { |
249 | r = -ENOBUFS; | 251 | r = -ENOBUFS; |
250 | goto err; | 252 | goto err; |
@@ -284,116 +286,7 @@ err: | |||
284 | 286 | ||
285 | /* Expects to be always run from workqueue - which acts as | 287 | /* Expects to be always run from workqueue - which acts as |
286 | * read-size critical section for our kind of RCU. */ | 288 | * read-size critical section for our kind of RCU. */ |
287 | static void handle_rx_big(struct vhost_net *net) | 289 | static void handle_rx(struct vhost_net *net) |
288 | { | ||
289 | struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_RX]; | ||
290 | unsigned out, in, log, s; | ||
291 | int head; | ||
292 | struct vhost_log *vq_log; | ||
293 | struct msghdr msg = { | ||
294 | .msg_name = NULL, | ||
295 | .msg_namelen = 0, | ||
296 | .msg_control = NULL, /* FIXME: get and handle RX aux data. */ | ||
297 | .msg_controllen = 0, | ||
298 | .msg_iov = vq->iov, | ||
299 | .msg_flags = MSG_DONTWAIT, | ||
300 | }; | ||
301 | struct virtio_net_hdr hdr = { | ||
302 | .flags = 0, | ||
303 | .gso_type = VIRTIO_NET_HDR_GSO_NONE | ||
304 | }; | ||
305 | size_t len, total_len = 0; | ||
306 | int err; | ||
307 | size_t hdr_size; | ||
308 | /* TODO: check that we are running from vhost_worker? */ | ||
309 | struct socket *sock = rcu_dereference_check(vq->private_data, 1); | ||
310 | |||
311 | if (!sock || skb_queue_empty(&sock->sk->sk_receive_queue)) | ||
312 | return; | ||
313 | |||
314 | mutex_lock(&vq->mutex); | ||
315 | vhost_disable_notify(vq); | ||
316 | hdr_size = vq->vhost_hlen; | ||
317 | |||
318 | vq_log = unlikely(vhost_has_feature(&net->dev, VHOST_F_LOG_ALL)) ? | ||
319 | vq->log : NULL; | ||
320 | |||
321 | for (;;) { | ||
322 | head = vhost_get_vq_desc(&net->dev, vq, vq->iov, | ||
323 | ARRAY_SIZE(vq->iov), | ||
324 | &out, &in, | ||
325 | vq_log, &log); | ||
326 | /* On error, stop handling until the next kick. */ | ||
327 | if (unlikely(head < 0)) | ||
328 | break; | ||
329 | /* OK, now we need to know about added descriptors. */ | ||
330 | if (head == vq->num) { | ||
331 | if (unlikely(vhost_enable_notify(vq))) { | ||
332 | /* They have slipped one in as we were | ||
333 | * doing that: check again. */ | ||
334 | vhost_disable_notify(vq); | ||
335 | continue; | ||
336 | } | ||
337 | /* Nothing new? Wait for eventfd to tell us | ||
338 | * they refilled. */ | ||
339 | break; | ||
340 | } | ||
341 | /* We don't need to be notified again. */ | ||
342 | if (out) { | ||
343 | vq_err(vq, "Unexpected descriptor format for RX: " | ||
344 | "out %d, int %d\n", | ||
345 | out, in); | ||
346 | break; | ||
347 | } | ||
348 | /* Skip header. TODO: support TSO/mergeable rx buffers. */ | ||
349 | s = move_iovec_hdr(vq->iov, vq->hdr, hdr_size, in); | ||
350 | msg.msg_iovlen = in; | ||
351 | len = iov_length(vq->iov, in); | ||
352 | /* Sanity check */ | ||
353 | if (!len) { | ||
354 | vq_err(vq, "Unexpected header len for RX: " | ||
355 | "%zd expected %zd\n", | ||
356 | iov_length(vq->hdr, s), hdr_size); | ||
357 | break; | ||
358 | } | ||
359 | err = sock->ops->recvmsg(NULL, sock, &msg, | ||
360 | len, MSG_DONTWAIT | MSG_TRUNC); | ||
361 | /* TODO: Check specific error and bomb out unless EAGAIN? */ | ||
362 | if (err < 0) { | ||
363 | vhost_discard_vq_desc(vq, 1); | ||
364 | break; | ||
365 | } | ||
366 | /* TODO: Should check and handle checksum. */ | ||
367 | if (err > len) { | ||
368 | pr_debug("Discarded truncated rx packet: " | ||
369 | " len %d > %zd\n", err, len); | ||
370 | vhost_discard_vq_desc(vq, 1); | ||
371 | continue; | ||
372 | } | ||
373 | len = err; | ||
374 | err = memcpy_toiovec(vq->hdr, (unsigned char *)&hdr, hdr_size); | ||
375 | if (err) { | ||
376 | vq_err(vq, "Unable to write vnet_hdr at addr %p: %d\n", | ||
377 | vq->iov->iov_base, err); | ||
378 | break; | ||
379 | } | ||
380 | len += hdr_size; | ||
381 | vhost_add_used_and_signal(&net->dev, vq, head, len); | ||
382 | if (unlikely(vq_log)) | ||
383 | vhost_log_write(vq, vq_log, log, len); | ||
384 | total_len += len; | ||
385 | if (unlikely(total_len >= VHOST_NET_WEIGHT)) { | ||
386 | vhost_poll_queue(&vq->poll); | ||
387 | break; | ||
388 | } | ||
389 | } | ||
390 | |||
391 | mutex_unlock(&vq->mutex); | ||
392 | } | ||
393 | |||
394 | /* Expects to be always run from workqueue - which acts as | ||
395 | * read-size critical section for our kind of RCU. */ | ||
396 | static void handle_rx_mergeable(struct vhost_net *net) | ||
397 | { | 290 | { |
398 | struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_RX]; | 291 | struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_RX]; |
399 | unsigned uninitialized_var(in), log; | 292 | unsigned uninitialized_var(in), log; |
@@ -433,7 +326,8 @@ static void handle_rx_mergeable(struct vhost_net *net) | |||
433 | sock_len += sock_hlen; | 326 | sock_len += sock_hlen; |
434 | vhost_len = sock_len + vhost_hlen; | 327 | vhost_len = sock_len + vhost_hlen; |
435 | headcount = get_rx_bufs(vq, vq->heads, vhost_len, | 328 | headcount = get_rx_bufs(vq, vq->heads, vhost_len, |
436 | &in, vq_log, &log); | 329 | &in, vq_log, &log, |
330 | likely(mergeable) ? UIO_MAXIOV : 1); | ||
437 | /* On error, stop handling until the next kick. */ | 331 | /* On error, stop handling until the next kick. */ |
438 | if (unlikely(headcount < 0)) | 332 | if (unlikely(headcount < 0)) |
439 | break; | 333 | break; |
@@ -499,14 +393,6 @@ static void handle_rx_mergeable(struct vhost_net *net) | |||
499 | mutex_unlock(&vq->mutex); | 393 | mutex_unlock(&vq->mutex); |
500 | } | 394 | } |
501 | 395 | ||
502 | static void handle_rx(struct vhost_net *net) | ||
503 | { | ||
504 | if (vhost_has_feature(&net->dev, VIRTIO_NET_F_MRG_RXBUF)) | ||
505 | handle_rx_mergeable(net); | ||
506 | else | ||
507 | handle_rx_big(net); | ||
508 | } | ||
509 | |||
510 | static void handle_tx_kick(struct vhost_work *work) | 396 | static void handle_tx_kick(struct vhost_work *work) |
511 | { | 397 | { |
512 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | 398 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |