Version:  2.0.40 2.2.26 2.4.37 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4

Linux/drivers/vhost/net.c

  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>
 13 #include <linux/miscdevice.h>
 14 #include <linux/module.h>
 15 #include <linux/moduleparam.h>
 16 #include <linux/mutex.h>
 17 #include <linux/workqueue.h>
 18 #include <linux/file.h>
 19 #include <linux/slab.h>
 20 #include <linux/vmalloc.h>
 21 
 22 #include <linux/net.h>
 23 #include <linux/if_packet.h>
 24 #include <linux/if_arp.h>
 25 #include <linux/if_tun.h>
 26 #include <linux/if_macvlan.h>
 27 #include <linux/if_vlan.h>
 28 
 29 #include <net/sock.h>
 30 
 31 #include "vhost.h"
 32 
 33 static int experimental_zcopytx = 1;
 34 module_param(experimental_zcopytx, int, 0444);
 35 MODULE_PARM_DESC(experimental_zcopytx, "Enable Zero Copy TX;"
 36                                        " 1 -Enable; 0 - Disable");
 37 
 38 /* 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 
 42 /* MAX number of TX used buffers for outstanding zerocopy */
 43 #define VHOST_MAX_PEND 128
 44 #define VHOST_GOODCOPY_LEN 256
 45 
 46 /*
 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 */
 51 #define VHOST_DMA_FAILED_LEN    ((__force __virtio32)3)
 52 /* Lower device DMA done */
 53 #define VHOST_DMA_DONE_LEN      ((__force __virtio32)2)
 54 /* Lower device DMA in progress */
 55 #define VHOST_DMA_IN_PROGRESS   ((__force __virtio32)1)
 56 /* Buffer unused */
 57 #define VHOST_DMA_CLEAR_LEN     ((__force __virtio32)0)
 58 
 59 #define VHOST_DMA_IS_DONE(len) ((__force u32)(len) >= (__force u32)VHOST_DMA_DONE_LEN)
 60 
 61 enum {
 62         VHOST_NET_FEATURES = VHOST_FEATURES |
 63                          (1ULL << VHOST_NET_F_VIRTIO_NET_HDR) |
 64                          (1ULL << VIRTIO_NET_F_MRG_RXBUF)
 65 };
 66 
 67 enum {
 68         VHOST_NET_VQ_RX = 0,
 69         VHOST_NET_VQ_TX = 1,
 70         VHOST_NET_VQ_MAX = 2,
 71 };
 72 
 73 struct vhost_net_ubuf_ref {
 74         /* refcount follows semantics similar to kref:
 75          *  0: object is released
 76          *  1: no outstanding ubufs
 77          * >1: outstanding ubufs
 78          */
 79         atomic_t refcount;
 80         wait_queue_head_t wait;
 81         struct vhost_virtqueue *vq;
 82 };
 83 
 84 struct vhost_net_virtqueue {
 85         struct vhost_virtqueue vq;
 86         size_t vhost_hlen;
 87         size_t sock_hlen;
 88         /* vhost zerocopy support fields below: */
 89         /* last used idx for outstanding DMA zerocopy buffers */
 90         int upend_idx;
 91         /* first used idx for DMA done zerocopy buffers */
 92         int done_idx;
 93         /* an array of userspace buffers info */
 94         struct ubuf_info *ubuf_info;
 95         /* Reference counting for outstanding ubufs.
 96          * Protected by vq mutex. Writers must also take device mutex. */
 97         struct vhost_net_ubuf_ref *ubufs;
 98 };
 99 
100 struct vhost_net {
101         struct vhost_dev dev;
102         struct vhost_net_virtqueue vqs[VHOST_NET_VQ_MAX];
103         struct vhost_poll poll[VHOST_NET_VQ_MAX];
104         /* Number of TX recently submitted.
105          * Protected by tx vq lock. */
106         unsigned tx_packets;
107         /* Number of times zerocopy TX recently failed.
