]>
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; | |
6dbd3e66 | 105 | size_t iov_len, payload_len; |
433fc58e AH |
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 | ||
6dbd3e66 SG |
150 | iov_len = iov_length(&vq->iov[out], in); |
151 | if (iov_len < sizeof(pkt->hdr)) { | |
152 | virtio_transport_free_pkt(pkt); | |
153 | vq_err(vq, "Buffer len [%zu] too small\n", iov_len); | |
154 | break; | |
155 | } | |
156 | ||
157 | iov_iter_init(&iov_iter, READ, &vq->iov[out], in, iov_len); | |
158 | payload_len = pkt->len - pkt->off; | |
159 | ||
160 | /* If the packet is greater than the space available in the | |
161 | * buffer, we split it using multiple buffers. | |
162 | */ | |
163 | if (payload_len > iov_len - sizeof(pkt->hdr)) | |
164 | payload_len = iov_len - sizeof(pkt->hdr); | |
165 | ||
166 | /* Set the correct length in the header */ | |
167 | pkt->hdr.len = cpu_to_le32(payload_len); | |
433fc58e AH |
168 | |
169 | nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); | |
170 | if (nbytes != sizeof(pkt->hdr)) { | |
171 | virtio_transport_free_pkt(pkt); | |
172 | vq_err(vq, "Faulted on copying pkt hdr\n"); | |
173 | break; | |
174 | } | |
175 | ||
6dbd3e66 SG |
176 | nbytes = copy_to_iter(pkt->buf + pkt->off, payload_len, |
177 | &iov_iter); | |
178 | if (nbytes != payload_len) { | |
433fc58e AH |
179 | virtio_transport_free_pkt(pkt); |
180 | vq_err(vq, "Faulted on copying pkt buf\n"); | |
181 | break; | |
182 | } | |
183 | ||
6dbd3e66 | 184 | vhost_add_used(vq, head, sizeof(pkt->hdr) + payload_len); |
433fc58e AH |
185 | added = true; |
186 | ||
82dfb540 GG |
187 | /* Deliver to monitoring devices all correctly transmitted |
188 | * packets. | |
189 | */ | |
190 | virtio_transport_deliver_tap_pkt(pkt); | |
191 | ||
6dbd3e66 SG |
192 | pkt->off += payload_len; |
193 | total_len += payload_len; | |
194 | ||
195 | /* If we didn't send all the payload we can requeue the packet | |
196 | * to send it with the next available buffer. | |
197 | */ | |
198 | if (pkt->off < pkt->len) { | |
199 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
200 | list_add(&pkt->list, &vsock->send_pkt_list); | |
201 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
202 | } else { | |
203 | if (pkt->reply) { | |
204 | int val; | |
205 | ||
206 | val = atomic_dec_return(&vsock->queued_replies); | |
207 | ||
208 | /* Do we have resources to resume tx | |
209 | * processing? | |
210 | */ | |
211 | if (val + 1 == tx_vq->num) | |
212 | restart_tx = true; | |
213 | } | |
214 | ||
215 | virtio_transport_free_pkt(pkt); | |
216 | } | |
e79b431f | 217 | } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); |
433fc58e AH |
218 | if (added) |
219 | vhost_signal(&vsock->dev, vq); | |
220 | ||
221 | out: | |
222 | mutex_unlock(&vq->mutex); | |
223 | ||
224 | if (restart_tx) | |
225 | vhost_poll_queue(&tx_vq->poll); | |
226 | } | |
227 | ||
228 | static void vhost_transport_send_pkt_work(struct vhost_work *work) | |
229 | { | |
230 | struct vhost_virtqueue *vq; | |
231 | struct vhost_vsock *vsock; | |
232 | ||
233 | vsock = container_of(work, struct vhost_vsock, send_pkt_work); | |
234 | vq = &vsock->vqs[VSOCK_VQ_RX]; | |
235 | ||
236 | vhost_transport_do_send_pkt(vsock, vq); | |
237 | } | |
238 | ||
239 | static int | |
240 | vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt) | |
241 | { | |
242 | struct vhost_vsock *vsock; | |
433fc58e AH |
243 | int len = pkt->len; |
244 | ||
834e772c SH |
245 | rcu_read_lock(); |
246 | ||
