]>
Commit | Line | Data |
---|---|---|
433fc58e AH |
1 | /* |
2 | * vhost transport for vsock | |
3 | * | |
4 | * Copyright (C) 2013-2015 Red Hat, Inc. | |
5 | * Author: Asias He <asias@redhat.com> | |
6 | * Stefan Hajnoczi <stefanha@redhat.com> | |
7 | * | |
8 | * This work is licensed under the terms of the GNU GPL, version 2. | |
9 | */ | |
10 | #include <linux/miscdevice.h> | |
11 | #include <linux/atomic.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/mutex.h> | |
14 | #include <linux/vmalloc.h> | |
15 | #include <net/sock.h> | |
16 | #include <linux/virtio_vsock.h> | |
17 | #include <linux/vhost.h> | |
18 | ||
19 | #include <net/af_vsock.h> | |
20 | #include "vhost.h" | |
21 | ||
22 | #define VHOST_VSOCK_DEFAULT_HOST_CID 2 | |
23 | ||
24 | enum { | |
25 | VHOST_VSOCK_FEATURES = VHOST_FEATURES, | |
26 | }; | |
27 | ||
28 | /* Used to track all the vhost_vsock instances on the system. */ | |
29 | static DEFINE_SPINLOCK(vhost_vsock_lock); | |
30 | static LIST_HEAD(vhost_vsock_list); | |
31 | ||
32 | struct vhost_vsock { | |
33 | struct vhost_dev dev; | |
34 | struct vhost_virtqueue vqs[2]; | |
35 | ||
36 | /* Link to global vhost_vsock_list, protected by vhost_vsock_lock */ | |
37 | struct list_head list; | |
38 | ||
39 | struct vhost_work send_pkt_work; | |
40 | spinlock_t send_pkt_list_lock; | |
41 | struct list_head send_pkt_list; /* host->guest pending packets */ | |
42 | ||
43 | atomic_t queued_replies; | |
44 | ||
45 | u32 guest_cid; | |
46 | }; | |
47 | ||
48 | static u32 vhost_transport_get_local_cid(void) | |
49 | { | |
50 | return VHOST_VSOCK_DEFAULT_HOST_CID; | |
51 | } | |
52 | ||
6c083c2b | 53 | static struct vhost_vsock *__vhost_vsock_get(u32 guest_cid) |
433fc58e AH |
54 | { |
55 | struct vhost_vsock *vsock; | |
56 | ||
433fc58e AH |
57 | list_for_each_entry(vsock, &vhost_vsock_list, list) { |
58 | u32 other_cid = vsock->guest_cid; | |
59 | ||
60 | /* Skip instances that have no CID yet */ | |
61 | if (other_cid == 0) | |
62 | continue; | |
63 | ||
64 | if (other_cid == guest_cid) { | |
433fc58e AH |
65 | return vsock; |
66 | } | |
67 | } | |
433fc58e AH |
68 | |
69 | return NULL; | |
70 | } | |
71 | ||
6c083c2b G |
72 | static struct vhost_vsock *vhost_vsock_get(u32 guest_cid) |
73 | { | |
74 | struct vhost_vsock *vsock; | |
75 | ||
76 | spin_lock_bh(&vhost_vsock_lock); | |
77 | vsock = __vhost_vsock_get(guest_cid); | |
78 | spin_unlock_bh(&vhost_vsock_lock); | |
79 | ||
80 | return vsock; | |
81 | } | |
82 | ||
433fc58e AH |
83 | static void |
84 | vhost_transport_do_send_pkt(struct vhost_vsock *vsock, | |
85 | struct vhost_virtqueue *vq) | |
86 | { | |
87 | struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX]; | |
88 | bool added = false; | |
89 | bool restart_tx = false; | |
90 | ||
91 | mutex_lock(&vq->mutex); | |
92 | ||
93 | if (!vq->private_data) | |
94 | goto out; | |
95 | ||
96 | /* Avoid further vmexits, we're already processing the virtqueue */ | |
97 | vhost_disable_notify(&vsock->dev, vq); | |
98 | ||
99 | for (;;) { | |
100 | struct virtio_vsock_pkt *pkt; | |
101 | struct iov_iter iov_iter; | |
102 | unsigned out, in; | |
103 | size_t nbytes; | |
104 | size_t len; | |
105 | int head; | |