108          * Protected by tx vq lock. */
109         unsigned tx_zcopy_err;
110         /* Flush in progress. Protected by tx vq lock. */
111         bool tx_flush;
112 };
113 
114 static unsigned vhost_net_zcopy_mask __read_mostly;
115 
116 static void vhost_net_enable_zcopy(int vq)
117 {
118         vhost_net_zcopy_mask |= 0x1 << vq;
119 }
120 
121 static struct vhost_net_ubuf_ref *
122 vhost_net_ubuf_alloc(struct vhost_virtqueue *vq, bool zcopy)
123 {
124         struct vhost_net_ubuf_ref *ubufs;
125         /* No zero copy backend? Nothing to count. */
126         if (!zcopy)
127                 return NULL;
128         ubufs = kmalloc(sizeof(*ubufs), GFP_KERNEL);
129         if (!ubufs)
130                 return ERR_PTR(-ENOMEM);
131         atomic_set(&ubufs->refcount, 1);
132         init_waitqueue_head(&ubufs->wait);
133         ubufs->vq = vq;
134         return ubufs;
135 }
136 
137 static int vhost_net_ubuf_put(struct vhost_net_ubuf_ref *ubufs)
138 {
139         int r = atomic_sub_return(1, &ubufs->refcount);
140         if (unlikely(!r))
141                 wake_up(&ubufs->wait);
142         return r;
143 }
144 
145 static void vhost_net_ubuf_put_and_wait(struct vhost_net_ubuf_ref *ubufs)
146 {
147         vhost_net_ubuf_put(ubufs);
148         wait_event(ubufs->wait, !atomic_read(&ubufs->refcount));
149 }
150 
151 static void vhost_net_ubuf_put_wait_and_free(struct vhost_net_ubuf_ref *ubufs)
152 {
153         vhost_net_ubuf_put_and_wait(ubufs);
154         kfree(ubufs);
155 }
156 
157 static void vhost_net_clear_ubuf_info(struct vhost_net *n)
158 {
159         int i;
160 
161         for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
162                 kfree(n->vqs[i].ubuf_info);
163                 n->vqs[i].ubuf_info = NULL;
164         }
165 }
166 
167 static int vhost_net_set_ubuf_info(struct vhost_net *n)
168 {
169         bool zcopy;
170         int i;
171 
172         for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
173                 zcopy = vhost_net_zcopy_mask & (0x1 << i);
174                 if (!zcopy)
175                         continue;
176                 n->vqs[i].ubuf_info = kmalloc(sizeof(*n->vqs[i].ubuf_info) *
177                                               UIO_MAXIOV, GFP_KERNEL);
178                 if  (!n->vqs[i].ubuf_info)
179                         goto err;
180         }
181         return 0;
182 
183 err:
184         vhost_net_clear_ubuf_info(n);
185         return -ENOMEM;
186 }
187 
188 static void vhost_net_vq_reset(struct vhost_net *n)
189 {
190         int i;
191 
192         vhost_net_clear_ubuf_info(n);
193 
194         for (i = 0; i < VHOST_NET_VQ_MAX; i++) {
195                 n->vqs[i].done_idx = 0;
196                 n->vqs[i].upend_idx = 0;
197                 n->vqs[i].ubufs = NULL;
198                 n->vqs[i].vhost_hlen = 0;
199                 n->vqs[i].sock_hlen = 0;
200         }
201 
202 }
203 
204 static void vhost_net_tx_packet(struct vhost_net *net)
205 {
206         ++net->tx_packets;
207         if (net->tx_packets < 1024)
208                 return;
209         net->tx_packets = 0;
210         net->tx_zcopy_err = 0;
211 }
212 
213 static void vhost_net_tx_err(struct vhost_net *net)
214 {
215         ++net->tx_zcopy_err;
216 }
217 
218 static bool vhost_net_tx_select_zcopy(struct vhost_net *net)
219 {
220         /* TX flush waits for outstanding DMAs to be done.
221          * Don't start new DMAs.
222          */
223         return !net->tx_flush &&
224                 net->tx_packets / 64 >= net->tx_zcopy_err;
225 }
226 
227 static bool vhost_sock_zcopy(struct socket *sock)
228 {
229         return unlikely(experimental_zcopytx) &&
230                 sock_flag(sock->sk, SOCK_ZEROCOPY);
231 }
232 
233 /* In case of DMA done not in order in lower device driver for some reason.
234  * upend_idx is used to track end of used idx, done_idx is used to track head
235  * of used idx. Once lower device DMA done contiguously, we will signal KVM
236  * guest used idx.
237  */
238 static void vhost_zerocopy_signal_used(struct vhost_net *net,
239                                        struct vhost_virtqueue *vq)
240 {
241         struct vhost_net_virtqueue *nvq =
242                 container_of(vq, struct vhost_net_virtqueue, vq);
243         int i, add;
244         int j = 0;
245 
246         for (i = nvq->done_idx; i != nvq->upend_idx; i = (i + 1) % UIO_MAXIOV) {
247                 if (vq->heads[i].len == VHOST_DMA_FAILED_LEN)
248                         vhost_net_tx_err(net);
249                 if (VHOST_DMA_IS_DONE(vq->heads[i].len)) {
250                         vq->heads[i].len = VHOST_DMA_CLEAR_LEN;
251                         ++j;
252                 } else
253                         break;
254         }
255         while (j) {
256                 add = min(UIO_MAXIOV - nvq->done_idx, j);
257                 vhost_add_used_and_signal_n(vq->dev, vq,
258                                             &vq->heads[nvq->done_idx], add);
259                 nvq->done_idx = (nvq->done_idx + add) % UIO_MAXIOV;
260                 j -= add;
261         }
262 }
263 
264 static void vhost_zerocopy_callback(struct ubuf_info *ubuf, bool success)
265 {
266         struct vhost_net_ubuf_ref *ubufs = ubuf->ctx;
267         struct vhost_virtqueue *vq = ubufs->vq;
268         int cnt;
269 
270         rcu_read_lock_bh();
271 
272         /* set len to mark this desc buffers done DMA */
273         vq->heads[ubuf->desc].len = success ?
274                 VHOST_DMA_DONE_LEN : VHOST_DMA_FAILED_LEN;
275         cnt = vhost_net_ubuf_put(ubufs);
276 
277         /*
278          * Trigger polling thread if guest stopped submitting new buffers:
279          * in this case, the refcount after decrement will eventually reach 1.
280          * We also trigger polling periodically after each 16 packets
281          * (the value 16 here is more or less arbitrary, it's tuned to trigger
282          * less than 10% of times).