433fc58e AH |
247 | /* Find the vhost_vsock according to guest context id */ |
248 | vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid)); | |
249 | if (!vsock) { | |
834e772c | 250 | rcu_read_unlock(); |
433fc58e AH |
251 | virtio_transport_free_pkt(pkt); |
252 | return -ENODEV; | |
253 | } | |
254 | ||
433fc58e AH |
255 | if (pkt->reply) |
256 | atomic_inc(&vsock->queued_replies); | |
257 | ||
258 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
259 | list_add_tail(&pkt->list, &vsock->send_pkt_list); | |
260 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
261 | ||
262 | vhost_work_queue(&vsock->dev, &vsock->send_pkt_work); | |
834e772c SH |
263 | |
264 | rcu_read_unlock(); | |
433fc58e AH |
265 | return len; |
266 | } | |
267 | ||
16320f36 PT |
268 | static int |
269 | vhost_transport_cancel_pkt(struct vsock_sock *vsk) | |
270 | { | |
271 | struct vhost_vsock *vsock; | |
272 | struct virtio_vsock_pkt *pkt, *n; | |
273 | int cnt = 0; | |
834e772c | 274 | int ret = -ENODEV; |
16320f36 PT |
275 | LIST_HEAD(freeme); |
276 | ||
834e772c SH |
277 | rcu_read_lock(); |
278 | ||
16320f36 PT |
279 | /* Find the vhost_vsock according to guest context id */ |
280 | vsock = vhost_vsock_get(vsk->remote_addr.svm_cid); | |
281 | if (!vsock) | |
834e772c | 282 | goto out; |
16320f36 PT |
283 | |
284 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
285 | list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) { | |
286 | if (pkt->vsk != vsk) | |
287 | continue; | |
288 | list_move(&pkt->list, &freeme); | |
289 | } | |
290 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
291 | ||
292 | list_for_each_entry_safe(pkt, n, &freeme, list) { | |
293 | if (pkt->reply) | |
294 | cnt++; | |
295 | list_del(&pkt->list); | |
296 | virtio_transport_free_pkt(pkt); | |
297 | } | |
298 | ||
299 | if (cnt) { | |
300 | struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX]; | |
301 | int new_cnt; | |
302 | ||
303 | new_cnt = atomic_sub_return(cnt, &vsock->queued_replies); | |
304 | if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num) | |
305 | vhost_poll_queue(&tx_vq->poll); | |
306 | } | |
307 | ||
834e772c SH |
308 | ret = 0; |
309 | out: | |
310 | rcu_read_unlock(); | |
311 | return ret; | |
16320f36 PT |
312 | } |
313 | ||
433fc58e AH |
314 | static struct virtio_vsock_pkt * |
315 | vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq, | |
316 | unsigned int out, unsigned int in) | |
317 | { | |
318 | struct virtio_vsock_pkt *pkt; | |
319 | struct iov_iter iov_iter; | |
320 | size_t nbytes; | |
321 | size_t len; | |
322 | ||
323 | if (in != 0) { | |
324 | vq_err(vq, "Expected 0 input buffers, got %u\n", in); | |
325 | return NULL; | |
326 | } | |
327 | ||
328 | pkt = kzalloc(sizeof(*pkt), GFP_KERNEL); | |
329 | if (!pkt) | |
330 | return NULL; | |
331 | ||
332 | len = iov_length(vq->iov, out); | |
333 | iov_iter_init(&iov_iter, WRITE, vq->iov, out, len); | |
334 | ||
335 | nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); | |
336 | if (nbytes != sizeof(pkt->hdr)) { | |
337 | vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n", | |
338 | sizeof(pkt->hdr), nbytes); | |
339 | kfree(pkt); | |
340 | return NULL; | |
341 | } | |
342 | ||
343 | if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM) | |
344 | pkt->len = le32_to_cpu(pkt->hdr.len); | |
345 | ||
346 | /* No payload */ | |
347 | if (!pkt->len) | |
348 | return pkt; | |
349 | ||
350 | /* The pkt is too big */ | |
