]>
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> | |
491e2809 | 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 | |
9564f70c 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. */ | |
38 | static DEFINE_SPINLOCK(vhost_vsock_lock); | |
491e2809 | 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 | ||
491e2809 SH |
45 | /* Link to global vhost_vsock_hash, writes use vhost_vsock_lock */ |
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 | ||
491e2809 SH |
62 | /* Callers that dereference the return value must hold vhost_vsock_lock or the |
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 | ||
491e2809 | 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 | ||
76 | if (other_cid == guest_cid) { | |
433fc58e AH |
77 | return vsock; |
78 | } | |
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]; | |
67b8b619 | 89 | int pkts = 0, total_len = 0; |
433fc58e AH |
90 | bool added = false; |
91 | bool restart_tx = false; | |
92 | ||
93 | mutex_lock(&vq->mutex); | |
94 | ||
95 | if (!vq->private_data) | |
96 | goto out; | |
97 | ||
98 | /* Avoid further vmexits, we're already processing the virtqueue */ | |
99 | vhost_disable_notify(&vsock->dev, vq); | |
100 | ||
67b8b619 | 101 | do { |
433fc58e AH |
102 | struct virtio_vsock_pkt *pkt; |
103 | struct iov_iter iov_iter; | |
104 | unsigned out, in; | |
105 | size_t nbytes; | |
10de4332 | 106 | size_t iov_len, payload_len; |
433fc58e AH |
107 | int head; |
108 | ||
109 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
110 | if (list_empty(&vsock->send_pkt_list)) { | |
111 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
112 | vhost_enable_notify(&vsock->dev, vq); | |
113 | break; | |
114 | } | |
115 | ||
116 | pkt = list_first_entry(&vsock->send_pkt_list, | |
117 | struct virtio_vsock_pkt, list); | |
118 | list_del_init(&pkt->list); | |
119 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
120 | ||
121 | head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), | |
122 | &out, &in, NULL, NULL); | |
123 | if (head < 0) { | |
124 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
125 | list_add(&pkt->list, &vsock->send_pkt_list); | |
126 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
127 | break; | |
128 | } | |
129 | ||
130 | if (head == vq->num) { | |
131 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
132 | list_add(&pkt->list, &vsock->send_pkt_list); | |
133 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
134 | ||
135 | /* We cannot finish yet if more buffers snuck in while | |
136 | * re-enabling notify. | |
137 | */ | |
138 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { | |
139 | vhost_disable_notify(&vsock->dev, vq); | |
140 | continue; | |
141 | } | |
142 | break; | |
143 | } | |
144 | ||
145 | if (out) { | |
146 | virtio_transport_free_pkt(pkt); | |
147 | vq_err(vq, "Expected 0 output buffers, got %u\n", out); | |
148 | break; | |
149 | } | |
150 | ||
10de4332 SG |
151 | iov_len = iov_length(&vq->iov[out], in); |
152 | if (iov_len < sizeof(pkt->hdr)) { | |
153 | virtio_transport_free_pkt(pkt); | |
154 | vq_err(vq, "Buffer len [%zu] too small\n", iov_len); | |
155 | break; | |
156 | } | |
157 | ||
158 | iov_iter_init(&iov_iter, READ, &vq->iov[out], in, iov_len); | |
159 | payload_len = pkt->len - pkt->off; | |
160 | ||
161 | /* If the packet is greater than the space available in the | |
