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> | |
834e772c | 18 | #include <linux/hashtable.h> |
433fc58e AH |
19 | |
20 | #include <net/af_vsock.h> | |
21 | #include "vhost.h" | |
22 | ||
23 | #define VHOST_VSOCK_DEFAULT_HOST_CID 2 | |
e82b9b07 JW |
24 | /* Max number of bytes transferred before requeueing the job. |
25 | * Using this limit prevents one virtqueue from starving others. */ | |
26 | #define VHOST_VSOCK_WEIGHT 0x80000 | |
27 | /* Max number of packets transferred before requeueing the job. | |
28 | * Using this limit prevents one virtqueue from starving others with | |
29 | * small pkts. | |
30 | */ | |
31 | #define VHOST_VSOCK_PKT_WEIGHT 256 | |
433fc58e AH |
32 | |
33 | enum { | |
34 | VHOST_VSOCK_FEATURES = VHOST_FEATURES, | |
35 | }; | |
36 | ||
37 | /* Used to track all the vhost_vsock instances on the system. */ | |
6db3d8dc | 38 | static DEFINE_MUTEX(vhost_vsock_mutex); |
834e772c | 39 | static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8); |
433fc58e AH |
40 | |
41 | struct vhost_vsock { | |
42 | struct vhost_dev dev; | |
43 | struct vhost_virtqueue vqs[2]; | |
44 | ||
6db3d8dc | 45 | /* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */ |
834e772c | 46 | struct hlist_node hash; |
433fc58e AH |
47 | |
48 | struct vhost_work send_pkt_work; | |
49 | spinlock_t send_pkt_list_lock; | |
50 | struct list_head send_pkt_list; /* host->guest pending packets */ | |
51 | ||
52 | atomic_t queued_replies; | |
53 | ||
54 | u32 guest_cid; | |
55 | }; | |
56 | ||
57 | static u32 vhost_transport_get_local_cid(void) | |
58 | { | |
59 | return VHOST_VSOCK_DEFAULT_HOST_CID; | |
60 | } | |
61 | ||
6db3d8dc | 62 | /* Callers that dereference the return value must hold vhost_vsock_mutex or the |
834e772c SH |
63 | * RCU read lock. |
64 | */ | |
65 | static struct vhost_vsock *vhost_vsock_get(u32 guest_cid) | |
433fc58e AH |
66 | { |
67 | struct vhost_vsock *vsock; | |
68 | ||
834e772c | 69 | hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) { |
433fc58e AH |
70 | u32 other_cid = vsock->guest_cid; |
71 | ||
72 | /* Skip instances that have no CID yet */ | |
73 | if (other_cid == 0) | |
74 | continue; | |
75 | ||
ff3c1b1a | 76 | if (other_cid == guest_cid) |
433fc58e | 77 | return vsock; |
ff3c1b1a | 78 | |
433fc58e | 79 | } |
433fc58e AH |
80 | |
81 | return NULL; | |
82 | } | |
83 | ||
84 | static void | |
85 | vhost_transport_do_send_pkt(struct vhost_vsock *vsock, | |
86 | struct vhost_virtqueue *vq) | |
87 | { | |
88 | struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX]; | |
89 | bool added = false; | |
90 | bool restart_tx = false; | |
91 | ||
92 | mutex_lock(&vq->mutex); | |
93 | ||
94 | if (!vq->private_data) | |
95 | goto out; | |
96 | ||
97 | /* Avoid further vmexits, we're already processing the virtqueue */ | |
98 | vhost_disable_notify(&vsock->dev, vq); | |
99 | ||
100 | for (;;) { | |
101 | struct virtio_vsock_pkt *pkt; | |
102 | struct iov_iter iov_iter; | |
103 | unsigned out, in; | |
104 | size_t nbytes; | |
105 | size_t len; | |
106 | int head; | |
107 | ||
108 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
109 | if (list_empty(&vsock->send_pkt_list)) { | |
110 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
111 | vhost_enable_notify(&vsock->dev, vq); | |
112 | break; | |
113 | } | |
114 | ||
115 | pkt = list_first_entry(&vsock->send_pkt_list, | |
116 | struct virtio_vsock_pkt, list); | |