351 | if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) { | |
352 | kfree(pkt); | |
353 | return NULL; | |
354 | } | |
355 | ||
356 | pkt->buf = kmalloc(pkt->len, GFP_KERNEL); | |
357 | if (!pkt->buf) { | |
358 | kfree(pkt); | |
359 | return NULL; | |
360 | } | |
361 | ||
473c7391 SG |
362 | pkt->buf_len = pkt->len; |
363 | ||
433fc58e AH |
364 | nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter); |
365 | if (nbytes != pkt->len) { | |
366 | vq_err(vq, "Expected %u byte payload, got %zu bytes\n", | |
367 | pkt->len, nbytes); | |
368 | virtio_transport_free_pkt(pkt); | |
369 | return NULL; | |
370 | } | |
371 | ||
372 | return pkt; | |
373 | } | |
374 | ||
375 | /* Is there space left for replies to rx packets? */ | |
376 | static bool vhost_vsock_more_replies(struct vhost_vsock *vsock) | |
377 | { | |
378 | struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX]; | |
379 | int val; | |
380 | ||
381 | smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */ | |
382 | val = atomic_read(&vsock->queued_replies); | |
383 | ||
384 | return val < vq->num; | |
385 | } | |
386 | ||
4c7246dc SG |
387 | static struct virtio_transport vhost_transport = { |
388 | .transport = { | |
6a2c0962 SG |
389 | .module = THIS_MODULE, |
390 | ||
4c7246dc SG |
391 | .get_local_cid = vhost_transport_get_local_cid, |
392 | ||
393 | .init = virtio_transport_do_socket_init, | |
394 | .destruct = virtio_transport_destruct, | |
395 | .release = virtio_transport_release, | |
396 | .connect = virtio_transport_connect, | |
397 | .shutdown = virtio_transport_shutdown, | |
398 | .cancel_pkt = vhost_transport_cancel_pkt, | |
399 | ||
400 | .dgram_enqueue = virtio_transport_dgram_enqueue, | |
401 | .dgram_dequeue = virtio_transport_dgram_dequeue, | |
402 | .dgram_bind = virtio_transport_dgram_bind, | |
403 | .dgram_allow = virtio_transport_dgram_allow, | |
404 | ||
405 | .stream_enqueue = virtio_transport_stream_enqueue, | |
406 | .stream_dequeue = virtio_transport_stream_dequeue, | |
407 | .stream_has_data = virtio_transport_stream_has_data, | |
408 | .stream_has_space = virtio_transport_stream_has_space, | |
409 | .stream_rcvhiwat = virtio_transport_stream_rcvhiwat, | |
410 | .stream_is_active = virtio_transport_stream_is_active, | |
411 | .stream_allow = virtio_transport_stream_allow, | |
412 | ||
413 | .notify_poll_in = virtio_transport_notify_poll_in, | |
414 | .notify_poll_out = virtio_transport_notify_poll_out, | |
415 | .notify_recv_init = virtio_transport_notify_recv_init, | |
416 | .notify_recv_pre_block = virtio_transport_notify_recv_pre_block, | |
417 | .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue, | |
418 | .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue, | |
419 | .notify_send_init = virtio_transport_notify_send_init, | |
420 | .notify_send_pre_block = virtio_transport_notify_send_pre_block, | |
421 | .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue, | |
422 | .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue, | |
b9f2b0ff | 423 | .notify_buffer_size = virtio_transport_notify_buffer_size, |
4c7246dc | 424 | |
4c7246dc SG |
425 | }, |
426 | ||
427 | .send_pkt = vhost_transport_send_pkt, | |
428 | }; | |
429 | ||
433fc58e AH |
430 | static void vhost_vsock_handle_tx_kick(struct vhost_work *work) |
431 | { | |
432 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
433 | poll.work); | |
434 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
435 | dev); | |
436 | struct virtio_vsock_pkt *pkt; | |
