]>
Commit | Line | Data |
---|---|---|
7a338472 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
433fc58e AH |
2 | /* |
3 | * vhost transport for vsock | |
4 | * | |
5 | * Copyright (C) 2013-2015 Red Hat, Inc. | |
6 | * Author: Asias He <asias@redhat.com> | |
7 | * Stefan Hajnoczi <stefanha@redhat.com> | |
433fc58e AH |
8 | */ |
9 | #include <linux/miscdevice.h> | |
10 | #include <linux/atomic.h> | |
11 | #include <linux/module.h> | |
12 | #include <linux/mutex.h> | |
13 | #include <linux/vmalloc.h> | |
14 | #include <net/sock.h> | |
15 | #include <linux/virtio_vsock.h> | |
16 | #include <linux/vhost.h> | |
834e772c | 17 | #include <linux/hashtable.h> |
433fc58e AH |
18 | |
19 | #include <net/af_vsock.h> | |
20 | #include "vhost.h" | |
21 | ||
22 | #define VHOST_VSOCK_DEFAULT_HOST_CID 2 | |
e82b9b07 JW |
23 | /* Max number of bytes transferred before requeueing the job. |
24 | * Using this limit prevents one virtqueue from starving others. */ | |
25 | #define VHOST_VSOCK_WEIGHT 0x80000 | |
26 | /* Max number of packets transferred before requeueing the job. | |
27 | * Using this limit prevents one virtqueue from starving others with | |
28 | * small pkts. | |
29 | */ | |
30 | #define VHOST_VSOCK_PKT_WEIGHT 256 | |
433fc58e AH |
31 | |
32 | enum { | |
33 | VHOST_VSOCK_FEATURES = VHOST_FEATURES, | |
34 | }; | |
35 | ||
36 | /* Used to track all the vhost_vsock instances on the system. */ | |
6db3d8dc | 37 | static DEFINE_MUTEX(vhost_vsock_mutex); |
834e772c | 38 | static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8); |
433fc58e AH |
39 | |
40 | struct vhost_vsock { | |
41 | struct vhost_dev dev; | |
42 | struct vhost_virtqueue vqs[2]; | |
43 | ||
6db3d8dc | 44 | /* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */ |
834e772c | 45 | struct hlist_node hash; |
433fc58e AH |
46 | |
47 | struct vhost_work send_pkt_work; | |
48 | spinlock_t send_pkt_list_lock; | |
49 | struct list_head send_pkt_list; /* host->guest pending packets */ | |
50 | ||
51 | atomic_t queued_replies; | |
52 | ||
53 | u32 guest_cid; | |
54 | }; | |
55 | ||
56 | static u32 vhost_transport_get_local_cid(void) | |
57 | { | |
58 | return VHOST_VSOCK_DEFAULT_HOST_CID; | |
59 | } | |
60 | ||
6db3d8dc | 61 | /* Callers that dereference the return value must hold vhost_vsock_mutex or the |
834e772c SH |
62 | * RCU read lock. |
63 | */ | |
64 | static struct vhost_vsock *vhost_vsock_get(u32 guest_cid) | |
433fc58e AH |
65 | { |
66 | struct vhost_vsock *vsock; | |
67 | ||
834e772c | 68 | hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) { |
433fc58e AH |
69 | u32 other_cid = vsock->guest_cid; |
70 | ||
71 | /* Skip instances that have no CID yet */ | |
72 | if (other_cid == 0) | |
73 | continue; | |
74 | ||
ff3c1b1a | 75 | if (other_cid == guest_cid) |
433fc58e | 76 | return vsock; |
ff3c1b1a | 77 | |
433fc58e | 78 | } |
433fc58e AH |
79 | |
80 | return NULL; | |
81 | } | |
82 | ||
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]; | |
e79b431f | 88 | int pkts = 0, total_len = 0; |
433fc58e AH |
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 | ||
e79b431f | 100 | do { |
433fc58e AH |
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 | ||
e79b431f | 185 | total_len += pkt->len; |
433fc58e | 186 | virtio_transport_free_pkt(pkt); |
e79b431f | 187 | } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); |
433fc58e AH |
188 | if (added) |
189 | vhost_signal(&vsock->dev, vq); | |
190 | ||
191 | out: | |
192 | mutex_unlock(&vq->mutex); | |
193 | ||
194 | if (restart_tx) | |
195 | vhost_poll_queue(&tx_vq->poll); | |
196 | } | |
197 | ||