283          */
284         if (cnt <= 1 || !(cnt % 16))
285                 vhost_poll_queue(&vq->poll);
286 
287         rcu_read_unlock_bh();
288 }
289 
290 /* Expects to be always run from workqueue - which acts as
291  * read-size critical section for our kind of RCU. */
292 static void handle_tx(struct vhost_net *net)
293 {
294         struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX];
295         struct vhost_virtqueue *vq = &nvq->vq;
296         unsigned out, in;
297         int head;
298         struct msghdr msg = {
299                 .msg_name = NULL,
300                 .msg_namelen = 0,
301                 .msg_control = NULL,
302                 .msg_controllen = 0,
303                 .msg_flags = MSG_DONTWAIT,
304         };
305         size_t len, total_len = 0;
306         int err;
307         size_t hdr_size;
308         struct socket *sock;
309         struct vhost_net_ubuf_ref *uninitialized_var(ubufs);
310         bool zcopy, zcopy_used;
311 
312         mutex_lock(&vq->mutex);
313         sock = vq->private_data;
314         if (!sock)
315                 goto out;
316 
317         vhost_disable_notify(&net->dev, vq);
318 
319         hdr_size = nvq->vhost_hlen;
320         zcopy = nvq->ubufs;
321 
322         for (;;) {
323                 /* Release DMAs done buffers first */
324                 if (zcopy)
325                         vhost_zerocopy_signal_used(net, vq);
326 
327                 /* If more outstanding DMAs, queue the work.
328                  * Handle upend_idx wrap around
329                  */
330                 if (unlikely((nvq->upend_idx + vq->num - VHOST_MAX_PEND)
331                               % UIO_MAXIOV == nvq->done_idx))
332                         break;
333 
334                 head = vhost_get_vq_desc(vq, vq->iov,
335                                          ARRAY_SIZE(vq->iov),
336                                          &out, &in,
337                                          NULL, NULL);
338                 /* On error, stop handling until the next kick. */
339                 if (unlikely(head < 0))
340                         break;
341                 /* Nothing new?  Wait for eventfd to tell us they refilled. */
342                 if (head == vq->num) {
343                         if (unlikely(vhost_enable_notify(&net->dev, vq))) {
344                                 vhost_disable_notify(&net->dev, vq);
345                                 continue;
346                         }
347                         break;
348                 }
349                 if (in) {
350                         vq_err(vq, "Unexpected descriptor format for TX: "
351                                "out %d, int %d\n", out, in);
352                         break;
353                 }
354                 /* Skip header. TODO: support TSO. */
355                 len = iov_length(vq->iov, out);
356                 iov_iter_init(&msg.msg_iter, WRITE, vq->iov, out, len);
357                 iov_iter_advance(&msg.msg_iter, hdr_size);
358                 /* Sanity check */
359                 if (!msg_data_left(&msg)) {
360                         vq_err(vq, "Unexpected header len for TX: "
361                                "%zd expected %zd\n",
362                                len, hdr_size);
363                         break;
364                 }
365                 len = msg_data_left(&msg);
366 
367                 zcopy_used = zcopy && len >= VHOST_GOODCOPY_LEN
368                                    && (nvq->upend_idx + 1) % UIO_MAXIOV !=
369                                       nvq->done_idx
370                                    && vhost_net_tx_select_zcopy(net);
371 
372                 /* use msg_control to pass vhost zerocopy ubuf info to skb */
373                 if (zcopy_used) {
374                         struct ubuf_info *ubuf;
375                         ubuf = nvq->ubuf_info + nvq->upend_idx;
376 
377                         vq->heads[nvq->upend_idx].id = cpu_to_vhost32(vq, head);
378                         vq->heads[nvq->upend_idx].len = VHOST_DMA_IN_PROGRESS;
379                         ubuf->callback = vhost_zerocopy_callback;
380                         ubuf->ctx = nvq->ubufs;
381                         ubuf->desc = nvq->upend_idx;
382                         msg.msg_control = ubuf;
383                         msg.msg_controllen = sizeof(ubuf);
384                         ubufs = nvq->ubufs;
385                         atomic_inc(&ubufs->refcount);
386                         nvq->upend_idx = (nvq->upend_idx + 1) % UIO_MAXIOV;
387                 } else {
388                         msg.msg_control = NULL;
389                         ubufs = NULL;
390                 }
391                 /* TODO: Check specific error and bomb out unless ENOBUFS? */
392                 err = sock->ops->sendmsg(sock, &msg, len);
393                 if (unlikely(err < 0)) {
394                         if (zcopy_used) {
395                                 vhost_net_ubuf_put(ubufs);
396                                 nvq->upend_idx = ((unsigned)nvq->upend_idx - 1)
397                                         % UIO_MAXIOV;
398                         }
399                         vhost_discard_vq_desc(vq, 1);
400                         break;
401                 }
402                 if (err != len)
403                         pr_debug("Truncated TX packet: "
404                                  " len %d != %zd\n", err, len);
405                 if (!zcopy_used)
406                         vhost_add_used_and_signal(&net->dev, vq, head, 0);
407                 else
408                         vhost_zerocopy_signal_used(net, vq);
409                 total_len += len;
410                 vhost_net_tx_packet(net);
411                 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
412                         vhost_poll_queue(&vq->poll);
413                         break;
414                 }
415         }
416 out:
417         mutex_unlock(&vq->mutex);
418 }
419 
420 static int peek_head_len(struct sock *sk)
421 {
422         struct sk_buff *head;
423         int len = 0;
424         unsigned long flags;
425 
426         spin_lock_irqsave(&sk->sk_receive_queue.lock, flags);
427         head = skb_peek(&sk->sk_receive_queue);
428         if (likely(head)) {
429                 len = head->len;
430                 if (skb_vlan_tag_present(head))
431                         len += VLAN_HLEN;
432         }
433 
434         spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags);
435         return len;
436 }
437 
438 /* This is a multi-buffer version of vhost_get_desc, that works if
439  *      vq has read descriptors only.