e79b431f | 437 | int head, pkts = 0, total_len = 0; |
433fc58e AH |
438 | unsigned int out, in; |
439 | bool added = false; | |
440 | ||
441 | mutex_lock(&vq->mutex); | |
442 | ||
443 | if (!vq->private_data) | |
444 | goto out; | |
445 | ||
446 | vhost_disable_notify(&vsock->dev, vq); | |
e79b431f | 447 | do { |
3fda5d6e SH |
448 | u32 len; |
449 | ||
433fc58e AH |
450 | if (!vhost_vsock_more_replies(vsock)) { |
451 | /* Stop tx until the device processes already | |
452 | * pending replies. Leave tx virtqueue | |
453 | * callbacks disabled. | |
454 | */ | |
455 | goto no_more_replies; | |
456 | } | |
457 | ||
458 | head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), | |
459 | &out, &in, NULL, NULL); | |
460 | if (head < 0) | |
461 | break; | |
462 | ||
463 | if (head == vq->num) { | |
464 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { | |
465 | vhost_disable_notify(&vsock->dev, vq); | |
466 | continue; | |
467 | } | |
468 | break; | |
469 | } | |
470 | ||
471 | pkt = vhost_vsock_alloc_pkt(vq, out, in); | |
472 | if (!pkt) { | |
473 | vq_err(vq, "Faulted on pkt\n"); | |
474 | continue; | |
475 | } | |
476 | ||
3fda5d6e SH |
477 | len = pkt->len; |
478 | ||
82dfb540 GG |
479 | /* Deliver to monitoring devices all received packets */ |
480 | virtio_transport_deliver_tap_pkt(pkt); | |
481 | ||
433fc58e | 482 | /* Only accept correctly addressed packets */ |
8a3cc29c SG |
483 | if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid && |
484 | le64_to_cpu(pkt->hdr.dst_cid) == | |
485 | vhost_transport_get_local_cid()) | |
4c7246dc | 486 | virtio_transport_recv_pkt(&vhost_transport, pkt); |
433fc58e AH |
487 | else |
488 | virtio_transport_free_pkt(pkt); | |
489 | ||
e79b431f JW |
490 | len += sizeof(pkt->hdr); |
491 | vhost_add_used(vq, head, len); | |
492 | total_len += len; | |
433fc58e | 493 | added = true; |
e79b431f | 494 | } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); |
433fc58e AH |
495 | |
496 | no_more_replies: | |
497 | if (added) | |
498 | vhost_signal(&vsock->dev, vq); | |
499 | ||
500 | out: | |
501 | mutex_unlock(&vq->mutex); | |
502 | } | |
503 | ||
504 | static void vhost_vsock_handle_rx_kick(struct vhost_work *work) | |
505 | { | |
506 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
507 | poll.work); | |
508 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
509 | dev); | |
510 | ||
511 | vhost_transport_do_send_pkt(vsock, vq); | |
512 | } | |
513 | ||
514 | static int vhost_vsock_start(struct vhost_vsock *vsock) | |
515 | { | |
0516ffd8 | 516 | struct vhost_virtqueue *vq; |
433fc58e AH |
517 | size_t i; |
518 | int ret; | |
519 | ||
520 | mutex_lock(&vsock->dev.mutex); | |
521 | ||
522 | ret = vhost_dev_check_owner(&vsock->dev); | |
523 | if (ret) | |
524 | goto err; | |
525 | ||
526 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
0516ffd8 | 527 | vq = &vsock->vqs[i]; |
433fc58e AH |
528 | |
529 | mutex_lock(&vq->mutex); | |
530 | ||
531 | if (!vhost_vq_access_ok(vq)) { | |
532 | ret = -EFAULT; | |
433fc58e AH |
533 | goto err_vq; |
534 | } | |
535 | ||
536 | if (!vq->private_data) { | |
537 | vq->private_data = vsock; | |
0516ffd8 SH |
538 | ret = vhost_vq_init_access(vq); |
539 | if (ret) | |
540 | goto err_vq; | |
433fc58e AH |
541 | } |
542 | ||
543 | mutex_unlock(&vq->mutex); | |
544 | } | |
545 | ||
546 | mutex_unlock(&vsock->dev.mutex); | |
547 | return 0; | |
548 | ||
549 | err_vq: | |
0516ffd8 SH |
550 | vq->private_data = NULL; |
551 | mutex_unlock(&vq->mutex); | |
552 | ||