106 | ||
107 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
108 | if (list_empty(&vsock->send_pkt_list)) { | |
109 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
110 | vhost_enable_notify(&vsock->dev, vq); | |
111 | break; | |
112 | } | |
113 | ||
114 | pkt = list_first_entry(&vsock->send_pkt_list, | |
115 | struct virtio_vsock_pkt, list); | |
116 | list_del_init(&pkt->list); | |
117 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
118 | ||
119 | head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), | |
120 | &out, &in, NULL, NULL); | |
121 | if (head < 0) { | |
122 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
123 | list_add(&pkt->list, &vsock->send_pkt_list); | |
124 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
125 | break; | |
126 | } | |
127 | ||
128 | if (head == vq->num) { | |
129 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
130 | list_add(&pkt->list, &vsock->send_pkt_list); | |
131 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
132 | ||
133 | /* We cannot finish yet if more buffers snuck in while | |
134 | * re-enabling notify. | |
135 | */ | |
136 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { | |
137 | vhost_disable_notify(&vsock->dev, vq); | |
138 | continue; | |
139 | } | |
140 | break; | |
141 | } | |
142 | ||
143 | if (out) { | |
144 | virtio_transport_free_pkt(pkt); | |
145 | vq_err(vq, "Expected 0 output buffers, got %u\n", out); | |
146 | break; | |
147 | } | |
148 | ||
149 | len = iov_length(&vq->iov[out], in); | |
150 | iov_iter_init(&iov_iter, READ, &vq->iov[out], in, len); | |
151 | ||
152 | nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); | |
153 | if (nbytes != sizeof(pkt->hdr)) { | |
154 | virtio_transport_free_pkt(pkt); | |
155 | vq_err(vq, "Faulted on copying pkt hdr\n"); | |
156 | break; | |
157 | } | |
158 | ||
159 | nbytes = copy_to_iter(pkt->buf, pkt->len, &iov_iter); | |
160 | if (nbytes != pkt->len) { | |
161 | virtio_transport_free_pkt(pkt); | |
162 | vq_err(vq, "Faulted on copying pkt buf\n"); | |
163 | break; | |
164 | } | |
165 | ||
166 | vhost_add_used(vq, head, sizeof(pkt->hdr) + pkt->len); | |
167 | added = true; | |
168 | ||
169 | if (pkt->reply) { | |
170 | int val; | |
171 | ||
172 | val = atomic_dec_return(&vsock->queued_replies); | |
173 | ||
174 | /* Do we have resources to resume tx processing? */ | |
175 | if (val + 1 == tx_vq->num) | |
176 | restart_tx = true; | |
177 | } | |
178 | ||
179 | virtio_transport_free_pkt(pkt); | |
180 | } | |
181 | if (added) | |
182 | vhost_signal(&vsock->dev, vq); | |
183 | ||
184 | out: | |
185 | mutex_unlock(&vq->mutex); | |
186 | ||
187 | if (restart_tx) | |
188 | vhost_poll_queue(&tx_vq->poll); | |
189 | } | |
190 | ||
191 | static void vhost_transport_send_pkt_work(struct vhost_work *work) | |
192 | { | |
193 | struct vhost_virtqueue *vq; | |
194 | struct vhost_vsock *vsock; | |
195 | ||
196 | vsock = container_of(work, struct vhost_vsock, send_pkt_work); | |
197 | vq = &vsock->vqs[VSOCK_VQ_RX]; | |
198 | ||
199 | vhost_transport_do_send_pkt(vsock, vq); | |
200 | } | |
201 | ||
202 | static int | |
203 | vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt) | |