440  * @vq          - the relevant virtqueue
441  * @datalen     - data length we'll be reading
442  * @iovcount    - returned count of io vectors we fill
443  * @log         - vhost log
444  * @log_num     - log offset
445  * @quota       - headcount quota, 1 for big buffer
446  *      returns number of buffer heads allocated, negative on error
447  */
448 static int get_rx_bufs(struct vhost_virtqueue *vq,
449                        struct vring_used_elem *heads,
450                        int datalen,
451                        unsigned *iovcount,
452                        struct vhost_log *log,
453                        unsigned *log_num,
454                        unsigned int quota)
455 {
456         unsigned int out, in;
457         int seg = 0;
458         int headcount = 0;
459         unsigned d;
460         int r, nlogs = 0;
461         /* len is always initialized before use since we are always called with
462          * datalen > 0.
463          */
464         u32 uninitialized_var(len);
465 
466         while (datalen > 0 && headcount < quota) {
467                 if (unlikely(seg >= UIO_MAXIOV)) {
468                         r = -ENOBUFS;
469                         goto err;
470                 }
471                 r = vhost_get_vq_desc(vq, vq->iov + seg,
472                                       ARRAY_SIZE(vq->iov) - seg, &out,
473                                       &in, log, log_num);
474                 if (unlikely(r < 0))
475                         goto err;
476 
477                 d = r;
478                 if (d == vq->num) {
479                         r = 0;
480                         goto err;
481                 }
482                 if (unlikely(out || in <= 0)) {
483                         vq_err(vq, "unexpected descriptor format for RX: "
484                                 "out %d, in %d\n", out, in);
485                         r = -EINVAL;
486                         goto err;
487                 }
488                 if (unlikely(log)) {
489                         nlogs += *log_num;
490                         log += *log_num;
491                 }
492                 heads[headcount].id = cpu_to_vhost32(vq, d);
493                 len = iov_length(vq->iov + seg, in);
494                 heads[headcount].len = cpu_to_vhost32(vq, len);
495                 datalen -= len;
496                 ++headcount;
497                 seg += in;
498         }
499         heads[headcount - 1].len = cpu_to_vhost32(vq, len + datalen);
500         *iovcount = seg;
501         if (unlikely(log))
502                 *log_num = nlogs;
503 
504         /* Detect overrun */
505         if (unlikely(datalen > 0)) {
506                 r = UIO_MAXIOV + 1;
507                 goto err;
508         }
509         return headcount;
510 err:
511         vhost_discard_vq_desc(vq, headcount);
512         return r;
513 }
514 
515 /* Expects to be always run from workqueue - which acts as
516  * read-size critical section for our kind of RCU. */
517 static void handle_rx(struct vhost_net *net)
518 {
519         struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_RX];
520         struct vhost_virtqueue *vq = &nvq->vq;
521         unsigned uninitialized_var(in), log;
522         struct vhost_log *vq_log;
523         struct msghdr msg = {
524                 .msg_name = NULL,
525                 .msg_namelen = 0,
526                 .msg_control = NULL, /* FIXME: get and handle RX aux data. */
527                 .msg_controllen = 0,
528                 .msg_flags = MSG_DONTWAIT,
529         };
530         struct virtio_net_hdr hdr = {
531                 .flags = 0,
532                 .gso_type = VIRTIO_NET_HDR_GSO_NONE
533         };
534         size_t total_len = 0;
535         int err, mergeable;
536         s16 headcount;
537         size_t vhost_hlen, sock_hlen;
538         size_t vhost_len, sock_len;
539         struct socket *sock;
540         struct iov_iter fixup;
541         __virtio16 num_buffers;
542 
543         mutex_lock(&vq->mutex);
544         sock = vq->private_data;
545         if (!sock)
546                 goto out;
547         vhost_disable_notify(&net->dev, vq);
548 
549         vhost_hlen = nvq->vhost_hlen;
550         sock_hlen = nvq->sock_hlen;
551 
552         vq_log = unlikely(vhost_has_feature(vq, VHOST_F_LOG_ALL)) ?