162 | * buffer, we split it using multiple buffers. | |
163 | */ | |
164 | if (payload_len > iov_len - sizeof(pkt->hdr)) | |
165 | payload_len = iov_len - sizeof(pkt->hdr); | |
166 | ||
167 | /* Set the correct length in the header */ | |
168 | pkt->hdr.len = cpu_to_le32(payload_len); | |
433fc58e AH |
169 | |
170 | nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); | |
171 | if (nbytes != sizeof(pkt->hdr)) { | |
172 | virtio_transport_free_pkt(pkt); | |
173 | vq_err(vq, "Faulted on copying pkt hdr\n"); | |
174 | break; | |
175 | } | |
176 | ||
10de4332 SG |
177 | nbytes = copy_to_iter(pkt->buf + pkt->off, payload_len, |
178 | &iov_iter); | |
179 | if (nbytes != payload_len) { | |
433fc58e AH |
180 | virtio_transport_free_pkt(pkt); |
181 | vq_err(vq, "Faulted on copying pkt buf\n"); | |
182 | break; | |
183 | } | |
184 | ||
10de4332 | 185 | vhost_add_used(vq, head, sizeof(pkt->hdr) + payload_len); |
433fc58e AH |
186 | added = true; |
187 | ||
82dfb540 GG |
188 | /* Deliver to monitoring devices all correctly transmitted |
189 | * packets. | |
190 | */ | |
191 | virtio_transport_deliver_tap_pkt(pkt); | |
192 | ||
10de4332 SG |
193 | pkt->off += payload_len; |
194 | total_len += payload_len; | |
195 | ||
196 | /* If we didn't send all the payload we can requeue the packet | |
197 | * to send it with the next available buffer. | |
198 | */ | |
199 | if (pkt->off < pkt->len) { | |
200 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
201 | list_add(&pkt->list, &vsock->send_pkt_list); | |
202 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
203 | } else { | |
204 | if (pkt->reply) { | |
205 | int val; | |
206 | ||
207 | val = atomic_dec_return(&vsock->queued_replies); | |
208 | ||
209 | /* Do we have resources to resume tx | |
210 | * processing? | |
211 | */ | |
212 | if (val + 1 == tx_vq->num) | |
213 | restart_tx = true; | |
214 | } | |
215 | ||
216 | virtio_transport_free_pkt(pkt); | |
217 | } | |
67b8b619 | 218 | } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); |
433fc58e AH |
219 | if (added) |
220 | vhost_signal(&vsock->dev, vq); | |
221 | ||
222 | out: | |
223 | mutex_unlock(&vq->mutex); | |
224 | ||
225 | if (restart_tx) | |
226 | vhost_poll_queue(&tx_vq->poll); | |
227 | } | |
228 | ||
229 | static void vhost_transport_send_pkt_work(struct vhost_work *work) | |
230 | { | |
231 | struct vhost_virtqueue *vq; | |
232 | struct vhost_vsock *vsock; | |
233 | ||
234 | vsock = container_of(work, struct vhost_vsock, send_pkt_work); | |
235 | vq = &vsock->vqs[VSOCK_VQ_RX]; | |
236 | ||
237 | vhost_transport_do_send_pkt(vsock, vq); | |
238 | } | |
239 | ||
240 | static int | |
241 | vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt) | |
242 | { | |
243 | struct vhost_vsock *vsock; | |
433fc58e AH |
244 | int len = pkt->len; |
245 | ||
491e2809 SH |
246 | rcu_read_lock(); |
247 | ||
433fc58e AH |
248 | /* Find the vhost_vsock according to guest context id */ |
249 | vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid)); | |
250 | if (!vsock) { | |
491e2809 | 251 | rcu_read_unlock(); |
433fc58e AH |
252 | virtio_transport_free_pkt(pkt); |
253 | return -ENODEV; | |
254 | } | |
255 | ||
433fc58e AH |
256 | if (pkt->reply) |
257 | atomic_inc(&vsock->queued_replies); | |
258 | ||
259 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