204 | { | |
205 | struct vhost_vsock *vsock; | |
433fc58e AH |
206 | int len = pkt->len; |
207 | ||
208 | /* Find the vhost_vsock according to guest context id */ | |
209 | vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid)); | |
210 | if (!vsock) { | |
211 | virtio_transport_free_pkt(pkt); | |
212 | return -ENODEV; | |
213 | } | |
214 | ||
433fc58e AH |
215 | if (pkt->reply) |
216 | atomic_inc(&vsock->queued_replies); | |
217 | ||
218 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
219 | list_add_tail(&pkt->list, &vsock->send_pkt_list); | |
220 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
221 | ||
222 | vhost_work_queue(&vsock->dev, &vsock->send_pkt_work); | |
223 | return len; | |
224 | } | |
225 | ||
16320f36 PT |
226 | static int |
227 | vhost_transport_cancel_pkt(struct vsock_sock *vsk) | |
228 | { | |
229 | struct vhost_vsock *vsock; | |
230 | struct virtio_vsock_pkt *pkt, *n; | |
231 | int cnt = 0; | |
232 | LIST_HEAD(freeme); | |
233 | ||
234 | /* Find the vhost_vsock according to guest context id */ | |
235 | vsock = vhost_vsock_get(vsk->remote_addr.svm_cid); | |
236 | if (!vsock) | |
237 | return -ENODEV; | |
238 | ||
239 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
240 | list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) { | |
241 | if (pkt->vsk != vsk) | |
242 | continue; | |
243 | list_move(&pkt->list, &freeme); | |
244 | } | |
245 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
246 | ||
247 | list_for_each_entry_safe(pkt, n, &freeme, list) { | |
248 | if (pkt->reply) | |
249 | cnt++; | |
250 | list_del(&pkt->list); | |
251 | virtio_transport_free_pkt(pkt); | |
252 | } | |
253 | ||
254 | if (cnt) { | |
255 | struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX]; | |
256 | int new_cnt; | |
257 | ||
258 | new_cnt = atomic_sub_return(cnt, &vsock->queued_replies); | |
259 | if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num) | |
260 | vhost_poll_queue(&tx_vq->poll); | |
261 | } | |
262 | ||
263 | return 0; | |
264 | } | |
265 | ||
433fc58e AH |
266 | static struct virtio_vsock_pkt * |
267 | vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq, | |
268 | unsigned int out, unsigned int in) | |
269 | { | |
270 | struct virtio_vsock_pkt *pkt; | |
271 | struct iov_iter iov_iter; | |
272 | size_t nbytes; | |
273 | size_t len; | |
274 | ||
275 | if (in != 0) { | |
276 | vq_err(vq, "Expected 0 input buffers, got %u\n", in); | |
277 | return NULL; | |
278 | } | |
279 | ||
280 | pkt = kzalloc(sizeof(*pkt), GFP_KERNEL); | |
281 | if (!pkt) | |
282 | return NULL; | |
283 | ||
284 | len = iov_length(vq->iov, out); | |
285 | iov_iter_init(&iov_iter, WRITE, vq->iov, out, len); | |
286 | ||
287 | nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); | |
288 | if (nbytes != sizeof(pkt->hdr)) { | |
289 | vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n", | |
290 | sizeof(pkt->hdr), nbytes); | |
291 | kfree(pkt); | |
292 | return NULL; | |
293 | } | |
294 | ||
295 | if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM) | |
296 | pkt->len = le32_to_cpu(pkt->hdr.len); | |
297 | ||
298 | /* No payload */ | |
299 | if (!pkt->len) | |
300 | return pkt; | |
301 | ||