433fc58e | 553 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
0516ffd8 | 554 | vq = &vsock->vqs[i]; |
433fc58e AH |
555 | |
556 | mutex_lock(&vq->mutex); | |
557 | vq->private_data = NULL; | |
558 | mutex_unlock(&vq->mutex); | |
559 | } | |
560 | err: | |
561 | mutex_unlock(&vsock->dev.mutex); | |
562 | return ret; | |
563 | } | |
564 | ||
565 | static int vhost_vsock_stop(struct vhost_vsock *vsock) | |
566 | { | |
567 | size_t i; | |
568 | int ret; | |
569 | ||
570 | mutex_lock(&vsock->dev.mutex); | |
571 | ||
572 | ret = vhost_dev_check_owner(&vsock->dev); | |
573 | if (ret) | |
574 | goto err; | |
575 | ||
576 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
577 | struct vhost_virtqueue *vq = &vsock->vqs[i]; | |
578 | ||
579 | mutex_lock(&vq->mutex); | |
580 | vq->private_data = NULL; | |
581 | mutex_unlock(&vq->mutex); | |
582 | } | |
583 | ||
584 | err: | |
585 | mutex_unlock(&vsock->dev.mutex); | |
586 | return ret; | |
587 | } | |
588 | ||
589 | static void vhost_vsock_free(struct vhost_vsock *vsock) | |
590 | { | |
b226acab | 591 | kvfree(vsock); |
433fc58e AH |
592 | } |
593 | ||
594 | static int vhost_vsock_dev_open(struct inode *inode, struct file *file) | |
595 | { | |
596 | struct vhost_virtqueue **vqs; | |
597 | struct vhost_vsock *vsock; | |
598 | int ret; | |
599 | ||
600 | /* This struct is large and allocation could fail, fall back to vmalloc | |
601 | * if there is no other way. | |
602 | */ | |
dcda9b04 | 603 | vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
6c5ab651 MH |
604 | if (!vsock) |
605 | return -ENOMEM; | |
433fc58e AH |
606 | |
607 | vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL); | |
608 | if (!vqs) { | |
609 | ret = -ENOMEM; | |
610 | goto out; | |
611 | } | |
612 | ||
a72b69dc SH |
613 | vsock->guest_cid = 0; /* no CID assigned yet */ |
614 | ||
433fc58e AH |
615 | atomic_set(&vsock->queued_replies, 0); |
616 | ||
617 | vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX]; | |
618 | vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX]; | |
619 | vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick; | |
620 | vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick; | |
621 | ||
e82b9b07 JW |
622 | vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs), |
623 | UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT, | |
792a4f2e | 624 | VHOST_VSOCK_WEIGHT, NULL); |
433fc58e AH |
625 | |
626 | file->private_data = vsock; | |
627 | spin_lock_init(&vsock->send_pkt_list_lock); | |
628 | INIT_LIST_HEAD(&vsock->send_pkt_list); | |
629 | vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work); | |
433fc58e AH |
630 | return 0; |
631 | ||
632 | out: | |
633 | vhost_vsock_free(vsock); | |
634 | return ret; | |
635 | } | |
636 | ||
637 | static void vhost_vsock_flush(struct vhost_vsock *vsock) | |
638 | { | |
639 | int i; | |
640 | ||
641 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) | |
642 | if (vsock->vqs[i].handle_kick) | |
643 | vhost_poll_flush(&vsock->vqs[i].poll); | |
644 | vhost_work_flush(&vsock->dev, &vsock->send_pkt_work); | |
645 | } | |
646 | ||
647 | static void vhost_vsock_reset_orphans(struct sock *sk) | |
648 | { | |
649 | struct vsock_sock *vsk = vsock_sk(sk); | |
650 | ||
651 | /* vmci_transport.c doesn't take sk_lock here either. At least we're | |
652 | * under vsock_table_lock so the sock cannot disappear while we're | |
653 | * executing. | |
654 | */ | |
655 | ||
c38f57da SH |
656 | /* If the peer is still valid, no need to reset connection */ |