198 | static void vhost_transport_send_pkt_work(struct vhost_work *work) | |
199 | { | |
200 | struct vhost_virtqueue *vq; | |
201 | struct vhost_vsock *vsock; | |
202 | ||
203 | vsock = container_of(work, struct vhost_vsock, send_pkt_work); | |
204 | vq = &vsock->vqs[VSOCK_VQ_RX]; | |
205 | ||
206 | vhost_transport_do_send_pkt(vsock, vq); | |
207 | } | |
208 | ||
209 | static int | |
210 | vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt) | |
211 | { | |
212 | struct vhost_vsock *vsock; | |
433fc58e AH |
213 | int len = pkt->len; |
214 | ||
834e772c SH |
215 | rcu_read_lock(); |
216 | ||
433fc58e AH |
217 | /* Find the vhost_vsock according to guest context id */ |
218 | vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid)); | |
219 | if (!vsock) { | |
834e772c | 220 | rcu_read_unlock(); |
433fc58e AH |
221 | virtio_transport_free_pkt(pkt); |
222 | return -ENODEV; | |
223 | } | |
224 | ||
433fc58e AH |
225 | if (pkt->reply) |
226 | atomic_inc(&vsock->queued_replies); | |
227 | ||
228 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
229 | list_add_tail(&pkt->list, &vsock->send_pkt_list); | |
230 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
231 | ||
232 | vhost_work_queue(&vsock->dev, &vsock->send_pkt_work); | |
834e772c SH |
233 | |
234 | rcu_read_unlock(); | |
433fc58e AH |
235 | return len; |
236 | } | |
237 | ||
16320f36 PT |
238 | static int |
239 | vhost_transport_cancel_pkt(struct vsock_sock *vsk) | |
240 | { | |
241 | struct vhost_vsock *vsock; | |
242 | struct virtio_vsock_pkt *pkt, *n; | |
243 | int cnt = 0; | |
834e772c | 244 | int ret = -ENODEV; |
16320f36 PT |
245 | LIST_HEAD(freeme); |
246 | ||
834e772c SH |
247 | rcu_read_lock(); |
248 | ||
16320f36 PT |
249 | /* Find the vhost_vsock according to guest context id */ |
250 | vsock = vhost_vsock_get(vsk->remote_addr.svm_cid); | |
251 | if (!vsock) | |
834e772c | 252 | goto out; |
16320f36 PT |
253 | |
254 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
255 | list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) { | |
256 | if (pkt->vsk != vsk) | |
257 | continue; | |
258 | list_move(&pkt->list, &freeme); | |
259 | } | |
260 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
261 | ||
262 | list_for_each_entry_safe(pkt, n, &freeme, list) { | |
263 | if (pkt->reply) | |
264 | cnt++; | |
265 | list_del(&pkt->list); | |
266 | virtio_transport_free_pkt(pkt); | |
267 | } | |
268 | ||
269 | if (cnt) { | |
270 | struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX]; | |
271 | int new_cnt; | |
272 | ||
273 | new_cnt = atomic_sub_return(cnt, &vsock->queued_replies); | |
274 | if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num) | |
275 | vhost_poll_queue(&tx_vq->poll); | |
276 | } | |
277 | ||
834e772c SH |
278 | ret = 0; |
279 | out: | |
280 | rcu_read_unlock(); | |
281 | return ret; | |
16320f36 PT |
282 | } |
283 | ||
433fc58e AH |
284 | static struct virtio_vsock_pkt * |
285 | vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq, | |
286 | unsigned int out, unsigned int in) | |
287 | { | |
288 | struct virtio_vsock_pkt *pkt; | |
289 | struct iov_iter iov_iter; | |
290 | size_t nbytes; | |
291 | size_t len; | |
292 | ||
293 | if (in != 0) { | |
294 | vq_err(vq, "Expected 0 input buffers, got %u\n", in); | |
295 | return NULL; | |
296 | } | |
297 | ||
298 | pkt = kzalloc(sizeof(*pkt), GFP_KERNEL); | |
299 | if (!pkt) | |
300 | return NULL; | |
301 | ||
302 | len = iov_length(vq->iov, out); | |
303 | iov_iter_init(&iov_iter, WRITE, vq->iov, out, len); | |
304 | ||
305 | nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); | |