117 | list_del_init(&pkt->list); | |
118 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
119 | ||
120 | head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), | |
121 | &out, &in, NULL, NULL); | |
122 | if (head < 0) { | |
123 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
124 | list_add(&pkt->list, &vsock->send_pkt_list); | |
125 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
126 | break; | |
127 | } | |
128 | ||
129 | if (head == vq->num) { | |
130 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
131 | list_add(&pkt->list, &vsock->send_pkt_list); | |
132 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
133 | ||
134 | /* We cannot finish yet if more buffers snuck in while | |
135 | * re-enabling notify. | |
136 | */ | |
137 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { | |
138 | vhost_disable_notify(&vsock->dev, vq); | |
139 | continue; | |
140 | } | |
141 | break; | |
142 | } | |
143 | ||
144 | if (out) { | |
145 | virtio_transport_free_pkt(pkt); | |
146 | vq_err(vq, "Expected 0 output buffers, got %u\n", out); | |
147 | break; | |
148 | } | |
149 | ||
150 | len = iov_length(&vq->iov[out], in); | |
151 | iov_iter_init(&iov_iter, READ, &vq->iov[out], in, len); | |
152 | ||
153 | nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); | |
154 | if (nbytes != sizeof(pkt->hdr)) { | |
155 | virtio_transport_free_pkt(pkt); | |
156 | vq_err(vq, "Faulted on copying pkt hdr\n"); | |
157 | break; | |
158 | } | |
159 | ||
160 | nbytes = copy_to_iter(pkt->buf, pkt->len, &iov_iter); | |
161 | if (nbytes != pkt->len) { | |
162 | virtio_transport_free_pkt(pkt); | |
163 | vq_err(vq, "Faulted on copying pkt buf\n"); | |
164 | break; | |
165 | } | |
166 | ||
167 | vhost_add_used(vq, head, sizeof(pkt->hdr) + pkt->len); | |
168 | added = true; | |
169 | ||
170 | if (pkt->reply) { | |
171 | int val; | |
172 | ||
173 | val = atomic_dec_return(&vsock->queued_replies); | |
174 | ||
175 | /* Do we have resources to resume tx processing? */ | |
176 | if (val + 1 == tx_vq->num) | |
177 | restart_tx = true; | |
178 | } | |
179 | ||
82dfb540 GG |
180 | /* Deliver to monitoring devices all correctly transmitted |
181 | * packets. | |
182 | */ | |
183 | virtio_transport_deliver_tap_pkt(pkt); | |
184 | ||
433fc58e AH |
185 | virtio_transport_free_pkt(pkt); |
186 | } | |
187 | if (added) | |
188 | vhost_signal(&vsock->dev, vq); | |
189 | ||
190 | out: | |
191 | mutex_unlock(&vq->mutex); | |
192 | ||
193 | if (restart_tx) | |
194 | vhost_poll_queue(&tx_vq->poll); | |
195 | } | |
196 | ||
197 | static void vhost_transport_send_pkt_work(struct vhost_work *work) | |
198 | { | |
199 | struct vhost_virtqueue *vq; | |
200 | struct vhost_vsock *vsock; | |
201 | ||
202 | vsock = container_of(work, struct vhost_vsock, send_pkt_work); | |
203 | vq = &vsock->vqs[VSOCK_VQ_RX]; | |
204 | ||
205 | vhost_transport_do_send_pkt(vsock, vq); | |
206 | } | |
207 | ||
208 | static int | |
209 | vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt) | |
210 | { | |
211 | struct vhost_vsock *vsock; | |
433fc58e AH |
212 | int len = pkt->len; |
213 | ||
834e772c SH |
214 | rcu_read_lock(); |
215 | ||
433fc58e AH |
216 | /* Find the vhost_vsock according to guest context id */ |
217 | vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid)); | |
218 | if (!vsock) { | |
834e772c | 219 | rcu_read_unlock(); |
433fc58e AH |
220 | virtio_transport_free_pkt(pkt); |
