1 | // SPDX-License-Identifier: GPL-2.0-only |
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> |
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> |
17 | #include <linux/hashtable.h> |
18 | |
19 | #include <net/af_vsock.h> |
20 | #include "vhost.h" |
21 | |
22 | #define VHOST_VSOCK_DEFAULT_HOST_CID 2 |
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 |
31 | |
32 | enum { |
33 | VHOST_VSOCK_FEATURES = VHOST_FEATURES | |
34 | (1ULL << VIRTIO_F_ACCESS_PLATFORM) | |
35 | (1ULL << VIRTIO_VSOCK_F_SEQPACKET) |
36 | }; |
37 | |
38 | enum { |
39 | VHOST_VSOCK_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2) |
40 | }; |
41 | |
42 | /* Used to track all the vhost_vsock instances on the system. */ |
43 | static DEFINE_MUTEX(vhost_vsock_mutex); |
44 | static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8); |
45 | |
46 | struct vhost_vsock { |
47 | struct vhost_dev dev; |
48 | struct vhost_virtqueue vqs[2]; |
49 | |
50 | /* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */ |
51 | struct hlist_node hash; |
52 | |
53 | struct vhost_work send_pkt_work; |
54 | struct sk_buff_head send_pkt_queue; /* host->guest pending packets */ |
55 | |
56 | atomic_t queued_replies; |
57 | |
58 | u32 guest_cid; |
59 | bool seqpacket_allow; |
60 | }; |
61 | |
62 | static u32 vhost_transport_get_local_cid(void) |
63 | { |
64 | return VHOST_VSOCK_DEFAULT_HOST_CID; |
65 | } |
66 | |
67 | /* Callers that dereference the return value must hold vhost_vsock_mutex or the |
68 | * RCU read lock. |
69 | */ |
70 | static struct vhost_vsock *vhost_vsock_get(u32 guest_cid) |
71 | { |
72 | struct vhost_vsock *vsock; |
73 | |
74 | hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) { |
75 | u32 other_cid = vsock->guest_cid; |
76 | |
77 | /* Skip instances that have no CID yet */ |
78 | if (other_cid == 0) |
79 | continue; |
80 | |
81 | if (other_cid == guest_cid) |
82 | return vsock; |
83 | |
84 | } |
85 | |
86 | return NULL; |
87 | } |
88 | |
89 | static void |
90 | vhost_transport_do_send_pkt(struct vhost_vsock *vsock, |
91 | struct vhost_virtqueue *vq) |
92 | { |
93 | struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX]; |
94 | int pkts = 0, total_len = 0; |
95 | bool added = false; |
96 | bool restart_tx = false; |
97 | |
98 | mutex_lock(&vq->mutex); |
99 | |
100 | if (!vhost_vq_get_backend(vq)) |
101 | goto out; |
102 | |
103 | if (!vq_meta_prefetch(vq)) |
104 | goto out; |
105 | |
106 | /* Avoid further vmexits, we're already processing the virtqueue */ |
107 | vhost_disable_notify(&vsock->dev, vq); |
108 | |
109 | do { |
110 | struct virtio_vsock_hdr *hdr; |
111 | size_t iov_len, payload_len; |
112 | struct iov_iter iov_iter; |
113 | u32 flags_to_restore = 0; |
114 | struct sk_buff *skb; |
115 | unsigned out, in; |
116 | size_t nbytes; |
117 | u32 offset; |
118 | int head; |
119 | |
120 | skb = virtio_vsock_skb_dequeue(list: &vsock->send_pkt_queue); |
121 | |
122 | if (!skb) { |
123 | vhost_enable_notify(&vsock->dev, vq); |
124 | break; |
125 | } |
126 | |
127 | head = vhost_get_vq_desc(vq, iov: vq->iov, ARRAY_SIZE(vq->iov), |
128 | out_num: &out, in_num: &in, NULL, NULL); |
129 | if (head < 0) { |
130 | virtio_vsock_skb_queue_head(list: &vsock->send_pkt_queue, skb); |
131 | break; |
132 | } |
133 | |