260 | list_add_tail(&pkt->list, &vsock->send_pkt_list); | |
261 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
262 | ||
263 | vhost_work_queue(&vsock->dev, &vsock->send_pkt_work); | |
491e2809 SH |
264 | |
265 | rcu_read_unlock(); | |
433fc58e AH |
266 | return len; |
267 | } | |
268 | ||
16320f36 PT |
269 | static int |
270 | vhost_transport_cancel_pkt(struct vsock_sock *vsk) | |
271 | { | |
272 | struct vhost_vsock *vsock; | |
273 | struct virtio_vsock_pkt *pkt, *n; | |
274 | int cnt = 0; | |
491e2809 | 275 | int ret = -ENODEV; |
16320f36 PT |
276 | LIST_HEAD(freeme); |
277 | ||
491e2809 SH |
278 | rcu_read_lock(); |
279 | ||
16320f36 PT |
280 | /* Find the vhost_vsock according to guest context id */ |
281 | vsock = vhost_vsock_get(vsk->remote_addr.svm_cid); | |
282 | if (!vsock) | |
491e2809 | 283 | goto out; |
16320f36 PT |
284 | |
285 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
286 | list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) { | |
287 | if (pkt->vsk != vsk) | |
288 | continue; | |
289 | list_move(&pkt->list, &freeme); | |
290 | } | |
291 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
292 | ||
293 | list_for_each_entry_safe(pkt, n, &freeme, list) { | |
294 | if (pkt->reply) | |
295 | cnt++; | |
296 | list_del(&pkt->list); | |
297 | virtio_transport_free_pkt(pkt); | |
298 | } | |
299 | ||
300 | if (cnt) { | |
301 | struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX]; | |
302 | int new_cnt; | |
303 | ||
304 | new_cnt = atomic_sub_return(cnt, &vsock->queued_replies); | |
305 | if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num) | |
306 | vhost_poll_queue(&tx_vq->poll); | |
307 | } | |
308 | ||
491e2809 SH |
309 | ret = 0; |
310 | out: | |
311 | rcu_read_unlock(); | |
312 | return ret; | |
16320f36 PT |
313 | } |
314 | ||
433fc58e AH |
315 | static struct virtio_vsock_pkt * |
316 | vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq, | |
317 | unsigned int out, unsigned int in) | |
318 | { | |
319 | struct virtio_vsock_pkt *pkt; | |
320 | struct iov_iter iov_iter; | |
321 | size_t nbytes; | |
322 | size_t len; | |
323 | ||
324 | if (in != 0) { | |
325 | vq_err(vq, "Expected 0 input buffers, got %u\n", in); | |
326 | return NULL; | |
327 | } | |
328 | ||
329 | pkt = kzalloc(sizeof(*pkt), GFP_KERNEL); | |
330 | if (!pkt) | |
331 | return NULL; | |
332 | ||
333 | len = iov_length(vq->iov, out); | |
334 | iov_iter_init(&iov_iter, WRITE, vq->iov, out, len); | |
335 | ||
336 | nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); | |
337 | if (nbytes != sizeof(pkt->hdr)) { | |
338 | vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n", | |
339 | sizeof(pkt->hdr), nbytes); | |
340 | kfree(pkt); | |
341 | return NULL; | |
342 | } | |
343 | ||
344 | if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM) | |
345 | pkt->len = le32_to_cpu(pkt->hdr.len); | |
346 | ||
347 | /* No payload */ | |
348 | if (!pkt->len) | |
349 | return pkt; | |
350 | ||
351 | /* The pkt is too big */ | |
352 | if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) { | |
353 | kfree(pkt); | |
354 | return NULL; | |
355 | } | |
356 | ||
357 | pkt->buf = kmalloc(pkt->len, GFP_KERNEL); | |
358 | if (!pkt->buf) { | |
359 | kfree(pkt); | |
360 | return NULL; | |
361 | } | |
362 | ||
363 | nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter); | |