221 | return -ENODEV; | |
222 | } | |
223 | ||
433fc58e AH |
224 | if (pkt->reply) |
225 | atomic_inc(&vsock->queued_replies); | |
226 | ||
227 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
228 | list_add_tail(&pkt->list, &vsock->send_pkt_list); | |
229 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
230 | ||
231 | vhost_work_queue(&vsock->dev, &vsock->send_pkt_work); | |
834e772c SH |
232 | |
233 | rcu_read_unlock(); | |
433fc58e AH |
234 | return len; |
235 | } | |
236 | ||
16320f36 PT |
237 | static int |
238 | vhost_transport_cancel_pkt(struct vsock_sock *vsk) | |
239 | { | |
240 | struct vhost_vsock *vsock; | |
241 | struct virtio_vsock_pkt *pkt, *n; | |
242 | int cnt = 0; | |
834e772c | 243 | int ret = -ENODEV; |
16320f36 PT |
244 | LIST_HEAD(freeme); |
245 | ||
834e772c SH |
246 | rcu_read_lock(); |
247 | ||
16320f36 PT |
248 | /* Find the vhost_vsock according to guest context id */ |
249 | vsock = vhost_vsock_get(vsk->remote_addr.svm_cid); | |
250 | if (!vsock) | |
834e772c | 251 | goto out; |
16320f36 PT |
252 | |
253 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
254 | list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) { | |
255 | if (pkt->vsk != vsk) | |
256 | continue; | |
257 | list_move(&pkt->list, &freeme); | |
258 | } | |
259 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
260 | ||
261 | list_for_each_entry_safe(pkt, n, &freeme, list) { | |
262 | if (pkt->reply) | |
263 | cnt++; | |
264 | list_del(&pkt->list); | |
265 | virtio_transport_free_pkt(pkt); | |
266 | } | |
267 | ||
268 | if (cnt) { | |
269 | struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX]; | |
270 | int new_cnt; | |
271 | ||
272 | new_cnt = atomic_sub_return(cnt, &vsock->queued_replies); | |
273 | if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num) | |
274 | vhost_poll_queue(&tx_vq->poll); | |
275 | } | |
276 | ||
834e772c SH |
277 | ret = 0; |
278 | out: | |
279 | rcu_read_unlock(); | |
280 | return ret; | |
16320f36 PT |
281 | } |
282 | ||
433fc58e AH |
283 | static struct virtio_vsock_pkt * |
284 | vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq, | |
285 | unsigned int out, unsigned int in) | |
286 | { | |
287 | struct virtio_vsock_pkt *pkt; | |
288 | struct iov_iter iov_iter; | |
289 | size_t nbytes; | |
290 | size_t len; | |
291 | ||
292 | if (in != 0) { | |
293 | vq_err(vq, "Expected 0 input buffers, got %u\n", in); | |
294 | return NULL; | |
295 | } | |
296 | ||
297 | pkt = kzalloc(sizeof(*pkt), GFP_KERNEL); | |
298 | if (!pkt) | |
299 | return NULL; | |
300 | ||
301 | len = iov_length(vq->iov, out); | |
302 | iov_iter_init(&iov_iter, WRITE, vq->iov, out, len); | |
303 | ||
304 | nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); | |
305 | if (nbytes != sizeof(pkt->hdr)) { | |
306 | vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n", | |
307 | sizeof(pkt->hdr), nbytes); | |
308 | kfree(pkt); | |
309 | return NULL; | |
310 | } | |
311 | ||
312 | if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM) | |
313 | pkt->len = le32_to_cpu(pkt->hdr.len); | |
314 | ||
315 | /* No payload */ | |
316 | if (!pkt->len) | |
317 | return pkt; | |
318 | ||
319 | /* The pkt is too big */ | |
320 | if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) { | |
321 | kfree(pkt); | |
322 | return NULL; | |
323 | } | |
324 | ||
325 | pkt->buf = kmalloc(pkt->len, GFP_KERNEL); | |
326 | if (!pkt->buf) { | |
327 | kfree(pkt); | |