302 | /* The pkt is too big */ | |
303 | if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) { | |
304 | kfree(pkt); | |
305 | return NULL; | |
306 | } | |
307 | ||
308 | pkt->buf = kmalloc(pkt->len, GFP_KERNEL); | |
309 | if (!pkt->buf) { | |
310 | kfree(pkt); | |
311 | return NULL; | |
312 | } | |
313 | ||
314 | nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter); | |
315 | if (nbytes != pkt->len) { | |
316 | vq_err(vq, "Expected %u byte payload, got %zu bytes\n", | |
317 | pkt->len, nbytes); | |
318 | virtio_transport_free_pkt(pkt); | |
319 | return NULL; | |
320 | } | |
321 | ||
322 | return pkt; | |
323 | } | |
324 | ||
325 | /* Is there space left for replies to rx packets? */ | |
326 | static bool vhost_vsock_more_replies(struct vhost_vsock *vsock) | |
327 | { | |
328 | struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX]; | |
329 | int val; | |
330 | ||
331 | smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */ | |
332 | val = atomic_read(&vsock->queued_replies); | |
333 | ||
334 | return val < vq->num; | |
335 | } | |
336 | ||
337 | static void vhost_vsock_handle_tx_kick(struct vhost_work *work) | |
338 | { | |
339 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
340 | poll.work); | |
341 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
342 | dev); | |
343 | struct virtio_vsock_pkt *pkt; | |
344 | int head; | |
345 | unsigned int out, in; | |
346 | bool added = false; | |
347 | ||
348 | mutex_lock(&vq->mutex); | |
349 | ||
350 | if (!vq->private_data) | |
351 | goto out; | |
352 | ||
353 | vhost_disable_notify(&vsock->dev, vq); | |
354 | for (;;) { | |
3fda5d6e SH |
355 | u32 len; |
356 | ||
433fc58e AH |
357 | if (!vhost_vsock_more_replies(vsock)) { |
358 | /* Stop tx until the device processes already | |
359 | * pending replies. Leave tx virtqueue | |
360 | * callbacks disabled. | |
361 | */ | |
362 | goto no_more_replies; | |
363 | } | |
364 | ||
365 | head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), | |
366 | &out, &in, NULL, NULL); | |
367 | if (head < 0) | |
368 | break; | |
369 | ||
370 | if (head == vq->num) { | |
371 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { | |
372 | vhost_disable_notify(&vsock->dev, vq); | |
373 | continue; | |
374 | } | |
375 | break; | |
376 | } | |
377 | ||
378 | pkt = vhost_vsock_alloc_pkt(vq, out, in); | |
379 | if (!pkt) { | |
380 | vq_err(vq, "Faulted on pkt\n"); | |
381 | continue; | |
382 | } | |
383 | ||
3fda5d6e SH |
384 | len = pkt->len; |
385 | ||
433fc58e AH |
386 | /* Only accept correctly addressed packets */ |
387 | if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid) | |
388 | virtio_transport_recv_pkt(pkt); | |
389 | else | |
390 | virtio_transport_free_pkt(pkt); | |
391 | ||
3fda5d6e | 392 | vhost_add_used(vq, head, sizeof(pkt->hdr) + len); |
433fc58e AH |
393 | added = true; |
394 | } | |
395 | ||
396 | no_more_replies: | |
397 | if (added) | |
398 | vhost_signal(&vsock->dev, vq); | |
399 | ||
400 | out: | |
401 | mutex_unlock(&vq->mutex); | |
402 | } | |
403 | ||
404 | static void vhost_vsock_handle_rx_kick(struct vhost_work *work) | |
405 | { | |
406 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