134 | if (head == vq->num) { |
135 | virtio_vsock_skb_queue_head(list: &vsock->send_pkt_queue, skb); |
136 | /* We cannot finish yet if more buffers snuck in while |
137 | * re-enabling notify. |
138 | */ |
139 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { |
140 | vhost_disable_notify(&vsock->dev, vq); |
141 | continue; |
142 | } |
143 | break; |
144 | } |
145 | |
146 | if (out) { |
147 | kfree_skb(skb); |
148 | vq_err(vq, "Expected 0 output buffers, got %u\n" , out); |
149 | break; |
150 | } |
151 | |
152 | iov_len = iov_length(iov: &vq->iov[out], nr_segs: in); |
153 | if (iov_len < sizeof(*hdr)) { |
154 | kfree_skb(skb); |
155 | vq_err(vq, "Buffer len [%zu] too small\n" , iov_len); |
156 | break; |
157 | } |
158 | |
159 | iov_iter_init(i: &iov_iter, ITER_DEST, iov: &vq->iov[out], nr_segs: in, count: iov_len); |
160 | offset = VIRTIO_VSOCK_SKB_CB(skb)->offset; |
161 | payload_len = skb->len - offset; |
162 | hdr = virtio_vsock_hdr(skb); |
163 | |
164 | /* If the packet is greater than the space available in the |
165 | * buffer, we split it using multiple buffers. |
166 | */ |
167 | if (payload_len > iov_len - sizeof(*hdr)) { |
168 | payload_len = iov_len - sizeof(*hdr); |
169 | |
170 | /* As we are copying pieces of large packet's buffer to |
171 | * small rx buffers, headers of packets in rx queue are |
172 | * created dynamically and are initialized with header |
173 | * of current packet(except length). But in case of |
174 | * SOCK_SEQPACKET, we also must clear message delimeter |
175 | * bit (VIRTIO_VSOCK_SEQ_EOM) and MSG_EOR bit |
176 | * (VIRTIO_VSOCK_SEQ_EOR) if set. Otherwise, |
177 | * there will be sequence of packets with these |
178 | * bits set. After initialized header will be copied to |
179 | * rx buffer, these required bits will be restored. |
180 | */ |
181 | if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { |
182 | hdr->flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM); |
183 | flags_to_restore |= VIRTIO_VSOCK_SEQ_EOM; |
184 | |
185 | if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) { |
186 | hdr->flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR); |
187 | flags_to_restore |= VIRTIO_VSOCK_SEQ_EOR; |
188 | } |
189 | } |
190 | } |
191 | |
192 | /* Set the correct length in the header */ |
193 | hdr->len = cpu_to_le32(payload_len); |
194 | |
195 | nbytes = copy_to_iter(addr: hdr, bytes: sizeof(*hdr), i: &iov_iter); |
196 | if (nbytes != sizeof(*hdr)) { |
197 | kfree_skb(skb); |
198 | vq_err(vq, "Faulted on copying pkt hdr\n" ); |
199 | break; |
200 | } |
201 | |
202 | if (skb_copy_datagram_iter(from: skb, |
203 | offset, |
204 | to: &iov_iter, |
205 | size: payload_len)) { |
206 | kfree_skb(skb); |
207 | vq_err(vq, "Faulted on copying pkt buf\n" ); |
208 | break; |
209 | } |
210 | |
211 | /* Deliver to monitoring devices all packets that we |
212 | * will transmit. |
213 | */ |
214 | virtio_transport_deliver_tap_pkt(skb); |
215 | |
216 | vhost_add_used(vq, head, len: sizeof(*hdr) + payload_len); |
217 | added = true; |
218 | |
219 | VIRTIO_VSOCK_SKB_CB(skb)->offset += payload_len; |
220 | total_len += payload_len; |
221 | |
222 | /* If we didn't send all the payload we can requeue the packet |
223 | * to send it with the next available buffer. |