328 | return NULL; | |
329 | } | |
330 | ||
331 | nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter); | |
332 | if (nbytes != pkt->len) { | |
333 | vq_err(vq, "Expected %u byte payload, got %zu bytes\n", | |
334 | pkt->len, nbytes); | |
335 | virtio_transport_free_pkt(pkt); | |
336 | return NULL; | |
337 | } | |
338 | ||
339 | return pkt; | |
340 | } | |
341 | ||
342 | /* Is there space left for replies to rx packets? */ | |
343 | static bool vhost_vsock_more_replies(struct vhost_vsock *vsock) | |
344 | { | |
345 | struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX]; | |
346 | int val; | |
347 | ||
348 | smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */ | |
349 | val = atomic_read(&vsock->queued_replies); | |
350 | ||
351 | return val < vq->num; | |
352 | } | |
353 | ||
354 | static void vhost_vsock_handle_tx_kick(struct vhost_work *work) | |
355 | { | |
356 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
357 | poll.work); | |
358 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
359 | dev); | |
360 | struct virtio_vsock_pkt *pkt; | |
361 | int head; | |
362 | unsigned int out, in; | |
363 | bool added = false; | |
364 | ||
365 | mutex_lock(&vq->mutex); | |
366 | ||
367 | if (!vq->private_data) | |
368 | goto out; | |
369 | ||
370 | vhost_disable_notify(&vsock->dev, vq); | |
371 | for (;;) { | |
3fda5d6e SH |
372 | u32 len; |
373 | ||
433fc58e AH |
374 | if (!vhost_vsock_more_replies(vsock)) { |
375 | /* Stop tx until the device processes already | |
376 | * pending replies. Leave tx virtqueue | |
377 | * callbacks disabled. | |
378 | */ | |
379 | goto no_more_replies; | |
380 | } | |
381 | ||
382 | head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), | |
383 | &out, &in, NULL, NULL); | |
384 | if (head < 0) | |
385 | break; | |
386 | ||
387 | if (head == vq->num) { | |
388 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { | |
389 | vhost_disable_notify(&vsock->dev, vq); | |
390 | continue; | |
391 | } | |
392 | break; | |
393 | } | |
394 | ||
395 | pkt = vhost_vsock_alloc_pkt(vq, out, in); | |
396 | if (!pkt) { | |
397 | vq_err(vq, "Faulted on pkt\n"); | |
398 | continue; | |
399 | } | |
400 | ||
3fda5d6e SH |
401 | len = pkt->len; |
402 | ||
82dfb540 GG |
403 | /* Deliver to monitoring devices all received packets */ |
404 | virtio_transport_deliver_tap_pkt(pkt); | |
405 | ||
433fc58e AH |
406 | /* Only accept correctly addressed packets */ |
407 | if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid) | |
408 | virtio_transport_recv_pkt(pkt); | |
409 | else | |
410 | virtio_transport_free_pkt(pkt); | |
411 | ||
3fda5d6e | 412 | vhost_add_used(vq, head, sizeof(pkt->hdr) + len); |
433fc58e AH |
413 | added = true; |
414 | } | |
415 | ||
416 | no_more_replies: | |
417 | if (added) | |
418 | vhost_signal(&vsock->dev, vq); | |
419 | ||
420 | out: | |
421 | mutex_unlock(&vq->mutex); | |
422 | } | |
423 | ||
424 | static void vhost_vsock_handle_rx_kick(struct vhost_work *work) | |
425 | { | |
426 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
427 | poll.work); | |
428 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
429 | dev); | |
430 | ||
431 | vhost_transport_do_send_pkt(vsock, vq); | |
432 | } | |
433 | ||
434 | static int vhost_vsock_start(struct vhost_vsock *vsock) | |
435 | { | |
0516ffd8 | 436 | struct vhost_virtqueue *vq; |
433fc58e AH |
437 | size_t i; |
438 | int ret; | |
439 | ||