364 | if (nbytes != pkt->len) { | |
365 | vq_err(vq, "Expected %u byte payload, got %zu bytes\n", | |
366 | pkt->len, nbytes); | |
367 | virtio_transport_free_pkt(pkt); | |
368 | return NULL; | |
369 | } | |
370 | ||
371 | return pkt; | |
372 | } | |
373 | ||
374 | /* Is there space left for replies to rx packets? */ | |
375 | static bool vhost_vsock_more_replies(struct vhost_vsock *vsock) | |
376 | { | |
377 | struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX]; | |
378 | int val; | |
379 | ||
380 | smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */ | |
381 | val = atomic_read(&vsock->queued_replies); | |
382 | ||
383 | return val < vq->num; | |
384 | } | |
385 | ||
386 | static void vhost_vsock_handle_tx_kick(struct vhost_work *work) | |
387 | { | |
388 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
389 | poll.work); | |
390 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
391 | dev); | |
392 | struct virtio_vsock_pkt *pkt; | |
67b8b619 | 393 | int head, pkts = 0, total_len = 0; |
433fc58e AH |
394 | unsigned int out, in; |
395 | bool added = false; | |
396 | ||
397 | mutex_lock(&vq->mutex); | |
398 | ||
399 | if (!vq->private_data) | |
400 | goto out; | |
401 | ||
402 | vhost_disable_notify(&vsock->dev, vq); | |
67b8b619 | 403 | do { |
3fda5d6e SH |
404 | u32 len; |
405 | ||
433fc58e AH |
406 | if (!vhost_vsock_more_replies(vsock)) { |
407 | /* Stop tx until the device processes already | |
408 | * pending replies. Leave tx virtqueue | |
409 | * callbacks disabled. | |
410 | */ | |
411 | goto no_more_replies; | |
412 | } | |
413 | ||
414 | head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), | |
415 | &out, &in, NULL, NULL); | |
416 | if (head < 0) | |
417 | break; | |
418 | ||
419 | if (head == vq->num) { | |
420 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { | |
421 | vhost_disable_notify(&vsock->dev, vq); | |
422 | continue; | |
423 | } | |
424 | break; | |
425 | } | |
426 | ||
427 | pkt = vhost_vsock_alloc_pkt(vq, out, in); | |
428 | if (!pkt) { | |
429 | vq_err(vq, "Faulted on pkt\n"); | |
430 | continue; | |
431 | } | |
432 | ||
3fda5d6e SH |
433 | len = pkt->len; |
434 | ||
82dfb540 GG |
435 | /* Deliver to monitoring devices all received packets */ |
436 | virtio_transport_deliver_tap_pkt(pkt); | |
437 | ||
433fc58e | 438 | /* Only accept correctly addressed packets */ |
27b5e941 SG |
439 | if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid && |
440 | le64_to_cpu(pkt->hdr.dst_cid) == | |
441 | vhost_transport_get_local_cid()) | |
433fc58e AH |
442 | virtio_transport_recv_pkt(pkt); |
443 | else | |
444 | virtio_transport_free_pkt(pkt); | |
445 | ||
67b8b619 JW |
446 | len += sizeof(pkt->hdr); |
447 | vhost_add_used(vq, head, len); | |
448 | total_len += len; | |
433fc58e | 449 | added = true; |
67b8b619 | 450 | } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); |
433fc58e AH |
451 | |
452 | no_more_replies: | |
453 | if (added) | |
454 | vhost_signal(&vsock->dev, vq); | |
455 | ||
456 | out: | |
457 | mutex_unlock(&vq->mutex); | |
458 | } | |
459 | ||
460 | static void vhost_vsock_handle_rx_kick(struct vhost_work *work) | |
461 | { | |
462 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
463 | poll.work); | |
464 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
465 | dev); | |
466 | ||