553                 vq->log : NULL;
554         mergeable = vhost_has_feature(vq, VIRTIO_NET_F_MRG_RXBUF);
555 
556         while ((sock_len = peek_head_len(sock->sk))) {
557                 sock_len += sock_hlen;
558                 vhost_len = sock_len + vhost_hlen;
559                 headcount = get_rx_bufs(vq, vq->heads, vhost_len,
560                                         &in, vq_log, &log,
561                                         likely(mergeable) ? UIO_MAXIOV : 1);
562                 /* On error, stop handling until the next kick. */
563                 if (unlikely(headcount < 0))
564                         break;
565                 /* On overrun, truncate and discard */
566                 if (unlikely(headcount > UIO_MAXIOV)) {
567                         iov_iter_init(&msg.msg_iter, READ, vq->iov, 1, 1);
568                         err = sock->ops->recvmsg(sock, &msg,
569                                                  1, MSG_DONTWAIT | MSG_TRUNC);
570                         pr_debug("Discarded rx packet: len %zd\n", sock_len);
571                         continue;
572                 }
573                 /* OK, now we need to know about added descriptors. */
574                 if (!headcount) {
575                         if (unlikely(vhost_enable_notify(&net->dev, vq))) {
576                                 /* They have slipped one in as we were
577                                  * doing that: check again. */
578                                 vhost_disable_notify(&net->dev, vq);
579                                 continue;
580                         }
581                         /* Nothing new?  Wait for eventfd to tell us
582                          * they refilled. */
583                         break;
584                 }
585                 /* We don't need to be notified again. */
586                 iov_iter_init(&msg.msg_iter, READ, vq->iov, in, vhost_len);
587                 fixup = msg.msg_iter;
588                 if (unlikely((vhost_hlen))) {
589                         /* We will supply the header ourselves
590                          * TODO: support TSO.
591                          */
592                         iov_iter_advance(&msg.msg_iter, vhost_hlen);
593                 }
594                 err = sock->ops->recvmsg(sock, &msg,
595                                          sock_len, MSG_DONTWAIT | MSG_TRUNC);
596                 /* Userspace might have consumed the packet meanwhile:
597                  * it's not supposed to do this usually, but might be hard
598                  * to prevent. Discard data we got (if any) and keep going. */
599                 if (unlikely(err != sock_len)) {
600                         pr_debug("Discarded rx packet: "
601                                  " len %d, expected %zd\n", err, sock_len);
602                         vhost_discard_vq_desc(vq, headcount);
603                         continue;
604                 }
605                 /* Supply virtio_net_hdr if VHOST_NET_F_VIRTIO_NET_HDR */
606                 if (unlikely(vhost_hlen)) {
607                         if (copy_to_iter(&hdr, sizeof(hdr),
608                                          &fixup) != sizeof(hdr)) {
609                                 vq_err(vq, "Unable to write vnet_hdr "
610                                        "at addr %p\n", vq->iov->iov_base);
611                                 break;
612                         }
613                 } else {
614                         /* Header came from socket; we'll need to patch
615                          * ->num_buffers over if VIRTIO_NET_F_MRG_RXBUF
616                          */
617                         iov_iter_advance(&fixup, sizeof(hdr));
618                 }
619                 /* TODO: Should check and handle checksum. */
620 
621                 num_buffers = cpu_to_vhost16(vq, headcount);
622                 if (likely(mergeable) &&
623                     copy_to_iter(&num_buffers, sizeof num_buffers,
624                                  &fixup) != sizeof num_buffers) {
625                         vq_err(vq, "Failed num_buffers write");
626                         vhost_discard_vq_desc(vq, headcount);
627                         break;
628                 }
629                 vhost_add_used_and_signal_n(&net->dev, vq, vq->heads,
630                                             headcount);
631                 if (unlikely(vq_log))
632                         vhost_log_write(vq, vq_log, log, vhost_len);
633                 total_len += vhost_len;
634                 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
635                         vhost_poll_queue(&vq->poll);
636                         break;
637                 }
638         }
639 out:
640         mutex_unlock(&vq->mutex);
641 }
642 
643 static void handle_tx_kick(struct vhost_work *work)
644 {
645         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
646                                                   poll.work);
647         struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
648 
649         handle_tx(net);
650 }
651 
652 static void handle_rx_kick(struct vhost_work *work)
653 {
654         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
655                                                   poll.work);
656         struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
657 
658         handle_rx(net);
659 }
660 
661 static void handle_tx_net(struct vhost_work *work)
662 {
663         struct vhost_net *net = container_of(work, struct vhost_net,
664                                              poll[VHOST_NET_VQ_TX].work);