440 | mutex_lock(&vsock->dev.mutex); | |
441 | ||
442 | ret = vhost_dev_check_owner(&vsock->dev); | |
443 | if (ret) | |
444 | goto err; | |
445 | ||
446 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
0516ffd8 | 447 | vq = &vsock->vqs[i]; |
433fc58e AH |
448 | |
449 | mutex_lock(&vq->mutex); | |
450 | ||
451 | if (!vhost_vq_access_ok(vq)) { | |
452 | ret = -EFAULT; | |
433fc58e AH |
453 | goto err_vq; |
454 | } | |
455 | ||
456 | if (!vq->private_data) { | |
457 | vq->private_data = vsock; | |
0516ffd8 SH |
458 | ret = vhost_vq_init_access(vq); |
459 | if (ret) | |
460 | goto err_vq; | |
433fc58e AH |
461 | } |
462 | ||
463 | mutex_unlock(&vq->mutex); | |
464 | } | |
465 | ||
466 | mutex_unlock(&vsock->dev.mutex); | |
467 | return 0; | |
468 | ||
469 | err_vq: | |
0516ffd8 SH |
470 | vq->private_data = NULL; |
471 | mutex_unlock(&vq->mutex); | |
472 | ||
433fc58e | 473 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
0516ffd8 | 474 | vq = &vsock->vqs[i]; |
433fc58e AH |
475 | |
476 | mutex_lock(&vq->mutex); | |
477 | vq->private_data = NULL; | |
478 | mutex_unlock(&vq->mutex); | |
479 | } | |
480 | err: | |
481 | mutex_unlock(&vsock->dev.mutex); | |
482 | return ret; | |
483 | } | |
484 | ||
485 | static int vhost_vsock_stop(struct vhost_vsock *vsock) | |
486 | { | |
487 | size_t i; | |
488 | int ret; | |
489 | ||
490 | mutex_lock(&vsock->dev.mutex); | |
491 | ||
492 | ret = vhost_dev_check_owner(&vsock->dev); | |
493 | if (ret) | |
494 | goto err; | |
495 | ||
496 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
497 | struct vhost_virtqueue *vq = &vsock->vqs[i]; | |
498 | ||
499 | mutex_lock(&vq->mutex); | |
500 | vq->private_data = NULL; | |
501 | mutex_unlock(&vq->mutex); | |
502 | } | |
503 | ||
504 | err: | |
505 | mutex_unlock(&vsock->dev.mutex); | |
506 | return ret; | |
507 | } | |
508 | ||
509 | static void vhost_vsock_free(struct vhost_vsock *vsock) | |
510 | { | |
b226acab | 511 | kvfree(vsock); |
433fc58e AH |
512 | } |
513 | ||
514 | static int vhost_vsock_dev_open(struct inode *inode, struct file *file) | |
515 | { | |
516 | struct vhost_virtqueue **vqs; | |
517 | struct vhost_vsock *vsock; | |
518 | int ret; | |
519 | ||
520 | /* This struct is large and allocation could fail, fall back to vmalloc | |
521 | * if there is no other way. | |
522 | */ | |
dcda9b04 | 523 | vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
6c5ab651 MH |
524 | if (!vsock) |
525 | return -ENOMEM; | |
433fc58e AH |
526 | |
527 | vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL); | |
528 | if (!vqs) { | |
529 | ret = -ENOMEM; | |
530 | goto out; | |
531 | } | |
532 | ||
a72b69dc SH |
533 | vsock->guest_cid = 0; /* no CID assigned yet */ |
534 | ||
433fc58e AH |
535 | atomic_set(&vsock->queued_replies, 0); |
536 | ||
537 | vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX]; | |
538 | vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX]; | |
539 | vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick; | |
540 | vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick; | |
541 | ||
e82b9b07 JW |
542 | vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs), |
543 | UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT, | |
544 | VHOST_VSOCK_WEIGHT); | |
433fc58e AH |
545 | |
546 | file->private_data = vsock; | |
547 | spin_lock_init(&vsock->send_pkt_list_lock); | |
548 | INIT_LIST_HEAD(&vsock->send_pkt_list); | |