407 | poll.work); | |
408 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
409 | dev); | |
410 | ||
411 | vhost_transport_do_send_pkt(vsock, vq); | |
412 | } | |
413 | ||
414 | static int vhost_vsock_start(struct vhost_vsock *vsock) | |
415 | { | |
0516ffd8 | 416 | struct vhost_virtqueue *vq; |
433fc58e AH |
417 | size_t i; |
418 | int ret; | |
419 | ||
420 | mutex_lock(&vsock->dev.mutex); | |
421 | ||
422 | ret = vhost_dev_check_owner(&vsock->dev); | |
423 | if (ret) | |
424 | goto err; | |
425 | ||
426 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
0516ffd8 | 427 | vq = &vsock->vqs[i]; |
433fc58e AH |
428 | |
429 | mutex_lock(&vq->mutex); | |
430 | ||
431 | if (!vhost_vq_access_ok(vq)) { | |
432 | ret = -EFAULT; | |
433fc58e AH |
433 | goto err_vq; |
434 | } | |
435 | ||
436 | if (!vq->private_data) { | |
437 | vq->private_data = vsock; | |
0516ffd8 SH |
438 | ret = vhost_vq_init_access(vq); |
439 | if (ret) | |
440 | goto err_vq; | |
433fc58e AH |
441 | } |
442 | ||
443 | mutex_unlock(&vq->mutex); | |
444 | } | |
445 | ||
446 | mutex_unlock(&vsock->dev.mutex); | |
447 | return 0; | |
448 | ||
449 | err_vq: | |
0516ffd8 SH |
450 | vq->private_data = NULL; |
451 | mutex_unlock(&vq->mutex); | |
452 | ||
433fc58e | 453 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
0516ffd8 | 454 | vq = &vsock->vqs[i]; |
433fc58e AH |
455 | |
456 | mutex_lock(&vq->mutex); | |
457 | vq->private_data = NULL; | |
458 | mutex_unlock(&vq->mutex); | |
459 | } | |
460 | err: | |
461 | mutex_unlock(&vsock->dev.mutex); | |
462 | return ret; | |
463 | } | |
464 | ||
465 | static int vhost_vsock_stop(struct vhost_vsock *vsock) | |
466 | { | |
467 | size_t i; | |
468 | int ret; | |
469 | ||
470 | mutex_lock(&vsock->dev.mutex); | |
471 | ||
472 | ret = vhost_dev_check_owner(&vsock->dev); | |
473 | if (ret) | |
474 | goto err; | |
475 | ||
476 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
477 | struct vhost_virtqueue *vq = &vsock->vqs[i]; | |
478 | ||
479 | mutex_lock(&vq->mutex); | |
480 | vq->private_data = NULL; | |
481 | mutex_unlock(&vq->mutex); | |
482 | } | |
483 | ||
484 | err: | |
485 | mutex_unlock(&vsock->dev.mutex); | |
486 | return ret; | |
487 | } | |
488 | ||
489 | static void vhost_vsock_free(struct vhost_vsock *vsock) | |
490 | { | |
b226acab | 491 | kvfree(vsock); |
433fc58e AH |
492 | } |
493 | ||
494 | static int vhost_vsock_dev_open(struct inode *inode, struct file *file) | |
495 | { | |
496 | struct vhost_virtqueue **vqs; | |
497 | struct vhost_vsock *vsock; | |
498 | int ret; | |
499 | ||
500 | /* This struct is large and allocation could fail, fall back to vmalloc | |
501 | * if there is no other way. | |
502 | */ | |
503 | vsock = kzalloc(sizeof(*vsock), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); | |
504 | if (!vsock) { | |
505 | vsock = vmalloc(sizeof(*vsock)); | |
506 | if (!vsock) | |
507 | return -ENOMEM; | |
508 | } | |
509 | ||
510 | vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL); | |
511 | if (!vqs) { | |
512 | ret = -ENOMEM; | |
513 | goto out; | |
514 | } | |
515 | ||
516 | atomic_set(&vsock->queued_replies, 0); | |
517 | ||
518 | vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX]; | |