306 | if (nbytes != sizeof(pkt->hdr)) { | |
307 | vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n", | |
308 | sizeof(pkt->hdr), nbytes); | |
309 | kfree(pkt); | |
310 | return NULL; | |
311 | } | |
312 | ||
313 | if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM) | |
314 | pkt->len = le32_to_cpu(pkt->hdr.len); | |
315 | ||
316 | /* No payload */ | |
317 | if (!pkt->len) | |
318 | return pkt; | |
319 | ||
320 | /* The pkt is too big */ | |
321 | if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) { | |
322 | kfree(pkt); | |
323 | return NULL; | |
324 | } | |
325 | ||
326 | pkt->buf = kmalloc(pkt->len, GFP_KERNEL); | |
327 | if (!pkt->buf) { | |
328 | kfree(pkt); | |
329 | return NULL; | |
330 | } | |
331 | ||
332 | nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter); | |
333 | if (nbytes != pkt->len) { | |
334 | vq_err(vq, "Expected %u byte payload, got %zu bytes\n", | |
335 | pkt->len, nbytes); | |
336 | virtio_transport_free_pkt(pkt); | |
337 | return NULL; | |
338 | } | |
339 | ||
340 | return pkt; | |
341 | } | |
342 | ||
343 | /* Is there space left for replies to rx packets? */ | |
344 | static bool vhost_vsock_more_replies(struct vhost_vsock *vsock) | |
345 | { | |
346 | struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX]; | |
347 | int val; | |
348 | ||
349 | smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */ | |
350 | val = atomic_read(&vsock->queued_replies); | |
351 | ||
352 | return val < vq->num; | |
353 | } | |
354 | ||
355 | static void vhost_vsock_handle_tx_kick(struct vhost_work *work) | |
356 | { | |
357 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
358 | poll.work); | |
359 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
360 | dev); | |
361 | struct virtio_vsock_pkt *pkt; | |
e79b431f | 362 | int head, pkts = 0, total_len = 0; |
433fc58e AH |
363 | unsigned int out, in; |
364 | bool added = false; | |
365 | ||
366 | mutex_lock(&vq->mutex); | |
367 | ||
368 | if (!vq->private_data) | |
369 | goto out; | |
370 | ||
371 | vhost_disable_notify(&vsock->dev, vq); | |
e79b431f | 372 | do { |
3fda5d6e SH |
373 | u32 len; |
374 | ||
433fc58e AH |
375 | if (!vhost_vsock_more_replies(vsock)) { |
376 | /* Stop tx until the device processes already | |
377 | * pending replies. Leave tx virtqueue | |
378 | * callbacks disabled. | |
379 | */ | |
380 | goto no_more_replies; | |
381 | } | |
382 | ||
383 | head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), | |
384 | &out, &in, NULL, NULL); | |
385 | if (head < 0) | |
386 | break; | |
387 | ||
388 | if (head == vq->num) { | |
389 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { | |
390 | vhost_disable_notify(&vsock->dev, vq); | |
391 | continue; | |
392 | } | |
393 | break; | |
394 | } | |
395 | ||
396 | pkt = vhost_vsock_alloc_pkt(vq, out, in); | |
397 | if (!pkt) { | |
398 | vq_err(vq, "Faulted on pkt\n"); | |
399 | continue; | |
400 | } | |
401 | ||
3fda5d6e SH |
402 | len = pkt->len; |
403 | ||
82dfb540 GG |
404 | /* Deliver to monitoring devices all received packets */ |
405 | virtio_transport_deliver_tap_pkt(pkt); | |
406 | ||
433fc58e AH |
407 | /* Only accept correctly addressed packets */ |
408 | if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid) | |
409 | virtio_transport_recv_pkt(pkt); | |
410 | else | |
411 | virtio_transport_free_pkt(pkt); | |
412 | ||
e79b431f JW |
413 | len += sizeof(pkt->hdr); |
414 | vhost_add_used(vq, head, len); | |
415 | total_len += len; | |
433fc58e | 416 | added = true; |
e79b431f | 417 | } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); |