467 | vhost_transport_do_send_pkt(vsock, vq); | |
468 | } | |
469 | ||
470 | static int vhost_vsock_start(struct vhost_vsock *vsock) | |
471 | { | |
0516ffd8 | 472 | struct vhost_virtqueue *vq; |
433fc58e AH |
473 | size_t i; |
474 | int ret; | |
475 | ||
476 | mutex_lock(&vsock->dev.mutex); | |
477 | ||
478 | ret = vhost_dev_check_owner(&vsock->dev); | |
479 | if (ret) | |
480 | goto err; | |
481 | ||
482 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
0516ffd8 | 483 | vq = &vsock->vqs[i]; |
433fc58e AH |
484 | |
485 | mutex_lock(&vq->mutex); | |
486 | ||
487 | if (!vhost_vq_access_ok(vq)) { | |
488 | ret = -EFAULT; | |
433fc58e AH |
489 | goto err_vq; |
490 | } | |
491 | ||
492 | if (!vq->private_data) { | |
493 | vq->private_data = vsock; | |
0516ffd8 SH |
494 | ret = vhost_vq_init_access(vq); |
495 | if (ret) | |
496 | goto err_vq; | |
433fc58e AH |
497 | } |
498 | ||
499 | mutex_unlock(&vq->mutex); | |
500 | } | |
501 | ||
502 | mutex_unlock(&vsock->dev.mutex); | |
503 | return 0; | |
504 | ||
505 | err_vq: | |
0516ffd8 SH |
506 | vq->private_data = NULL; |
507 | mutex_unlock(&vq->mutex); | |
508 | ||
433fc58e | 509 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
0516ffd8 | 510 | vq = &vsock->vqs[i]; |
433fc58e AH |
511 | |
512 | mutex_lock(&vq->mutex); | |
513 | vq->private_data = NULL; | |
514 | mutex_unlock(&vq->mutex); | |
515 | } | |
516 | err: | |
517 | mutex_unlock(&vsock->dev.mutex); | |
518 | return ret; | |
519 | } | |
520 | ||
521 | static int vhost_vsock_stop(struct vhost_vsock *vsock) | |
522 | { | |
523 | size_t i; | |
524 | int ret; | |
525 | ||
526 | mutex_lock(&vsock->dev.mutex); | |
527 | ||
528 | ret = vhost_dev_check_owner(&vsock->dev); | |
529 | if (ret) | |
530 | goto err; | |
531 | ||
532 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
533 | struct vhost_virtqueue *vq = &vsock->vqs[i]; | |
534 | ||
535 | mutex_lock(&vq->mutex); | |
536 | vq->private_data = NULL; | |
537 | mutex_unlock(&vq->mutex); | |
538 | } | |
539 | ||
540 | err: | |
541 | mutex_unlock(&vsock->dev.mutex); | |
542 | return ret; | |
543 | } | |
544 | ||
545 | static void vhost_vsock_free(struct vhost_vsock *vsock) | |
546 | { | |
b226acab | 547 | kvfree(vsock); |
433fc58e AH |
548 | } |
549 | ||
550 | static int vhost_vsock_dev_open(struct inode *inode, struct file *file) | |
551 | { | |
552 | struct vhost_virtqueue **vqs; | |
553 | struct vhost_vsock *vsock; | |
554 | int ret; | |
555 | ||
556 | /* This struct is large and allocation could fail, fall back to vmalloc | |
557 | * if there is no other way. | |
558 | */ | |
dcda9b04 | 559 | vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
6c5ab651 MH |
560 | if (!vsock) |
561 | return -ENOMEM; | |
433fc58e AH |
562 | |
563 | vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL); | |
564 | if (!vqs) { | |
565 | ret = -ENOMEM; | |
566 | goto out; | |
567 | } | |
568 | ||
a72b69dc SH |
569 | vsock->guest_cid = 0; /* no CID assigned yet */ |
570 | ||
433fc58e AH |
571 | atomic_set(&vsock->queued_replies, 0); |
572 | ||
573 | vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX]; | |
574 | vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX]; | |
575 | vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick; | |
576 | vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick; | |
577 | ||
9564f70c JW |