519 | vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX]; | |
520 | vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick; | |
521 | vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick; | |
522 | ||
523 | vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs)); | |
524 | ||
525 | file->private_data = vsock; | |
526 | spin_lock_init(&vsock->send_pkt_list_lock); | |
527 | INIT_LIST_HEAD(&vsock->send_pkt_list); | |
528 | vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work); | |
529 | ||
530 | spin_lock_bh(&vhost_vsock_lock); | |
531 | list_add_tail(&vsock->list, &vhost_vsock_list); | |
532 | spin_unlock_bh(&vhost_vsock_lock); | |
533 | return 0; | |
534 | ||
535 | out: | |
536 | vhost_vsock_free(vsock); | |
537 | return ret; | |
538 | } | |
539 | ||
540 | static void vhost_vsock_flush(struct vhost_vsock *vsock) | |
541 | { | |
542 | int i; | |
543 | ||
544 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) | |
545 | if (vsock->vqs[i].handle_kick) | |
546 | vhost_poll_flush(&vsock->vqs[i].poll); | |
547 | vhost_work_flush(&vsock->dev, &vsock->send_pkt_work); | |
548 | } | |
549 | ||
550 | static void vhost_vsock_reset_orphans(struct sock *sk) | |
551 | { | |
552 | struct vsock_sock *vsk = vsock_sk(sk); | |
553 | ||
554 | /* vmci_transport.c doesn't take sk_lock here either. At least we're | |
555 | * under vsock_table_lock so the sock cannot disappear while we're | |
556 | * executing. | |
557 | */ | |
558 | ||
c4587631 | 559 | if (!vhost_vsock_get(vsk->remote_addr.svm_cid)) { |
433fc58e AH |
560 | sock_set_flag(sk, SOCK_DONE); |
561 | vsk->peer_shutdown = SHUTDOWN_MASK; | |
562 | sk->sk_state = SS_UNCONNECTED; | |
563 | sk->sk_err = ECONNRESET; | |
564 | sk->sk_error_report(sk); | |
565 | } | |
566 | } | |
567 | ||
568 | static int vhost_vsock_dev_release(struct inode *inode, struct file *file) | |
569 | { | |
570 | struct vhost_vsock *vsock = file->private_data; | |
571 | ||
572 | spin_lock_bh(&vhost_vsock_lock); | |
573 | list_del(&vsock->list); | |
574 | spin_unlock_bh(&vhost_vsock_lock); | |
575 | ||
576 | /* Iterating over all connections for all CIDs to find orphans is | |
577 | * inefficient. Room for improvement here. */ | |
578 | vsock_for_each_connected_socket(vhost_vsock_reset_orphans); | |
579 | ||
580 | vhost_vsock_stop(vsock); | |
581 | vhost_vsock_flush(vsock); | |
582 | vhost_dev_stop(&vsock->dev); | |
583 | ||
584 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
585 | while (!list_empty(&vsock->send_pkt_list)) { | |
586 | struct virtio_vsock_pkt *pkt; | |
587 | ||
588 | pkt = list_first_entry(&vsock->send_pkt_list, | |
589 | struct virtio_vsock_pkt, list); | |
590 | list_del_init(&pkt->list); | |
591 | virtio_transport_free_pkt(pkt); | |
592 | } | |
593 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
594 | ||
595 | vhost_dev_cleanup(&vsock->dev, false); | |
596 | kfree(vsock->dev.vqs); | |
597 | vhost_vsock_free(vsock); | |
598 | return 0; | |
599 | } | |
600 | ||
601 | static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid) | |
602 | { | |
603 | struct vhost_vsock *other; | |
604 | ||
605 | /* Refuse reserved CIDs */ | |
606 | if (guest_cid <= VMADDR_CID_HOST || | |