433fc58e AH |
418 | |
419 | no_more_replies: | |
420 | if (added) | |
421 | vhost_signal(&vsock->dev, vq); | |
422 | ||
423 | out: | |
424 | mutex_unlock(&vq->mutex); | |
425 | } | |
426 | ||
427 | static void vhost_vsock_handle_rx_kick(struct vhost_work *work) | |
428 | { | |
429 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
430 | poll.work); | |
431 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
432 | dev); | |
433 | ||
434 | vhost_transport_do_send_pkt(vsock, vq); | |
435 | } | |
436 | ||
437 | static int vhost_vsock_start(struct vhost_vsock *vsock) | |
438 | { | |
0516ffd8 | 439 | struct vhost_virtqueue *vq; |
433fc58e AH |
440 | size_t i; |
441 | int ret; | |
442 | ||
443 | mutex_lock(&vsock->dev.mutex); | |
444 | ||
445 | ret = vhost_dev_check_owner(&vsock->dev); | |
446 | if (ret) | |
447 | goto err; | |
448 | ||
449 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
0516ffd8 | 450 | vq = &vsock->vqs[i]; |
433fc58e AH |
451 | |
452 | mutex_lock(&vq->mutex); | |
453 | ||
454 | if (!vhost_vq_access_ok(vq)) { | |
455 | ret = -EFAULT; | |
433fc58e AH |
456 | goto err_vq; |
457 | } | |
458 | ||
459 | if (!vq->private_data) { | |
460 | vq->private_data = vsock; | |
0516ffd8 SH |
461 | ret = vhost_vq_init_access(vq); |
462 | if (ret) | |
463 | goto err_vq; | |
433fc58e AH |
464 | } |
465 | ||
466 | mutex_unlock(&vq->mutex); | |
467 | } | |
468 | ||
469 | mutex_unlock(&vsock->dev.mutex); | |
470 | return 0; | |
471 | ||
472 | err_vq: | |
0516ffd8 SH |
473 | vq->private_data = NULL; |
474 | mutex_unlock(&vq->mutex); | |
475 | ||
433fc58e | 476 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
0516ffd8 | 477 | vq = &vsock->vqs[i]; |
433fc58e AH |
478 | |
479 | mutex_lock(&vq->mutex); | |
480 | vq->private_data = NULL; | |
481 | mutex_unlock(&vq->mutex); | |
482 | } | |
483 | err: | |
484 | mutex_unlock(&vsock->dev.mutex); | |
485 | return ret; | |
486 | } | |
487 | ||
488 | static int vhost_vsock_stop(struct vhost_vsock *vsock) | |
489 | { | |
490 | size_t i; | |
491 | int ret; | |
492 | ||
493 | mutex_lock(&vsock->dev.mutex); | |
494 | ||
495 | ret = vhost_dev_check_owner(&vsock->dev); | |
496 | if (ret) | |
497 | goto err; | |
498 | ||
499 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
500 | struct vhost_virtqueue *vq = &vsock->vqs[i]; | |
501 | ||
502 | mutex_lock(&vq->mutex); | |
503 | vq->private_data = NULL; | |
504 | mutex_unlock(&vq->mutex); | |
505 | } | |
506 | ||
507 | err: | |
508 | mutex_unlock(&vsock->dev.mutex); | |
509 | return ret; | |
510 | } | |
511 | ||
512 | static void vhost_vsock_free(struct vhost_vsock *vsock) | |
513 | { | |
b226acab | 514 | kvfree(vsock); |
433fc58e AH |
515 | } |
516 | ||
517 | static int vhost_vsock_dev_open(struct inode *inode, struct file *file) | |
518 | { | |
519 | struct vhost_virtqueue **vqs; | |
520 | struct vhost_vsock *vsock; | |
521 | int ret; | |
522 | ||
523 | /* This struct is large and allocation could fail, fall back to vmalloc | |
524 | * if there is no other way. | |
525 | */ | |
dcda9b04 | 526 | vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
6c5ab651 MH |
527 | if (!vsock) |
528 | return -ENOMEM; | |
433fc58e AH |
529 | |
530 | vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL); | |
531 | if (!vqs) { | |
532 | ret = -ENOMEM; | |
533 | goto out; | |
534 | } | |
535 | ||
a72b69dc SH |
536 | vsock->guest_cid = 0; /* no CID assigned yet */ |
537 | ||
433fc58e AH |
538 | atomic_set(&vsock->queued_replies, 0); |
539 | ||
540 | vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX]; | |
541 | vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX]; | |