549 | vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work); | |
433fc58e AH |
550 | return 0; |
551 | ||
552 | out: | |
553 | vhost_vsock_free(vsock); | |
554 | return ret; | |
555 | } | |
556 | ||
557 | static void vhost_vsock_flush(struct vhost_vsock *vsock) | |
558 | { | |
559 | int i; | |
560 | ||
561 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) | |
562 | if (vsock->vqs[i].handle_kick) | |
563 | vhost_poll_flush(&vsock->vqs[i].poll); | |
564 | vhost_work_flush(&vsock->dev, &vsock->send_pkt_work); | |
565 | } | |
566 | ||
567 | static void vhost_vsock_reset_orphans(struct sock *sk) | |
568 | { | |
569 | struct vsock_sock *vsk = vsock_sk(sk); | |
570 | ||
571 | /* vmci_transport.c doesn't take sk_lock here either. At least we're | |
572 | * under vsock_table_lock so the sock cannot disappear while we're | |
573 | * executing. | |
574 | */ | |
575 | ||
c38f57da SH |
576 | /* If the peer is still valid, no need to reset connection */ |
577 | if (vhost_vsock_get(vsk->remote_addr.svm_cid)) | |
578 | return; | |
579 | ||
580 | /* If the close timeout is pending, let it expire. This avoids races | |
581 | * with the timeout callback. | |
582 | */ | |
583 | if (vsk->close_work_scheduled) | |
584 | return; | |
585 | ||
586 | sock_set_flag(sk, SOCK_DONE); | |
587 | vsk->peer_shutdown = SHUTDOWN_MASK; | |
588 | sk->sk_state = SS_UNCONNECTED; | |
589 | sk->sk_err = ECONNRESET; | |
590 | sk->sk_error_report(sk); | |
433fc58e AH |
591 | } |
592 | ||
593 | static int vhost_vsock_dev_release(struct inode *inode, struct file *file) | |
594 | { | |
595 | struct vhost_vsock *vsock = file->private_data; | |
596 | ||
6db3d8dc | 597 | mutex_lock(&vhost_vsock_mutex); |
834e772c SH |
598 | if (vsock->guest_cid) |
599 | hash_del_rcu(&vsock->hash); | |
6db3d8dc | 600 | mutex_unlock(&vhost_vsock_mutex); |
433fc58e | 601 | |
834e772c SH |
602 | /* Wait for other CPUs to finish using vsock */ |
603 | synchronize_rcu(); | |
604 | ||
433fc58e AH |
605 | /* Iterating over all connections for all CIDs to find orphans is |
606 | * inefficient. Room for improvement here. */ | |
607 | vsock_for_each_connected_socket(vhost_vsock_reset_orphans); | |
608 | ||
609 | vhost_vsock_stop(vsock); | |
610 | vhost_vsock_flush(vsock); | |
611 | vhost_dev_stop(&vsock->dev); | |
612 | ||
613 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
614 | while (!list_empty(&vsock->send_pkt_list)) { | |
615 | struct virtio_vsock_pkt *pkt; | |
616 | ||
617 | pkt = list_first_entry(&vsock->send_pkt_list, | |
618 | struct virtio_vsock_pkt, list); | |
619 | list_del_init(&pkt->list); | |
620 | virtio_transport_free_pkt(pkt); | |
621 | } | |
622 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
623 | ||
f6f93f75 | 624 | vhost_dev_cleanup(&vsock->dev); |
433fc58e AH |
625 | kfree(vsock->dev.vqs); |
626 | vhost_vsock_free(vsock); | |
627 | return 0; | |
628 | } | |
629 | ||
630 | static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid) | |
631 | { | |
632 | struct vhost_vsock *other; | |
633 | ||
634 | /* Refuse reserved CIDs */ | |
635 | if (guest_cid <= VMADDR_CID_HOST || | |
636 | guest_cid == U32_MAX) | |
637 | return -EINVAL; | |
638 | ||
639 | /* 64-bit CIDs are not yet supported */ | |
640 | if (guest_cid > U32_MAX) | |
641 | return -EINVAL; | |
642 | ||
643 | /* Refuse if CID is already in use */ | |
6db3d8dc | 644 | mutex_lock(&vhost_vsock_mutex); |
834e772c | 645 | other = vhost_vsock_get(guest_cid); |