224 | */ |
225 | if (VIRTIO_VSOCK_SKB_CB(skb)->offset < skb->len) { |
226 | hdr->flags |= cpu_to_le32(flags_to_restore); |
227 | |
228 | /* We are queueing the same skb to handle |
229 | * the remaining bytes, and we want to deliver it |
230 | * to monitoring devices in the next iteration. |
231 | */ |
232 | virtio_vsock_skb_clear_tap_delivered(skb); |
233 | virtio_vsock_skb_queue_head(list: &vsock->send_pkt_queue, skb); |
234 | } else { |
235 | if (virtio_vsock_skb_reply(skb)) { |
236 | int val; |
237 | |
238 | val = atomic_dec_return(v: &vsock->queued_replies); |
239 | |
240 | /* Do we have resources to resume tx |
241 | * processing? |
242 | */ |
243 | if (val + 1 == tx_vq->num) |
244 | restart_tx = true; |
245 | } |
246 | |
247 | consume_skb(skb); |
248 | } |
249 | } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); |
250 | if (added) |
251 | vhost_signal(&vsock->dev, vq); |
252 | |
253 | out: |
254 | mutex_unlock(lock: &vq->mutex); |
255 | |
256 | if (restart_tx) |
257 | vhost_poll_queue(poll: &tx_vq->poll); |
258 | } |
259 | |
260 | static void vhost_transport_send_pkt_work(struct vhost_work *work) |
261 | { |
262 | struct vhost_virtqueue *vq; |
263 | struct vhost_vsock *vsock; |
264 | |
265 | vsock = container_of(work, struct vhost_vsock, send_pkt_work); |
266 | vq = &vsock->vqs[VSOCK_VQ_RX]; |
267 | |
268 | vhost_transport_do_send_pkt(vsock, vq); |
269 | } |
270 | |
271 | static int |
272 | vhost_transport_send_pkt(struct sk_buff *skb) |
273 | { |
274 | struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); |
275 | struct vhost_vsock *vsock; |
276 | int len = skb->len; |
277 | |
278 | rcu_read_lock(); |
279 | |
280 | /* Find the vhost_vsock according to guest context id */ |
281 | vsock = vhost_vsock_get(le64_to_cpu(hdr->dst_cid)); |
282 | if (!vsock) { |
283 | rcu_read_unlock(); |
284 | kfree_skb(skb); |
285 | return -ENODEV; |
286 | } |
287 | |
288 | if (virtio_vsock_skb_reply(skb)) |
289 | atomic_inc(v: &vsock->queued_replies); |
290 | |
291 | virtio_vsock_skb_queue_tail(list: &vsock->send_pkt_queue, skb); |
292 | vhost_vq_work_queue(vq: &vsock->vqs[VSOCK_VQ_RX], work: &vsock->send_pkt_work); |
293 | |
294 | rcu_read_unlock(); |
295 | return len; |
296 | } |
297 | |
298 | static int |
299 | vhost_transport_cancel_pkt(struct vsock_sock *vsk) |
300 | { |
301 | struct vhost_vsock *vsock; |
302 | int cnt = 0; |
303 | int ret = -ENODEV; |
304 | |
305 | rcu_read_lock(); |
306 | |
307 | /* Find the vhost_vsock according to guest context id */ |
308 | vsock = vhost_vsock_get(guest_cid: vsk->remote_addr.svm_cid); |
309 | if (!vsock) |
310 | goto out; |
311 | |
312 | cnt = virtio_transport_purge_skbs(vsk, list: &vsock->send_pkt_queue); |
313 | |
314 | if (cnt) { |
315 | struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX]; |
316 | int new_cnt; |
317 | |
318 | new_cnt = atomic_sub_return(i: cnt, v: &vsock->queued_replies); |
319 | if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num) |
320 | vhost_poll_queue(poll: &tx_vq->poll); |
321 | } |
322 | |
323 | ret = 0; |
324 | out: |
325 | rcu_read_unlock(); |
326 | return ret; |
327 | } |
328 | |
329 | static struct sk_buff * |
330 | vhost_vsock_alloc_skb(struct vhost_virtqueue *vq, |
331 | unsigned int out, unsigned int in) |
332 | { |