542 | vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick; | |
543 | vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick; | |
544 | ||
e82b9b07 JW |
545 | vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs), |
546 | UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT, | |
547 | VHOST_VSOCK_WEIGHT); | |
433fc58e AH |
548 | |
549 | file->private_data = vsock; | |
550 | spin_lock_init(&vsock->send_pkt_list_lock); | |
551 | INIT_LIST_HEAD(&vsock->send_pkt_list); | |
552 | vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work); | |
433fc58e AH |
553 | return 0; |
554 | ||
555 | out: | |
556 | vhost_vsock_free(vsock); | |
557 | return ret; | |
558 | } | |
559 | ||
560 | static void vhost_vsock_flush(struct vhost_vsock *vsock) | |
561 | { | |
562 | int i; | |
563 | ||
564 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) | |
565 | if (vsock->vqs[i].handle_kick) | |
566 | vhost_poll_flush(&vsock->vqs[i].poll); | |
567 | vhost_work_flush(&vsock->dev, &vsock->send_pkt_work); | |
568 | } | |
569 | ||
570 | static void vhost_vsock_reset_orphans(struct sock *sk) | |
571 | { | |
572 | struct vsock_sock *vsk = vsock_sk(sk); | |
573 | ||
574 | /* vmci_transport.c doesn't take sk_lock here either. At least we're | |
575 | * under vsock_table_lock so the sock cannot disappear while we're | |
576 | * executing. | |
577 | */ | |
578 | ||
c38f57da SH |
579 | /* If the peer is still valid, no need to reset connection */ |
580 | if (vhost_vsock_get(vsk->remote_addr.svm_cid)) | |
581 | return; | |
582 | ||
583 | /* If the close timeout is pending, let it expire. This avoids races | |
584 | * with the timeout callback. | |
585 | */ | |
586 | if (vsk->close_work_scheduled) | |
587 | return; | |
588 | ||
589 | sock_set_flag(sk, SOCK_DONE); | |
590 | vsk->peer_shutdown = SHUTDOWN_MASK; | |
591 | sk->sk_state = SS_UNCONNECTED; | |
592 | sk->sk_err = ECONNRESET; | |
593 | sk->sk_error_report(sk); | |
433fc58e AH |
594 | } |
595 | ||
596 | static int vhost_vsock_dev_release(struct inode *inode, struct file *file) | |
597 | { | |
598 | struct vhost_vsock *vsock = file->private_data; | |
599 | ||
6db3d8dc | 600 | mutex_lock(&vhost_vsock_mutex); |
834e772c SH |
601 | if (vsock->guest_cid) |
602 | hash_del_rcu(&vsock->hash); | |
6db3d8dc | 603 | mutex_unlock(&vhost_vsock_mutex); |
433fc58e | 604 | |
834e772c SH |
605 | /* Wait for other CPUs to finish using vsock */ |
606 | synchronize_rcu(); | |
607 | ||
433fc58e AH |
608 | /* Iterating over all connections for all CIDs to find orphans is |
609 | * inefficient. Room for improvement here. */ | |
610 | vsock_for_each_connected_socket(vhost_vsock_reset_orphans); | |
611 | ||
612 | vhost_vsock_stop(vsock); | |
613 | vhost_vsock_flush(vsock); | |
614 | vhost_dev_stop(&vsock->dev); | |
615 | ||
616 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
617 | while (!list_empty(&vsock->send_pkt_list)) { | |
618 | struct virtio_vsock_pkt *pkt; | |
619 | ||
620 | pkt = list_first_entry(&vsock->send_pkt_list, | |
621 | struct virtio_vsock_pkt, list); | |
622 | list_del_init(&pkt->list); | |
623 | virtio_transport_free_pkt(pkt); | |
624 | } | |
625 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
626 | ||
f6f93f75 | 627 | vhost_dev_cleanup(&vsock->dev); |
433fc58e AH |
628 | kfree(vsock->dev.vqs); |
629 | vhost_vsock_free(vsock); | |
630 | return 0; | |
631 | } | |
632 | ||
633 | static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid) | |
634 | { | |
635 | struct vhost_vsock *other; | |
636 | ||
637 | /* Refuse reserved CIDs */ | |
638 | if (guest_cid <= VMADDR_CID_HOST || | |