6c083c2b | 646 | if (other && other != vsock) { |
6db3d8dc | 647 | mutex_unlock(&vhost_vsock_mutex); |
6c083c2b G |
648 | return -EADDRINUSE; |
649 | } | |
834e772c SH |
650 | |
651 | if (vsock->guest_cid) | |
652 | hash_del_rcu(&vsock->hash); | |
653 | ||
433fc58e | 654 | vsock->guest_cid = guest_cid; |
7fbe078c | 655 | hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid); |
6db3d8dc | 656 | mutex_unlock(&vhost_vsock_mutex); |
433fc58e AH |
657 | |
658 | return 0; | |
659 | } | |
660 | ||
661 | static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features) | |
662 | { | |
663 | struct vhost_virtqueue *vq; | |
664 | int i; | |
665 | ||
666 | if (features & ~VHOST_VSOCK_FEATURES) | |
667 | return -EOPNOTSUPP; | |
668 | ||
669 | mutex_lock(&vsock->dev.mutex); | |
670 | if ((features & (1 << VHOST_F_LOG_ALL)) && | |
671 | !vhost_log_access_ok(&vsock->dev)) { | |
672 | mutex_unlock(&vsock->dev.mutex); | |
673 | return -EFAULT; | |
674 | } | |
675 | ||
676 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
677 | vq = &vsock->vqs[i]; | |
678 | mutex_lock(&vq->mutex); | |
679 | vq->acked_features = features; | |
680 | mutex_unlock(&vq->mutex); | |
681 | } | |
682 | mutex_unlock(&vsock->dev.mutex); | |
683 | return 0; | |
684 | } | |
685 | ||
686 | static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl, | |
687 | unsigned long arg) | |
688 | { | |
689 | struct vhost_vsock *vsock = f->private_data; | |
690 | void __user *argp = (void __user *)arg; | |
691 | u64 guest_cid; | |
692 | u64 features; | |
693 | int start; | |
694 | int r; | |
695 | ||
696 | switch (ioctl) { | |
697 | case VHOST_VSOCK_SET_GUEST_CID: | |
698 | if (copy_from_user(&guest_cid, argp, sizeof(guest_cid))) | |
699 | return -EFAULT; | |
700 | return vhost_vsock_set_cid(vsock, guest_cid); | |
701 | case VHOST_VSOCK_SET_RUNNING: | |
702 | if (copy_from_user(&start, argp, sizeof(start))) | |
703 | return -EFAULT; | |
704 | if (start) | |
705 | return vhost_vsock_start(vsock); | |
706 | else | |
707 | return vhost_vsock_stop(vsock); | |
708 | case VHOST_GET_FEATURES: | |
709 | features = VHOST_VSOCK_FEATURES; | |
710 | if (copy_to_user(argp, &features, sizeof(features))) | |
711 | return -EFAULT; | |
712 | return 0; | |
713 | case VHOST_SET_FEATURES: | |
714 | if (copy_from_user(&features, argp, sizeof(features))) | |
715 | return -EFAULT; | |
716 | return vhost_vsock_set_features(vsock, features); | |
717 | default: | |
718 | mutex_lock(&vsock->dev.mutex); | |
719 | r = vhost_dev_ioctl(&vsock->dev, ioctl, argp); | |
720 | if (r == -ENOIOCTLCMD) | |
721 | r = vhost_vring_ioctl(&vsock->dev, ioctl, argp); | |
722 | else | |
723 | vhost_vsock_flush(vsock); | |
724 | mutex_unlock(&vsock->dev.mutex); | |
725 | return r; | |
726 | } | |
727 | } | |
728 | ||
dc32bb67 SR |
729 | #ifdef CONFIG_COMPAT |
730 | static long vhost_vsock_dev_compat_ioctl(struct file *f, unsigned int ioctl, | |
731 | unsigned long arg) | |
732 | { | |
733 | return vhost_vsock_dev_ioctl(f, ioctl, (unsigned long)compat_ptr(arg)); | |
734 | } | |
735 | #endif | |
736 | ||
433fc58e AH |
737 | static const struct file_operations vhost_vsock_fops = { |
738 | .owner = THIS_MODULE, | |
739 | .open = vhost_vsock_dev_open, | |
740 | .release = vhost_vsock_dev_release, | |
741 | .llseek = noop_llseek, | |
742 | .unlocked_ioctl = vhost_vsock_dev_ioctl, | |
dc32bb67 SR |
743 | #ifdef CONFIG_COMPAT |