333 | struct virtio_vsock_hdr *hdr; |
334 | struct iov_iter iov_iter; |
335 | struct sk_buff *skb; |
336 | size_t payload_len; |
337 | size_t nbytes; |
338 | size_t len; |
339 | |
340 | if (in != 0) { |
341 | vq_err(vq, "Expected 0 input buffers, got %u\n" , in); |
342 | return NULL; |
343 | } |
344 | |
345 | len = iov_length(iov: vq->iov, nr_segs: out); |
346 | |
347 | /* len contains both payload and hdr */ |
348 | skb = virtio_vsock_alloc_skb(size: len, GFP_KERNEL); |
349 | if (!skb) |
350 | return NULL; |
351 | |
352 | iov_iter_init(i: &iov_iter, ITER_SOURCE, iov: vq->iov, nr_segs: out, count: len); |
353 | |
354 | hdr = virtio_vsock_hdr(skb); |
355 | nbytes = copy_from_iter(addr: hdr, bytes: sizeof(*hdr), i: &iov_iter); |
356 | if (nbytes != sizeof(*hdr)) { |
357 | vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n" , |
358 | sizeof(*hdr), nbytes); |
359 | kfree_skb(skb); |
360 | return NULL; |
361 | } |
362 | |
363 | payload_len = le32_to_cpu(hdr->len); |
364 | |
365 | /* No payload */ |
366 | if (!payload_len) |
367 | return skb; |
368 | |
369 | /* The pkt is too big or the length in the header is invalid */ |
370 | if (payload_len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || |
371 | payload_len + sizeof(*hdr) > len) { |
372 | kfree_skb(skb); |
373 | return NULL; |
374 | } |
375 | |
376 | virtio_vsock_skb_rx_put(skb); |
377 | |
378 | nbytes = copy_from_iter(addr: skb->data, bytes: payload_len, i: &iov_iter); |
379 | if (nbytes != payload_len) { |
380 | vq_err(vq, "Expected %zu byte payload, got %zu bytes\n" , |
381 | payload_len, nbytes); |
382 | kfree_skb(skb); |
383 | return NULL; |
384 | } |
385 | |
386 | return skb; |
387 | } |
388 | |
389 | /* Is there space left for replies to rx packets? */ |
390 | static bool vhost_vsock_more_replies(struct vhost_vsock *vsock) |
391 | { |
392 | struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX]; |
393 | int val; |
394 | |
395 | smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */ |
396 | val = atomic_read(v: &vsock->queued_replies); |
397 | |
398 | return val < vq->num; |
399 | } |
400 | |
401 | static bool vhost_transport_msgzerocopy_allow(void) |
402 | { |
403 | return true; |
404 | } |
405 | |
406 | static bool vhost_transport_seqpacket_allow(u32 remote_cid); |
407 | |
408 | static struct virtio_transport vhost_transport = { |
409 | .transport = { |
410 | .module = THIS_MODULE, |
411 | |
412 | .get_local_cid = vhost_transport_get_local_cid, |
413 | |
414 | .init = virtio_transport_do_socket_init, |
415 | .destruct = virtio_transport_destruct, |
416 | .release = virtio_transport_release, |
417 | .connect = virtio_transport_connect, |
418 | .shutdown = virtio_transport_shutdown, |
419 | .cancel_pkt = vhost_transport_cancel_pkt, |
420 | |
421 | .dgram_enqueue = virtio_transport_dgram_enqueue, |
422 | .dgram_dequeue = virtio_transport_dgram_dequeue, |
423 | .dgram_bind = virtio_transport_dgram_bind, |
424 | .dgram_allow = virtio_transport_dgram_allow, |
425 | |
426 | .stream_enqueue = virtio_transport_stream_enqueue, |
427 | .stream_dequeue = virtio_transport_stream_dequeue, |
428 | .stream_has_data = virtio_transport_stream_has_data, |
429 | .stream_has_space = virtio_transport_stream_has_space, |
430 | .stream_rcvhiwat = virtio_transport_stream_rcvhiwat, |