578 | vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs), |
579 | VHOST_VSOCK_PKT_WEIGHT, VHOST_VSOCK_WEIGHT); | |
433fc58e AH |
580 | |
581 | file->private_data = vsock; | |
582 | spin_lock_init(&vsock->send_pkt_list_lock); | |
583 | INIT_LIST_HEAD(&vsock->send_pkt_list); | |
584 | vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work); | |
433fc58e AH |
585 | return 0; |
586 | ||
587 | out: | |
588 | vhost_vsock_free(vsock); | |
589 | return ret; | |
590 | } | |
591 | ||
592 | static void vhost_vsock_flush(struct vhost_vsock *vsock) | |
593 | { | |
594 | int i; | |
595 | ||
596 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) | |
597 | if (vsock->vqs[i].handle_kick) | |
598 | vhost_poll_flush(&vsock->vqs[i].poll); | |
599 | vhost_work_flush(&vsock->dev, &vsock->send_pkt_work); | |
600 | } | |
601 | ||
602 | static void vhost_vsock_reset_orphans(struct sock *sk) | |
603 | { | |
604 | struct vsock_sock *vsk = vsock_sk(sk); | |
605 | ||
606 | /* vmci_transport.c doesn't take sk_lock here either. At least we're | |
607 | * under vsock_table_lock so the sock cannot disappear while we're | |
608 | * executing. | |
609 | */ | |
610 | ||
7ef12395 SH |
611 | /* If the peer is still valid, no need to reset connection */ |
612 | if (vhost_vsock_get(vsk->remote_addr.svm_cid)) | |
613 | return; | |
614 | ||
615 | /* If the close timeout is pending, let it expire. This avoids races | |
616 | * with the timeout callback. | |
617 | */ | |
618 | if (vsk->close_work_scheduled) | |
619 | return; | |
620 | ||
621 | sock_set_flag(sk, SOCK_DONE); | |
622 | vsk->peer_shutdown = SHUTDOWN_MASK; | |
623 | sk->sk_state = SS_UNCONNECTED; | |
624 | sk->sk_err = ECONNRESET; | |
625 | sk->sk_error_report(sk); | |
433fc58e AH |
626 | } |
627 | ||
628 | static int vhost_vsock_dev_release(struct inode *inode, struct file *file) | |
629 | { | |
630 | struct vhost_vsock *vsock = file->private_data; | |
631 | ||
632 | spin_lock_bh(&vhost_vsock_lock); | |
491e2809 SH |
633 | if (vsock->guest_cid) |
634 | hash_del_rcu(&vsock->hash); | |
433fc58e AH |
635 | spin_unlock_bh(&vhost_vsock_lock); |
636 | ||
491e2809 SH |
637 | /* Wait for other CPUs to finish using vsock */ |
638 | synchronize_rcu(); | |
639 | ||
433fc58e AH |
640 | /* Iterating over all connections for all CIDs to find orphans is |
641 | * inefficient. Room for improvement here. */ | |
642 | vsock_for_each_connected_socket(vhost_vsock_reset_orphans); | |
643 | ||
644 | vhost_vsock_stop(vsock); | |
645 | vhost_vsock_flush(vsock); | |
646 | vhost_dev_stop(&vsock->dev); | |
647 | ||
648 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
649 | while (!list_empty(&vsock->send_pkt_list)) { | |
650 | struct virtio_vsock_pkt *pkt; | |
651 | ||
652 | pkt = list_first_entry(&vsock->send_pkt_list, | |
653 | struct virtio_vsock_pkt, list); | |
654 | list_del_init(&pkt->list); | |
655 | virtio_transport_free_pkt(pkt); | |
656 | } | |
657 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
658 | ||
659 | vhost_dev_cleanup(&vsock->dev, false); | |
660 | kfree(vsock->dev.vqs); | |
661 | vhost_vsock_free(vsock); | |
662 | return 0; | |
663 | } | |
664 | ||
665 | static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid) | |
666 | { | |
667 | struct vhost_vsock *other; | |
668 | ||
669 | /* Refuse reserved CIDs */ | |
670 | if (guest_cid <= VMADDR_CID_HOST || | |