665         handle_tx(net);
666 }
667 
668 static void handle_rx_net(struct vhost_work *work)
669 {
670         struct vhost_net *net = container_of(work, struct vhost_net,
671                                              poll[VHOST_NET_VQ_RX].work);
672         handle_rx(net);
673 }
674 
675 static int vhost_net_open(struct inode *inode, struct file *f)
676 {
677         struct vhost_net *n;
678         struct vhost_dev *dev;
679         struct vhost_virtqueue **vqs;
680         int i;
681 
682         n = kmalloc(sizeof *n, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
683         if (!n) {
684                 n = vmalloc(sizeof *n);
685                 if (!n)
686                         return -ENOMEM;
687         }
688         vqs = kmalloc(VHOST_NET_VQ_MAX * sizeof(*vqs), GFP_KERNEL);
689         if (!vqs) {
690                 kvfree(n);
691                 return -ENOMEM;
692         }
693 
694         dev = &n->dev;
695         vqs[VHOST_NET_VQ_TX] = &n->vqs[VHOST_NET_VQ_TX].vq;
696         vqs[VHOST_NET_VQ_RX] = &n->vqs[VHOST_NET_VQ_RX].vq;
697         n->vqs[VHOST_NET_VQ_TX].vq.handle_kick = handle_tx_kick;
698         n->vqs[VHOST_NET_VQ_RX].vq.handle_kick = handle_rx_kick;
699         for (i = 0; i < VHOST_NET_VQ_MAX; i++) {
700                 n->vqs[i].ubufs = NULL;
701                 n->vqs[i].ubuf_info = NULL;
702                 n->vqs[i].upend_idx = 0;
703                 n->vqs[i].done_idx = 0;
704                 n->vqs[i].vhost_hlen = 0;
705                 n->vqs[i].sock_hlen = 0;
706         }
707         vhost_dev_init(dev, vqs, VHOST_NET_VQ_MAX);
708 
709         vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT, dev);
710         vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN, dev);
711 
712         f->private_data = n;
713 
714         return 0;
715 }
716 
717 static void vhost_net_disable_vq(struct vhost_net *n,
718                                  struct vhost_virtqueue *vq)
719 {
720         struct vhost_net_virtqueue *nvq =
721                 container_of(vq, struct vhost_net_virtqueue, vq);
722         struct vhost_poll *poll = n->poll + (nvq - n->vqs);
723         if (!vq->private_data)
724                 return;
725         vhost_poll_stop(poll);
726 }
727 
728 static int vhost_net_enable_vq(struct vhost_net *n,
729                                 struct vhost_virtqueue *vq)
730 {
731         struct vhost_net_virtqueue *nvq =
732                 container_of(vq, struct vhost_net_virtqueue, vq);
733         struct vhost_poll *poll = n->poll + (nvq - n->vqs);
734         struct socket *sock;
735 
736         sock = vq->private_data;
737         if (!sock)
738                 return 0;
739 
740         return vhost_poll_start(poll, sock->file);
741 }
742 
743 static struct socket *vhost_net_stop_vq(struct vhost_net *n,
744                                         struct vhost_virtqueue *vq)
745 {
746         struct socket *sock;
747 
748         mutex_lock(&vq->mutex);
749         sock = vq->private_data;
750         vhost_net_disable_vq(n, vq);
751         vq->private_data = NULL;
752         mutex_unlock(&vq->mutex);
753         return sock;
754 }
755 
756 static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock,
757                            struct socket **rx_sock)
758 {
759         *tx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_TX].vq);
760         *rx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_RX].vq);
761 }
762 
763 static void vhost_net_flush_vq(struct vhost_net *n, int index)
764 {
765         vhost_poll_flush(n->poll + index);
766         vhost_poll_flush(&n->vqs[index].vq.poll);
767 }
768 
769 static void vhost_net_flush(struct vhost_net *n)
770 {
771         vhost_net_flush_vq(n, VHOST_NET_VQ_TX);
772         vhost_net_flush_vq(n, VHOST_NET_VQ_RX);
773         if (n->vqs[VHOST_NET_VQ_TX].ubufs) {
774                 mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
775                 n->tx_flush = true;
776                 mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
777                 /* Wait for all lower device DMAs done. */
778                 vhost_net_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].ubufs);
779                 mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
780                 n->tx_flush = false;
781                 atomic_set(&n->vqs[VHOST_NET_VQ_TX].ubufs->refcount, 1);
782                 mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
783         }
784 }
785 
786 static int vhost_net_release(struct inode *inode, struct file *f)
787 {
788         struct vhost_net *n = f->private_data;
789         struct socket *tx_sock;
790         struct socket *rx_sock;
791 
792         vhost_net_stop(n, &tx_sock, &rx_sock);
793         vhost_net_flush(n);
794         vhost_dev_stop(&n->dev);
795         vhost_dev_cleanup(&n->dev, false);
796         vhost_net_vq_reset(n);
797         if (tx_sock)
798                 sockfd_put(tx_sock);
799         if (rx_sock)
800                 sockfd_put(rx_sock);
801         /* Make sure no callbacks are outstanding */
802         synchronize_rcu_bh();
803         /* We do an extra flush before freeing memory,
804          * since jobs can re-queue themselves. */
805         vhost_net_flush(n);
806         kfree(n->dev.vqs);
807         kvfree(n);
808         return 0;
809 }
810 
811 static struct socket *get_raw_socket(int fd)
812 {
813         struct {
814                 struct sockaddr_ll sa;
815                 char  buf[MAX_ADDR_LEN];
816         } uaddr;
817         int uaddr_len = sizeof uaddr, r;
818         struct socket *sock = sockfd_lookup(fd, &r);