431 | .stream_is_active = virtio_transport_stream_is_active, |
432 | .stream_allow = virtio_transport_stream_allow, |
433 | |
434 | .seqpacket_dequeue = virtio_transport_seqpacket_dequeue, |
435 | .seqpacket_enqueue = virtio_transport_seqpacket_enqueue, |
436 | .seqpacket_allow = vhost_transport_seqpacket_allow, |
437 | .seqpacket_has_data = virtio_transport_seqpacket_has_data, |
438 | |
439 | .msgzerocopy_allow = vhost_transport_msgzerocopy_allow, |
440 | |
441 | .notify_poll_in = virtio_transport_notify_poll_in, |
442 | .notify_poll_out = virtio_transport_notify_poll_out, |
443 | .notify_recv_init = virtio_transport_notify_recv_init, |
444 | .notify_recv_pre_block = virtio_transport_notify_recv_pre_block, |
445 | .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue, |
446 | .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue, |
447 | .notify_send_init = virtio_transport_notify_send_init, |
448 | .notify_send_pre_block = virtio_transport_notify_send_pre_block, |
449 | .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue, |
450 | .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue, |
451 | .notify_buffer_size = virtio_transport_notify_buffer_size, |
452 | |
453 | .read_skb = virtio_transport_read_skb, |
454 | }, |
455 | |
456 | .send_pkt = vhost_transport_send_pkt, |
457 | }; |
458 | |
459 | static bool vhost_transport_seqpacket_allow(u32 remote_cid) |
460 | { |
461 | struct vhost_vsock *vsock; |
462 | bool seqpacket_allow = false; |
463 | |
464 | rcu_read_lock(); |
465 | vsock = vhost_vsock_get(guest_cid: remote_cid); |
466 | |
467 | if (vsock) |
468 | seqpacket_allow = vsock->seqpacket_allow; |
469 | |
470 | rcu_read_unlock(); |
471 | |
472 | return seqpacket_allow; |
473 | } |
474 | |
475 | static void vhost_vsock_handle_tx_kick(struct vhost_work *work) |
476 | { |
477 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
478 | poll.work); |
479 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, |
480 | dev); |
481 | int head, pkts = 0, total_len = 0; |
482 | unsigned int out, in; |
483 | struct sk_buff *skb; |
484 | bool added = false; |
485 | |
486 | mutex_lock(&vq->mutex); |
487 | |
488 | if (!vhost_vq_get_backend(vq)) |
489 | goto out; |
490 | |
491 | if (!vq_meta_prefetch(vq)) |
492 | goto out; |
493 | |
494 | vhost_disable_notify(&vsock->dev, vq); |
495 | do { |
496 | struct virtio_vsock_hdr *hdr; |
497 | |
498 | if (!vhost_vsock_more_replies(vsock)) { |
499 | /* Stop tx until the device processes already |
500 | * pending replies. Leave tx virtqueue |
501 | * callbacks disabled. |
502 | */ |
503 | goto no_more_replies; |
504 | } |
505 | |
506 | head = vhost_get_vq_desc(vq, iov: vq->iov, ARRAY_SIZE(vq->iov), |
507 | out_num: &out, in_num: &in, NULL, NULL); |
508 | if (head < 0) |
509 | break; |
510 | |
511 | if (head == vq->num) { |
512 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { |
513 | vhost_disable_notify(&vsock->dev, vq); |
514 | continue; |
515 | } |
516 | break; |
517 | } |
518 | |
519 | skb = vhost_vsock_alloc_skb(vq, out, in); |
520 | if (!skb) { |
521 | vq_err(vq, "Faulted on pkt\n" ); |
522 | continue; |
523 | } |
524 | |
525 | total_len += sizeof(*hdr) + skb->len; |
526 | |
527 | /* Deliver to monitoring devices all received packets */ |