671 | guest_cid == U32_MAX) | |
672 | return -EINVAL; | |
673 | ||
674 | /* 64-bit CIDs are not yet supported */ | |
675 | if (guest_cid > U32_MAX) | |
676 | return -EINVAL; | |
677 | ||
678 | /* Refuse if CID is already in use */ | |
433fc58e | 679 | spin_lock_bh(&vhost_vsock_lock); |
491e2809 | 680 | other = vhost_vsock_get(guest_cid); |
6c083c2b G |
681 | if (other && other != vsock) { |
682 | spin_unlock_bh(&vhost_vsock_lock); | |
683 | return -EADDRINUSE; | |
684 | } | |
491e2809 SH |
685 | |
686 | if (vsock->guest_cid) | |
687 | hash_del_rcu(&vsock->hash); | |
688 | ||
433fc58e | 689 | vsock->guest_cid = guest_cid; |
2f592f0c | 690 | hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid); |
433fc58e AH |
691 | spin_unlock_bh(&vhost_vsock_lock); |
692 | ||
693 | return 0; | |
694 | } | |
695 | ||
696 | static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features) | |
697 | { | |
698 | struct vhost_virtqueue *vq; | |
699 | int i; | |
700 | ||
701 | if (features & ~VHOST_VSOCK_FEATURES) | |
702 | return -EOPNOTSUPP; | |
703 | ||
704 | mutex_lock(&vsock->dev.mutex); | |
705 | if ((features & (1 << VHOST_F_LOG_ALL)) && | |
706 | !vhost_log_access_ok(&vsock->dev)) { | |
707 | mutex_unlock(&vsock->dev.mutex); | |
708 | return -EFAULT; | |
709 | } | |
710 | ||
711 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
712 | vq = &vsock->vqs[i]; | |
713 | mutex_lock(&vq->mutex); | |
714 | vq->acked_features = features; | |
715 | mutex_unlock(&vq->mutex); | |
716 | } | |
717 | mutex_unlock(&vsock->dev.mutex); | |
718 | return 0; | |
719 | } | |
720 | ||
721 | static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl, | |
722 | unsigned long arg) | |
723 | { | |
724 | struct vhost_vsock *vsock = f->private_data; | |
725 | void __user *argp = (void __user *)arg; | |
726 | u64 guest_cid; | |
727 | u64 features; | |
728 | int start; | |
729 | int r; | |
730 | ||
731 | switch (ioctl) { | |
732 | case VHOST_VSOCK_SET_GUEST_CID: | |
733 | if (copy_from_user(&guest_cid, argp, sizeof(guest_cid))) | |
734 | return -EFAULT; | |
735 | return vhost_vsock_set_cid(vsock, guest_cid); | |
736 | case VHOST_VSOCK_SET_RUNNING: | |
737 | if (copy_from_user(&start, argp, sizeof(start))) | |
738 | return -EFAULT; | |
739 | if (start) | |
740 | return vhost_vsock_start(vsock); | |
741 | else | |
742 | return vhost_vsock_stop(vsock); | |
743 | case VHOST_GET_FEATURES: | |
744 | features = VHOST_VSOCK_FEATURES; | |
745 | if (copy_to_user(argp, &features, sizeof(features))) | |
746 | return -EFAULT; | |
747 | return 0; | |
748 | case VHOST_SET_FEATURES: | |
749 | if (copy_from_user(&features, argp, sizeof(features))) | |
750 | return -EFAULT; | |
751 | return vhost_vsock_set_features(vsock, features); | |
752 | default: | |
753 | mutex_lock(&vsock->dev.mutex); | |
754 | r = vhost_dev_ioctl(&vsock->dev, ioctl, argp); | |
755 | if (r == -ENOIOCTLCMD) | |
756 | r = vhost_vring_ioctl(&vsock->dev, ioctl, argp); | |
757 | else | |
758 | vhost_vsock_flush(vsock); | |
759 | mutex_unlock(&vsock->dev.mutex); | |
760 | return r; | |
761 | } | |
762 | } | |
763 | ||
764 | static const struct file_operations vhost_vsock_fops = { | |
765 | .owner = THIS_MODULE, | |
766 | .open = vhost_vsock_dev_open, | |
767 | .release = vhost_vsock_dev_release, | |
768 | .llseek = noop_llseek, | |
769 | .unlocked_ioctl = vhost_vsock_dev_ioctl, | |