607 | guest_cid == U32_MAX) | |
608 | return -EINVAL; | |
609 | ||
610 | /* 64-bit CIDs are not yet supported */ | |
611 | if (guest_cid > U32_MAX) | |
612 | return -EINVAL; | |
613 | ||
614 | /* Refuse if CID is already in use */ | |
433fc58e | 615 | spin_lock_bh(&vhost_vsock_lock); |
6c083c2b G |
616 | other = __vhost_vsock_get(guest_cid); |
617 | if (other && other != vsock) { | |
618 | spin_unlock_bh(&vhost_vsock_lock); | |
619 | return -EADDRINUSE; | |
620 | } | |
433fc58e AH |
621 | vsock->guest_cid = guest_cid; |
622 | spin_unlock_bh(&vhost_vsock_lock); | |
623 | ||
624 | return 0; | |
625 | } | |
626 | ||
627 | static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features) | |
628 | { | |
629 | struct vhost_virtqueue *vq; | |
630 | int i; | |
631 | ||
632 | if (features & ~VHOST_VSOCK_FEATURES) | |
633 | return -EOPNOTSUPP; | |
634 | ||
635 | mutex_lock(&vsock->dev.mutex); | |
636 | if ((features & (1 << VHOST_F_LOG_ALL)) && | |
637 | !vhost_log_access_ok(&vsock->dev)) { | |
638 | mutex_unlock(&vsock->dev.mutex); | |
639 | return -EFAULT; | |
640 | } | |
641 | ||
642 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
643 | vq = &vsock->vqs[i]; | |
644 | mutex_lock(&vq->mutex); | |
645 | vq->acked_features = features; | |
646 | mutex_unlock(&vq->mutex); | |
647 | } | |
648 | mutex_unlock(&vsock->dev.mutex); | |
649 | return 0; | |
650 | } | |
651 | ||
652 | static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl, | |
653 | unsigned long arg) | |
654 | { | |
655 | struct vhost_vsock *vsock = f->private_data; | |
656 | void __user *argp = (void __user *)arg; | |
657 | u64 guest_cid; | |
658 | u64 features; | |
659 | int start; | |
660 | int r; | |
661 | ||
662 | switch (ioctl) { | |
663 | case VHOST_VSOCK_SET_GUEST_CID: | |
664 | if (copy_from_user(&guest_cid, argp, sizeof(guest_cid))) | |
665 | return -EFAULT; | |
666 | return vhost_vsock_set_cid(vsock, guest_cid); | |
667 | case VHOST_VSOCK_SET_RUNNING: | |
668 | if (copy_from_user(&start, argp, sizeof(start))) | |
669 | return -EFAULT; | |
670 | if (start) | |
671 | return vhost_vsock_start(vsock); | |
672 | else | |
673 | return vhost_vsock_stop(vsock); | |
674 | case VHOST_GET_FEATURES: | |
675 | features = VHOST_VSOCK_FEATURES; | |
676 | if (copy_to_user(argp, &features, sizeof(features))) | |
677 | return -EFAULT; | |
678 | return 0; | |
679 | case VHOST_SET_FEATURES: | |
680 | if (copy_from_user(&features, argp, sizeof(features))) | |
681 | return -EFAULT; | |
682 | return vhost_vsock_set_features(vsock, features); | |
683 | default: | |
684 | mutex_lock(&vsock->dev.mutex); | |
685 | r = vhost_dev_ioctl(&vsock->dev, ioctl, argp); | |
686 | if (r == -ENOIOCTLCMD) | |
687 | r = vhost_vring_ioctl(&vsock->dev, ioctl, argp); | |
688 | else | |
689 | vhost_vsock_flush(vsock); | |
690 | mutex_unlock(&vsock->dev.mutex); | |
691 | return r; | |
692 | } | |
693 | } | |
694 | ||
695 | static const struct file_operations vhost_vsock_fops = { | |
696 | .owner = THIS_MODULE, | |
697 | .open = vhost_vsock_dev_open, | |
698 | .release = vhost_vsock_dev_release, | |
699 | .llseek = noop_llseek, | |
700 | .unlocked_ioctl = vhost_vsock_dev_ioctl, | |
701 | }; | |