639 | guest_cid == U32_MAX) | |
640 | return -EINVAL; | |
641 | ||
642 | /* 64-bit CIDs are not yet supported */ | |
643 | if (guest_cid > U32_MAX) | |
644 | return -EINVAL; | |
645 | ||
646 | /* Refuse if CID is already in use */ | |
6db3d8dc | 647 | mutex_lock(&vhost_vsock_mutex); |
834e772c | 648 | other = vhost_vsock_get(guest_cid); |
6c083c2b | 649 | if (other && other != vsock) { |
6db3d8dc | 650 | mutex_unlock(&vhost_vsock_mutex); |
6c083c2b G |
651 | return -EADDRINUSE; |
652 | } | |
834e772c SH |
653 | |
654 | if (vsock->guest_cid) | |
655 | hash_del_rcu(&vsock->hash); | |
656 | ||
433fc58e | 657 | vsock->guest_cid = guest_cid; |
7fbe078c | 658 | hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid); |
6db3d8dc | 659 | mutex_unlock(&vhost_vsock_mutex); |
433fc58e AH |
660 | |
661 | return 0; | |
662 | } | |
663 | ||
664 | static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features) | |
665 | { | |
666 | struct vhost_virtqueue *vq; | |
667 | int i; | |
668 | ||
669 | if (features & ~VHOST_VSOCK_FEATURES) | |
670 | return -EOPNOTSUPP; | |
671 | ||
672 | mutex_lock(&vsock->dev.mutex); | |
673 | if ((features & (1 << VHOST_F_LOG_ALL)) && | |
674 | !vhost_log_access_ok(&vsock->dev)) { | |
675 | mutex_unlock(&vsock->dev.mutex); | |
676 | return -EFAULT; | |
677 | } | |
678 | ||
679 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
680 | vq = &vsock->vqs[i]; | |
681 | mutex_lock(&vq->mutex); | |
682 | vq->acked_features = features; | |
683 | mutex_unlock(&vq->mutex); | |
684 | } | |
685 | mutex_unlock(&vsock->dev.mutex); | |
686 | return 0; | |
687 | } | |
688 | ||
689 | static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl, | |
690 | unsigned long arg) | |
691 | { | |
692 | struct vhost_vsock *vsock = f->private_data; | |
693 | void __user *argp = (void __user *)arg; | |
694 | u64 guest_cid; | |
695 | u64 features; | |
696 | int start; | |
697 | int r; | |
698 | ||
699 | switch (ioctl) { | |
700 | case VHOST_VSOCK_SET_GUEST_CID: | |
701 | if (copy_from_user(&guest_cid, argp, sizeof(guest_cid))) | |
702 | return -EFAULT; | |
703 | return vhost_vsock_set_cid(vsock, guest_cid); | |
704 | case VHOST_VSOCK_SET_RUNNING: | |
705 | if (copy_from_user(&start, argp, sizeof(start))) | |
706 | return -EFAULT; | |
707 | if (start) | |
708 | return vhost_vsock_start(vsock); | |
709 | else | |
710 | return vhost_vsock_stop(vsock); | |
711 | case VHOST_GET_FEATURES: | |
712 | features = VHOST_VSOCK_FEATURES; | |
713 | if (copy_to_user(argp, &features, sizeof(features))) | |
714 | return -EFAULT; | |
715 | return 0; | |
716 | case VHOST_SET_FEATURES: | |
717 | if (copy_from_user(&features, argp, sizeof(features))) | |
718 | return -EFAULT; | |
719 | return vhost_vsock_set_features(vsock, features); | |
720 | default: | |
721 | mutex_lock(&vsock->dev.mutex); | |
722 | r = vhost_dev_ioctl(&vsock->dev, ioctl, argp); | |
723 | if (r == -ENOIOCTLCMD) | |
724 | r = vhost_vring_ioctl(&vsock->dev, ioctl, argp); | |
725 | else | |
726 | vhost_vsock_flush(vsock); | |
727 | mutex_unlock(&vsock->dev.mutex); | |
728 | return r; | |
729 | } | |
730 | } | |
731 | ||
dc32bb67 SR |
732 | #ifdef CONFIG_COMPAT |
733 | static long vhost_vsock_dev_compat_ioctl(struct file *f, unsigned int ioctl, | |
734 | unsigned long arg) | |
735 | { | |
736 | return vhost_vsock_dev_ioctl(f, ioctl, (unsigned long)compat_ptr(arg)); | |
737 | } | |
738 | #endif | |
739 | ||
433fc58e AH |
740 | static const struct file_operations vhost_vsock_fops = { |
741 | .owner = THIS_MODULE, | |
742 | .open = vhost_vsock_dev_open, | |