657 | if (vhost_vsock_get(vsk->remote_addr.svm_cid)) | |
658 | return; | |
659 | ||
660 | /* If the close timeout is pending, let it expire. This avoids races | |
661 | * with the timeout callback. | |
662 | */ | |
663 | if (vsk->close_work_scheduled) | |
664 | return; | |
665 | ||
666 | sock_set_flag(sk, SOCK_DONE); | |
667 | vsk->peer_shutdown = SHUTDOWN_MASK; | |
668 | sk->sk_state = SS_UNCONNECTED; | |
669 | sk->sk_err = ECONNRESET; | |
670 | sk->sk_error_report(sk); | |
433fc58e AH |
671 | } |
672 | ||
673 | static int vhost_vsock_dev_release(struct inode *inode, struct file *file) | |
674 | { | |
675 | struct vhost_vsock *vsock = file->private_data; | |
676 | ||
6db3d8dc | 677 | mutex_lock(&vhost_vsock_mutex); |
834e772c SH |
678 | if (vsock->guest_cid) |
679 | hash_del_rcu(&vsock->hash); | |
6db3d8dc | 680 | mutex_unlock(&vhost_vsock_mutex); |
433fc58e | 681 | |
834e772c SH |
682 | /* Wait for other CPUs to finish using vsock */ |
683 | synchronize_rcu(); | |
684 | ||
433fc58e AH |
685 | /* Iterating over all connections for all CIDs to find orphans is |
686 | * inefficient. Room for improvement here. */ | |
687 | vsock_for_each_connected_socket(vhost_vsock_reset_orphans); | |
688 | ||
689 | vhost_vsock_stop(vsock); | |
690 | vhost_vsock_flush(vsock); | |
691 | vhost_dev_stop(&vsock->dev); | |
692 | ||
693 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
694 | while (!list_empty(&vsock->send_pkt_list)) { | |
695 | struct virtio_vsock_pkt *pkt; | |
696 | ||
697 | pkt = list_first_entry(&vsock->send_pkt_list, | |
698 | struct virtio_vsock_pkt, list); | |
699 | list_del_init(&pkt->list); | |
700 | virtio_transport_free_pkt(pkt); | |
701 | } | |
702 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
703 | ||
f6f93f75 | 704 | vhost_dev_cleanup(&vsock->dev); |
433fc58e AH |
705 | kfree(vsock->dev.vqs); |
706 | vhost_vsock_free(vsock); | |
707 | return 0; | |
708 | } | |
709 | ||
710 | static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid) | |
711 | { | |
712 | struct vhost_vsock *other; | |
713 | ||
714 | /* Refuse reserved CIDs */ | |
715 | if (guest_cid <= VMADDR_CID_HOST || | |
716 | guest_cid == U32_MAX) | |
717 | return -EINVAL; | |
718 | ||
719 | /* 64-bit CIDs are not yet supported */ | |
720 | if (guest_cid > U32_MAX) | |
721 | return -EINVAL; | |
722 | ||
ed8640a9 SG |
723 | /* Refuse if CID is assigned to the guest->host transport (i.e. nested |
724 | * VM), to make the loopback work. | |
725 | */ | |
726 | if (vsock_find_cid(guest_cid)) | |
727 | return -EADDRINUSE; | |
728 | ||
433fc58e | 729 | /* Refuse if CID is already in use */ |
6db3d8dc | 730 | mutex_lock(&vhost_vsock_mutex); |
834e772c | 731 | other = vhost_vsock_get(guest_cid); |
6c083c2b | 732 | if (other && other != vsock) { |
6db3d8dc | 733 | mutex_unlock(&vhost_vsock_mutex); |
6c083c2b G |
734 | return -EADDRINUSE; |
735 | } | |
834e772c SH |
736 | |
737 | if (vsock->guest_cid) | |
738 | hash_del_rcu(&vsock->hash); | |
739 | ||
433fc58e | 740 | vsock->guest_cid = guest_cid; |
7fbe078c | 741 | hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid); |
6db3d8dc | 742 | mutex_unlock(&vhost_vsock_mutex); |
433fc58e AH |
743 | |
744 | return 0; | |
745 | } | |
746 | ||
747 | static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features) | |
748 | { | |
749 | struct vhost_virtqueue *vq; | |
750 | int i; | |
751 | ||
752 | if (features & ~VHOST_VSOCK_FEATURES) | |
753 | return -EOPNOTSUPP; | |
754 | ||