702 | ||
703 | static struct miscdevice vhost_vsock_misc = { | |
704 | .minor = MISC_DYNAMIC_MINOR, | |
705 | .name = "vhost-vsock", | |
706 | .fops = &vhost_vsock_fops, | |
707 | }; | |
708 | ||
709 | static struct virtio_transport vhost_transport = { | |
710 | .transport = { | |
711 | .get_local_cid = vhost_transport_get_local_cid, | |
712 | ||
713 | .init = virtio_transport_do_socket_init, | |
714 | .destruct = virtio_transport_destruct, | |
715 | .release = virtio_transport_release, | |
716 | .connect = virtio_transport_connect, | |
717 | .shutdown = virtio_transport_shutdown, | |
16320f36 | 718 | .cancel_pkt = vhost_transport_cancel_pkt, |
433fc58e AH |
719 | |
720 | .dgram_enqueue = virtio_transport_dgram_enqueue, | |
721 | .dgram_dequeue = virtio_transport_dgram_dequeue, | |
722 | .dgram_bind = virtio_transport_dgram_bind, | |
723 | .dgram_allow = virtio_transport_dgram_allow, | |
724 | ||
725 | .stream_enqueue = virtio_transport_stream_enqueue, | |
726 | .stream_dequeue = virtio_transport_stream_dequeue, | |
727 | .stream_has_data = virtio_transport_stream_has_data, | |
728 | .stream_has_space = virtio_transport_stream_has_space, | |
729 | .stream_rcvhiwat = virtio_transport_stream_rcvhiwat, | |
730 | .stream_is_active = virtio_transport_stream_is_active, | |
731 | .stream_allow = virtio_transport_stream_allow, | |
732 | ||
733 | .notify_poll_in = virtio_transport_notify_poll_in, | |
734 | .notify_poll_out = virtio_transport_notify_poll_out, | |
735 | .notify_recv_init = virtio_transport_notify_recv_init, | |
736 | .notify_recv_pre_block = virtio_transport_notify_recv_pre_block, | |
737 | .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue, | |
738 | .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue, | |
739 | .notify_send_init = virtio_transport_notify_send_init, | |
740 | .notify_send_pre_block = virtio_transport_notify_send_pre_block, | |
741 | .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue, | |
742 | .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue, | |
743 | ||
744 | .set_buffer_size = virtio_transport_set_buffer_size, | |
745 | .set_min_buffer_size = virtio_transport_set_min_buffer_size, | |
746 | .set_max_buffer_size = virtio_transport_set_max_buffer_size, | |
747 | .get_buffer_size = virtio_transport_get_buffer_size, | |
748 | .get_min_buffer_size = virtio_transport_get_min_buffer_size, | |
749 | .get_max_buffer_size = virtio_transport_get_max_buffer_size, | |
750 | }, | |
751 | ||
752 | .send_pkt = vhost_transport_send_pkt, | |
753 | }; | |
754 | ||
755 | static int __init vhost_vsock_init(void) | |
756 | { | |
757 | int ret; | |
758 | ||
759 | ret = vsock_core_init(&vhost_transport.transport); | |
760 | if (ret < 0) | |
761 | return ret; | |
762 | return misc_register(&vhost_vsock_misc); | |
763 | }; | |
764 | ||
765 | static void __exit vhost_vsock_exit(void) | |
766 | { | |
767 | misc_deregister(&vhost_vsock_misc); | |
768 | vsock_core_exit(); | |
769 | }; | |
770 | ||
771 | module_init(vhost_vsock_init); | |
772 | module_exit(vhost_vsock_exit); | |
773 | MODULE_LICENSE("GPL v2"); | |
774 | MODULE_AUTHOR("Asias He"); | |
775 | MODULE_DESCRIPTION("vhost transport for vsock "); |