743 | .release = vhost_vsock_dev_release, | |
744 | .llseek = noop_llseek, | |
745 | .unlocked_ioctl = vhost_vsock_dev_ioctl, | |
dc32bb67 SR |
746 | #ifdef CONFIG_COMPAT |
747 | .compat_ioctl = vhost_vsock_dev_compat_ioctl, | |
748 | #endif | |
433fc58e AH |
749 | }; |
750 | ||
751 | static struct miscdevice vhost_vsock_misc = { | |
f4660cc9 | 752 | .minor = VHOST_VSOCK_MINOR, |
433fc58e AH |
753 | .name = "vhost-vsock", |
754 | .fops = &vhost_vsock_fops, | |
755 | }; | |
756 | ||
757 | static struct virtio_transport vhost_transport = { | |
758 | .transport = { | |
759 | .get_local_cid = vhost_transport_get_local_cid, | |
760 | ||
761 | .init = virtio_transport_do_socket_init, | |
762 | .destruct = virtio_transport_destruct, | |
763 | .release = virtio_transport_release, | |
764 | .connect = virtio_transport_connect, | |
765 | .shutdown = virtio_transport_shutdown, | |
16320f36 | 766 | .cancel_pkt = vhost_transport_cancel_pkt, |
433fc58e AH |
767 | |
768 | .dgram_enqueue = virtio_transport_dgram_enqueue, | |
769 | .dgram_dequeue = virtio_transport_dgram_dequeue, | |
770 | .dgram_bind = virtio_transport_dgram_bind, | |
771 | .dgram_allow = virtio_transport_dgram_allow, | |
772 | ||
773 | .stream_enqueue = virtio_transport_stream_enqueue, | |
774 | .stream_dequeue = virtio_transport_stream_dequeue, | |
775 | .stream_has_data = virtio_transport_stream_has_data, | |
776 | .stream_has_space = virtio_transport_stream_has_space, | |
777 | .stream_rcvhiwat = virtio_transport_stream_rcvhiwat, | |
778 | .stream_is_active = virtio_transport_stream_is_active, | |
779 | .stream_allow = virtio_transport_stream_allow, | |
780 | ||
781 | .notify_poll_in = virtio_transport_notify_poll_in, | |
782 | .notify_poll_out = virtio_transport_notify_poll_out, | |
783 | .notify_recv_init = virtio_transport_notify_recv_init, | |
784 | .notify_recv_pre_block = virtio_transport_notify_recv_pre_block, | |
785 | .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue, | |
786 | .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue, | |
787 | .notify_send_init = virtio_transport_notify_send_init, | |
788 | .notify_send_pre_block = virtio_transport_notify_send_pre_block, | |
789 | .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue, | |
790 | .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue, | |
791 | ||
792 | .set_buffer_size = virtio_transport_set_buffer_size, | |
793 | .set_min_buffer_size = virtio_transport_set_min_buffer_size, | |
794 | .set_max_buffer_size = virtio_transport_set_max_buffer_size, | |
795 | .get_buffer_size = virtio_transport_get_buffer_size, | |
796 | .get_min_buffer_size = virtio_transport_get_min_buffer_size, | |
797 | .get_max_buffer_size = virtio_transport_get_max_buffer_size, | |
798 | }, | |
799 | ||
800 | .send_pkt = vhost_transport_send_pkt, | |
801 | }; | |
802 | ||
803 | static int __init vhost_vsock_init(void) | |
804 | { | |
805 | int ret; | |
806 | ||
807 | ret = vsock_core_init(&vhost_transport.transport); | |
808 | if (ret < 0) | |
809 | return ret; | |
810 | return misc_register(&vhost_vsock_misc); | |
811 | }; | |
812 | ||
813 | static void __exit vhost_vsock_exit(void) | |
814 | { | |
815 | misc_deregister(&vhost_vsock_misc); | |
816 | vsock_core_exit(); | |
817 | }; | |
818 | ||
819 | module_init(vhost_vsock_init); | |
820 | module_exit(vhost_vsock_exit); | |
821 | MODULE_LICENSE("GPL v2"); | |
822 | MODULE_AUTHOR("Asias He"); | |
823 | MODULE_DESCRIPTION("vhost transport for vsock "); | |
f4660cc9 SH |
824 | MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR); |
825 | MODULE_ALIAS("devname:vhost-vsock"); |