819 
820         if (!sock)
821                 return ERR_PTR(-ENOTSOCK);
822 
823         /* Parameter checking */
824         if (sock->sk->sk_type != SOCK_RAW) {
825                 r = -ESOCKTNOSUPPORT;
826                 goto err;
827         }
828 
829         r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa,
830                                &uaddr_len, 0);
831         if (r)
832                 goto err;
833 
834         if (uaddr.sa.sll_family != AF_PACKET) {
835                 r = -EPFNOSUPPORT;
836                 goto err;
837         }
838         return sock;
839 err:
840         sockfd_put(sock);
841         return ERR_PTR(r);
842 }
843 
844 static struct socket *get_tap_socket(int fd)
845 {
846         struct file *file = fget(fd);
847         struct socket *sock;
848 
849         if (!file)
850                 return ERR_PTR(-EBADF);
851         sock = tun_get_socket(file);
852         if (!IS_ERR(sock))
853                 return sock;
854         sock = macvtap_get_socket(file);
855         if (IS_ERR(sock))
856                 fput(file);
857         return sock;
858 }
859 
860 static struct socket *get_socket(int fd)
861 {
862         struct socket *sock;
863 
864         /* special case to disable backend */
865         if (fd == -1)
866                 return NULL;
867         sock = get_raw_socket(fd);
868         if (!IS_ERR(sock))
869                 return sock;
870         sock = get_tap_socket(fd);
871         if (!IS_ERR(sock))
872                 return sock;
873         return ERR_PTR(-ENOTSOCK);
874 }
875 
876 static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
877 {
878         struct socket *sock, *oldsock;
879         struct vhost_virtqueue *vq;
880         struct vhost_net_virtqueue *nvq;
881         struct vhost_net_ubuf_ref *ubufs, *oldubufs = NULL;
882         int r;
883 
884         mutex_lock(&n->dev.mutex);
885         r = vhost_dev_check_owner(&n->dev);
886         if (r)
887                 goto err;
888 
889         if (index >= VHOST_NET_VQ_MAX) {
890                 r = -ENOBUFS;
891                 goto err;
892         }
893         vq = &n->vqs[index].vq;
894         nvq = &n->vqs[index];
895         mutex_lock(&vq->mutex);
896 
897         /* Verify that ring has been setup correctly. */
898         if (!vhost_vq_access_ok(vq)) {
899                 r = -EFAULT;
900                 goto err_vq;
901         }
902         sock = get_socket(fd);
903         if (IS_ERR(sock)) {
904                 r = PTR_ERR(sock);
905                 goto err_vq;
906         }
907 
908         /* start polling new socket */
909         oldsock = vq->private_data;
910         if (sock != oldsock) {
911                 ubufs = vhost_net_ubuf_alloc(vq,
912                                              sock && vhost_sock_zcopy(sock));
913                 if (IS_ERR(ubufs)) {
914                         r = PTR_ERR(ubufs);
915                         goto err_ubufs;
916                 }
917 
918                 vhost_net_disable_vq(n, vq);
919                 vq->private_data = sock;
920                 r = vhost_init_used(vq);
921                 if (r)
922                         goto err_used;
923                 r = vhost_net_enable_vq(n, vq);
924                 if (r)
925                         goto err_used;
926 
927                 oldubufs = nvq->ubufs;
928                 nvq->ubufs = ubufs;
929 
930                 n->tx_packets = 0;
931                 n->tx_zcopy_err = 0;
932                 n->tx_flush = false;
933         }
934 
935         mutex_unlock(&vq->mutex);
936 
937         if (oldubufs) {
938                 vhost_net_ubuf_put_wait_and_free(oldubufs);
939                 mutex_lock(&vq->mutex);
940                 vhost_zerocopy_signal_used(n, vq);
941                 mutex_unlock(&vq->mutex);
942         }
943 
944         if (oldsock) {
945                 vhost_net_flush_vq(n, index);
946                 sockfd_put(oldsock);
947         }
948 
949         mutex_unlock(&n->dev.mutex);
950         return 0;
951 
952 err_used:
953         vq->private_data = oldsock;
954         vhost_net_enable_vq(n, vq);
955         if (ubufs)
956                 vhost_net_ubuf_put_wait_and_free(ubufs);
957 err_ubufs:
958         sockfd_put(sock);
959 err_vq:
960         mutex_unlock(&vq->mutex);
961 err:
962         mutex_unlock(&n->dev.mutex);
963         return r;
964 }
965 
966 static long vhost_net_reset_owner(struct vhost_net *n)
967 {
968         struct socket *tx_sock = NULL;
969         struct socket *rx_sock = NULL;
970         long err;
971         struct vhost_memory *memory;
972 
973         mutex_lock(&n->dev.mutex);
974         err = vhost_dev_check_owner(&n->dev);
975         if (err)
976                 goto done;
977         memory = vhost_dev_reset_owner_prepare();
978         if (!memory) {
979                 err = -ENOMEM;
980                 goto done;
981         }
982         vhost_net_stop(n, &tx_sock, &rx_sock);
983         vhost_net_flush(n);
984         vhost_dev_reset_owner(&n->dev, memory);
985         vhost_net_vq_reset(n);
986 done:
987         mutex_unlock(&n->dev.mutex);
988         if (tx_sock)
989                 sockfd_put(tx_sock);
990         if (rx_sock)
991                 sockfd_put(rx_sock);
992         return err;
993 }
994 
995 static int vhost_net_set_features(struct vhost_net *n, u64 features)
996 {
997         size_t vhost_hlen, sock_hlen, hdr_len;
998         int i;
999 
1000         hdr_len = (features & ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
1001                                (1ULL << VIRTIO_F_VERSION_1))) ?