755 | mutex_lock(&vsock->dev.mutex); | |
756 | if ((features & (1 << VHOST_F_LOG_ALL)) && | |
757 | !vhost_log_access_ok(&vsock->dev)) { | |
758 | mutex_unlock(&vsock->dev.mutex); | |
759 | return -EFAULT; | |
760 | } | |
761 | ||
762 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
763 | vq = &vsock->vqs[i]; | |
764 | mutex_lock(&vq->mutex); | |
765 | vq->acked_features = features; | |
766 | mutex_unlock(&vq->mutex); | |
767 | } | |
768 | mutex_unlock(&vsock->dev.mutex); | |
769 | return 0; | |
770 | } | |
771 | ||
772 | static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl, | |
773 | unsigned long arg) | |
774 | { | |
775 | struct vhost_vsock *vsock = f->private_data; | |
776 | void __user *argp = (void __user *)arg; | |
777 | u64 guest_cid; | |
778 | u64 features; | |
779 | int start; | |
780 | int r; | |
781 | ||
782 | switch (ioctl) { | |
783 | case VHOST_VSOCK_SET_GUEST_CID: | |
784 | if (copy_from_user(&guest_cid, argp, sizeof(guest_cid))) | |
785 | return -EFAULT; | |
786 | return vhost_vsock_set_cid(vsock, guest_cid); | |
787 | case VHOST_VSOCK_SET_RUNNING: | |
788 | if (copy_from_user(&start, argp, sizeof(start))) | |
789 | return -EFAULT; | |
790 | if (start) | |
791 | return vhost_vsock_start(vsock); | |
792 | else | |
793 | return vhost_vsock_stop(vsock); | |
794 | case VHOST_GET_FEATURES: | |
795 | features = VHOST_VSOCK_FEATURES; | |
796 | if (copy_to_user(argp, &features, sizeof(features))) | |
797 | return -EFAULT; | |
798 | return 0; | |
799 | case VHOST_SET_FEATURES: | |
800 | if (copy_from_user(&features, argp, sizeof(features))) | |
801 | return -EFAULT; | |
802 | return vhost_vsock_set_features(vsock, features); | |
803 | default: | |
804 | mutex_lock(&vsock->dev.mutex); | |
805 | r = vhost_dev_ioctl(&vsock->dev, ioctl, argp); | |
806 | if (r == -ENOIOCTLCMD) | |
807 | r = vhost_vring_ioctl(&vsock->dev, ioctl, argp); | |
808 | else | |
809 | vhost_vsock_flush(vsock); | |
810 | mutex_unlock(&vsock->dev.mutex); | |
811 | return r; | |
812 | } | |
813 | } | |
814 | ||
815 | static const struct file_operations vhost_vsock_fops = { | |
816 | .owner = THIS_MODULE, | |
817 | .open = vhost_vsock_dev_open, | |
818 | .release = vhost_vsock_dev_release, | |
819 | .llseek = noop_llseek, | |
820 | .unlocked_ioctl = vhost_vsock_dev_ioctl, | |
407e9ef7 | 821 | .compat_ioctl = compat_ptr_ioctl, |
433fc58e AH |
822 | }; |
823 | ||
824 | static struct miscdevice vhost_vsock_misc = { | |
f4660cc9 | 825 | .minor = VHOST_VSOCK_MINOR, |
433fc58e AH |
826 | .name = "vhost-vsock", |
827 | .fops = &vhost_vsock_fops, | |
828 | }; | |
829 | ||
433fc58e AH |
830 | static int __init vhost_vsock_init(void) |
831 | { | |
832 | int ret; | |
833 | ||
c0cfa2d8 SG |
834 | ret = vsock_core_register(&vhost_transport.transport, |
835 | VSOCK_TRANSPORT_F_H2G); | |
433fc58e AH |
836 | if (ret < 0) |
837 | return ret; | |
838 | return misc_register(&vhost_vsock_misc); | |
839 | }; | |
840 | ||
841 | static void __exit vhost_vsock_exit(void) | |
842 | { | |
843 | misc_deregister(&vhost_vsock_misc); | |
c0cfa2d8 | 844 | vsock_core_unregister(&vhost_transport.transport); |
433fc58e AH |
845 | }; |
846 | ||
847 | module_init(vhost_vsock_init); | |
848 | module_exit(vhost_vsock_exit); | |
849 | MODULE_LICENSE("GPL v2"); | |
850 | MODULE_AUTHOR("Asias He"); | |
851 | MODULE_DESCRIPTION("vhost transport for vsock "); | |
f4660cc9 SH |
852 | MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR); |
853 | MODULE_ALIAS("devname:vhost-vsock"); |