744 | .compat_ioctl = vhost_vsock_dev_compat_ioctl, | |
745 | #endif | |
433fc58e AH |
746 | }; |
747 | ||
748 | static struct miscdevice vhost_vsock_misc = { | |
f4660cc9 | 749 | .minor = VHOST_VSOCK_MINOR, |
433fc58e AH |
750 | .name = "vhost-vsock", |
751 | .fops = &vhost_vsock_fops, | |
752 | }; | |
753 | ||
754 | static struct virtio_transport vhost_transport = { | |
755 | .transport = { | |
756 | .get_local_cid = vhost_transport_get_local_cid, | |
757 | ||
758 | .init = virtio_transport_do_socket_init, | |
759 | .destruct = virtio_transport_destruct, | |
760 | .release = virtio_transport_release, | |
761 | .connect = virtio_transport_connect, | |
762 | .shutdown = virtio_transport_shutdown, | |
16320f36 | 763 | .cancel_pkt = vhost_transport_cancel_pkt, |
433fc58e AH |
764 | |
765 | .dgram_enqueue = virtio_transport_dgram_enqueue, | |
766 | .dgram_dequeue = virtio_transport_dgram_dequeue, | |
767 | .dgram_bind = virtio_transport_dgram_bind, | |
768 | .dgram_allow = virtio_transport_dgram_allow, | |
769 | ||
770 | .stream_enqueue = virtio_transport_stream_enqueue, | |
771 | .stream_dequeue = virtio_transport_stream_dequeue, | |
772 | .stream_has_data = virtio_transport_stream_has_data, | |
773 | .stream_has_space = virtio_transport_stream_has_space, | |
774 | .stream_rcvhiwat = virtio_transport_stream_rcvhiwat, | |
775 | .stream_is_active = virtio_transport_stream_is_active, | |
776 | .stream_allow = virtio_transport_stream_allow, | |
777 | ||
778 | .notify_poll_in = virtio_transport_notify_poll_in, | |
779 | .notify_poll_out = virtio_transport_notify_poll_out, | |
780 | .notify_recv_init = virtio_transport_notify_recv_init, | |
781 | .notify_recv_pre_block = virtio_transport_notify_recv_pre_block, | |
782 | .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue, | |
783 | .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue, | |
784 | .notify_send_init = virtio_transport_notify_send_init, | |
785 | .notify_send_pre_block = virtio_transport_notify_send_pre_block, | |
786 | .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue, | |
787 | .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue, | |
788 | ||
789 | .set_buffer_size = virtio_transport_set_buffer_size, | |
790 | .set_min_buffer_size = virtio_transport_set_min_buffer_size, | |
791 | .set_max_buffer_size = virtio_transport_set_max_buffer_size, | |
792 | .get_buffer_size = virtio_transport_get_buffer_size, | |
793 | .get_min_buffer_size = virtio_transport_get_min_buffer_size, | |
794 | .get_max_buffer_size = virtio_transport_get_max_buffer_size, | |
795 | }, | |
796 | ||
797 | .send_pkt = vhost_transport_send_pkt, | |
798 | }; | |
799 | ||
800 | static int __init vhost_vsock_init(void) | |
801 | { | |
802 | int ret; | |
803 | ||
804 | ret = vsock_core_init(&vhost_transport.transport); | |
805 | if (ret < 0) | |
806 | return ret; | |
807 | return misc_register(&vhost_vsock_misc); | |
808 | }; | |
809 | ||
810 | static void __exit vhost_vsock_exit(void) | |
811 | { | |
812 | misc_deregister(&vhost_vsock_misc); | |
813 | vsock_core_exit(); | |
814 | }; | |
815 | ||
816 | module_init(vhost_vsock_init); | |
817 | module_exit(vhost_vsock_exit); | |
818 | MODULE_LICENSE("GPL v2"); | |
819 | MODULE_AUTHOR("Asias He"); | |
820 | MODULE_DESCRIPTION("vhost transport for vsock "); | |
f4660cc9 SH |
821 | MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR); |
822 | MODULE_ALIAS("devname:vhost-vsock"); |