1002                         sizeof(struct virtio_net_hdr_mrg_rxbuf) :
1003                         sizeof(struct virtio_net_hdr);
1004         if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) {
1005                 /* vhost provides vnet_hdr */
1006                 vhost_hlen = hdr_len;
1007                 sock_hlen = 0;
1008         } else {
1009                 /* socket provides vnet_hdr */
1010                 vhost_hlen = 0;
1011                 sock_hlen = hdr_len;
1012         }
1013         mutex_lock(&n->dev.mutex);
1014         if ((features & (1 << VHOST_F_LOG_ALL)) &&
1015             !vhost_log_access_ok(&n->dev)) {
1016                 mutex_unlock(&n->dev.mutex);
1017                 return -EFAULT;
1018         }
1019         for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
1020                 mutex_lock(&n->vqs[i].vq.mutex);
1021                 n->vqs[i].vq.acked_features = features;
1022                 n->vqs[i].vhost_hlen = vhost_hlen;
1023                 n->vqs[i].sock_hlen = sock_hlen;
1024                 mutex_unlock(&n->vqs[i].vq.mutex);
1025         }
1026         mutex_unlock(&n->dev.mutex);
1027         return 0;
1028 }
1029 
1030 static long vhost_net_set_owner(struct vhost_net *n)
1031 {
1032         int r;
1033 
1034         mutex_lock(&n->dev.mutex);
1035         if (vhost_dev_has_owner(&n->dev)) {
1036                 r = -EBUSY;
1037                 goto out;
1038         }
1039         r = vhost_net_set_ubuf_info(n);
1040         if (r)
1041                 goto out;
1042         r = vhost_dev_set_owner(&n->dev);
1043         if (r)
1044                 vhost_net_clear_ubuf_info(n);
1045         vhost_net_flush(n);
1046 out:
1047         mutex_unlock(&n->dev.mutex);
1048         return r;
1049 }
1050 
1051 static long vhost_net_ioctl(struct file *f, unsigned int ioctl,
1052                             unsigned long arg)
1053 {
1054         struct vhost_net *n = f->private_data;
1055         void __user *argp = (void __user *)arg;
1056         u64 __user *featurep = argp;
1057         struct vhost_vring_file backend;
1058         u64 features;
1059         int r;
1060 
1061         switch (ioctl) {
1062         case VHOST_NET_SET_BACKEND:
1063                 if (copy_from_user(&backend, argp, sizeof backend))
1064                         return -EFAULT;
1065                 return vhost_net_set_backend(n, backend.index, backend.fd);
1066         case VHOST_GET_FEATURES:
1067                 features = VHOST_NET_FEATURES;
1068                 if (copy_to_user(featurep, &features, sizeof features))
1069                         return -EFAULT;
1070                 return 0;
1071         case VHOST_SET_FEATURES:
1072                 if (copy_from_user(&features, featurep, sizeof features))
1073                         return -EFAULT;
1074                 if (features & ~VHOST_NET_FEATURES)
1075                         return -EOPNOTSUPP;
1076                 return vhost_net_set_features(n, features);
1077         case VHOST_RESET_OWNER:
1078                 return vhost_net_reset_owner(n);
1079         case VHOST_SET_OWNER:
1080                 return vhost_net_set_owner(n);
1081         default:
1082                 mutex_lock(&n->dev.mutex);
1083                 r = vhost_dev_ioctl(&n->dev, ioctl, argp);
1084                 if (r == -ENOIOCTLCMD)
1085                         r = vhost_vring_ioctl(&n->dev, ioctl, argp);
1086                 else
1087                         vhost_net_flush(n);
1088                 mutex_unlock(&n->dev.mutex);
1089                 return r;
1090         }
1091 }
1092 
1093 #ifdef CONFIG_COMPAT
1094 static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl,
1095                                    unsigned long arg)
1096 {
1097         return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1098 }
1099 #endif
1100 
1101 static const struct file_operations vhost_net_fops = {
1102         .owner          = THIS_MODULE,
1103         .release        = vhost_net_release,
1104         .unlocked_ioctl = vhost_net_ioctl,
1105 #ifdef CONFIG_COMPAT
1106         .compat_ioctl   = vhost_net_compat_ioctl,
1107 #endif
1108         .open           = vhost_net_open,
1109         .llseek         = noop_llseek,
1110 };
1111 
1112 static struct miscdevice vhost_net_misc = {
1113         .minor = VHOST_NET_MINOR,
1114         .name = "vhost-net",
1115         .fops = &vhost_net_fops,
1116 };
1117 
1118 static int vhost_net_init(void)
1119 {
1120         if (experimental_zcopytx)
1121                 vhost_net_enable_zcopy(VHOST_NET_VQ_TX);
1122         return misc_register(&vhost_net_misc);
1123 }
1124 module_init(vhost_net_init);
1125 
1126 static void vhost_net_exit(void)
1127 {
1128         misc_deregister(&vhost_net_misc);
1129 }
1130 module_exit(vhost_net_exit);
1131 
1132 MODULE_VERSION("0.0.1");
1133 MODULE_LICENSE("GPL v2");
1134 MODULE_AUTHOR("Michael S. Tsirkin");
1135 MODULE_DESCRIPTION("Host kernel accelerator for virtio net");
1136 MODULE_ALIAS_MISCDEV(VHOST_NET_MINOR);
1137 MODULE_ALIAS("devname:vhost-net");
1138 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us