770 | }; | |
771 | ||
772 | static struct miscdevice vhost_vsock_misc = { | |
f4660cc9 | 773 | .minor = VHOST_VSOCK_MINOR, |
433fc58e AH |
774 | .name = "vhost-vsock", |
775 | .fops = &vhost_vsock_fops, | |
776 | }; | |
777 | ||
778 | static struct virtio_transport vhost_transport = { | |
779 | .transport = { | |
780 | .get_local_cid = vhost_transport_get_local_cid, | |
781 | ||
782 | .init = virtio_transport_do_socket_init, | |
783 | .destruct = virtio_transport_destruct, | |
784 | .release = virtio_transport_release, | |
785 | .connect = virtio_transport_connect, | |
786 | .shutdown = virtio_transport_shutdown, | |
16320f36 | 787 | .cancel_pkt = vhost_transport_cancel_pkt, |
433fc58e AH |
788 | |
789 | .dgram_enqueue = virtio_transport_dgram_enqueue, | |
790 | .dgram_dequeue = virtio_transport_dgram_dequeue, | |
791 | .dgram_bind = virtio_transport_dgram_bind, | |
792 | .dgram_allow = virtio_transport_dgram_allow, | |
793 | ||
794 | .stream_enqueue = virtio_transport_stream_enqueue, | |
795 | .stream_dequeue = virtio_transport_stream_dequeue, | |
796 | .stream_has_data = virtio_transport_stream_has_data, | |
797 | .stream_has_space = virtio_transport_stream_has_space, | |
798 | .stream_rcvhiwat = virtio_transport_stream_rcvhiwat, | |
799 | .stream_is_active = virtio_transport_stream_is_active, | |
800 | .stream_allow = virtio_transport_stream_allow, | |
801 | ||
802 | .notify_poll_in = virtio_transport_notify_poll_in, | |
803 | .notify_poll_out = virtio_transport_notify_poll_out, | |
804 | .notify_recv_init = virtio_transport_notify_recv_init, | |
805 | .notify_recv_pre_block = virtio_transport_notify_recv_pre_block, | |
806 | .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue, | |
807 | .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue, | |
808 | .notify_send_init = virtio_transport_notify_send_init, | |
809 | .notify_send_pre_block = virtio_transport_notify_send_pre_block, | |
810 | .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue, | |
811 | .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue, | |
812 | ||
813 | .set_buffer_size = virtio_transport_set_buffer_size, | |
814 | .set_min_buffer_size = virtio_transport_set_min_buffer_size, | |
815 | .set_max_buffer_size = virtio_transport_set_max_buffer_size, | |
816 | .get_buffer_size = virtio_transport_get_buffer_size, | |
817 | .get_min_buffer_size = virtio_transport_get_min_buffer_size, | |
818 | .get_max_buffer_size = virtio_transport_get_max_buffer_size, | |
819 | }, | |
820 | ||
821 | .send_pkt = vhost_transport_send_pkt, | |
822 | }; | |
823 | ||
824 | static int __init vhost_vsock_init(void) | |
825 | { | |
826 | int ret; | |
827 | ||
828 | ret = vsock_core_init(&vhost_transport.transport); | |
829 | if (ret < 0) | |
830 | return ret; | |
831 | return misc_register(&vhost_vsock_misc); | |
832 | }; | |
833 | ||
834 | static void __exit vhost_vsock_exit(void) | |
835 | { | |
836 | misc_deregister(&vhost_vsock_misc); | |
837 | vsock_core_exit(); | |
838 | }; | |
839 | ||
840 | module_init(vhost_vsock_init); | |
841 | module_exit(vhost_vsock_exit); | |
842 | MODULE_LICENSE("GPL v2"); | |
843 | MODULE_AUTHOR("Asias He"); | |
844 | MODULE_DESCRIPTION("vhost transport for vsock "); | |
f4660cc9 SH |
845 | MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR); |
846 | MODULE_ALIAS("devname:vhost-vsock"); |