528 | virtio_transport_deliver_tap_pkt(skb); |
529 | |
530 | hdr = virtio_vsock_hdr(skb); |
531 | |
532 | /* Only accept correctly addressed packets */ |
533 | if (le64_to_cpu(hdr->src_cid) == vsock->guest_cid && |
534 | le64_to_cpu(hdr->dst_cid) == |
535 | vhost_transport_get_local_cid()) |
536 | virtio_transport_recv_pkt(t: &vhost_transport, skb); |
537 | else |
538 | kfree_skb(skb); |
539 | |
540 | vhost_add_used(vq, head, len: 0); |
541 | added = true; |
542 | } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); |
543 | |
544 | no_more_replies: |
545 | if (added) |
546 | vhost_signal(&vsock->dev, vq); |
547 | |
548 | out: |
549 | mutex_unlock(lock: &vq->mutex); |
550 | } |
551 | |
552 | static void vhost_vsock_handle_rx_kick(struct vhost_work *work) |
553 | { |
554 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
555 | poll.work); |
556 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, |
557 | dev); |
558 | |
559 | vhost_transport_do_send_pkt(vsock, vq); |
560 | } |
561 | |
562 | static int vhost_vsock_start(struct vhost_vsock *vsock) |
563 | { |
564 | struct vhost_virtqueue *vq; |
565 | size_t i; |
566 | int ret; |
567 | |
568 | mutex_lock(&vsock->dev.mutex); |
569 | |
570 | ret = vhost_dev_check_owner(&vsock->dev); |
571 | if (ret) |
572 | goto err; |
573 | |
574 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
575 | vq = &vsock->vqs[i]; |
576 | |
577 | mutex_lock(&vq->mutex); |
578 | |
579 | if (!vhost_vq_access_ok(vq)) { |
580 | ret = -EFAULT; |
581 | goto err_vq; |
582 | } |
583 | |
584 | if (!vhost_vq_get_backend(vq)) { |
585 | vhost_vq_set_backend(vq, private_data: vsock); |
586 | ret = vhost_vq_init_access(vq); |
587 | if (ret) |
588 | goto err_vq; |
589 | } |
590 | |
591 | mutex_unlock(lock: &vq->mutex); |
592 | } |
593 | |
594 | /* Some packets may have been queued before the device was started, |
595 | * let's kick the send worker to send them. |
596 | */ |
597 | vhost_vq_work_queue(vq: &vsock->vqs[VSOCK_VQ_RX], work: &vsock->send_pkt_work); |
598 | |
599 | mutex_unlock(lock: &vsock->dev.mutex); |
600 | return 0; |
601 | |
602 | err_vq: |
603 | vhost_vq_set_backend(vq, NULL); |
604 | mutex_unlock(lock: &vq->mutex); |
605 | |
606 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
607 | vq = &vsock->vqs[i]; |
608 | |
609 | mutex_lock(&vq->mutex); |
610 | vhost_vq_set_backend(vq, NULL); |
611 | mutex_unlock(lock: &vq->mutex); |
612 | } |
613 | err: |
614 | mutex_unlock(lock: &vsock->dev.mutex); |
615 | return ret; |
616 | } |
617 | |
618 | static int vhost_vsock_stop(struct vhost_vsock *vsock, bool check_owner) |
619 | { |
620 | size_t i; |
621 | int ret = 0; |
622 | |
623 | mutex_lock(&vsock->dev.mutex); |
624 | |
625 | if (check_owner) { |
626 | ret = vhost_dev_check_owner(&vsock->dev); |
627 | if (ret) |
628 | goto err; |
629 | } |
630 | |
631 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
632 | struct vhost_virtqueue *vq = &vsock->vqs[i]; |
633 | |
634 | mutex_lock(&vq->mutex); |
635 | vhost_vq_set_backend(vq, NULL); |
636 | mutex_unlock(lock: &vq->mutex); |
637 | } |
638 | |
639 | err: |
640 | mutex_unlock(lock: &vsock->dev.mutex); |
641 | return ret; |
642 | } |
643 | |
644 | static void vhost_vsock_free(struct vhost_vsock *vsock) |
645 | { |
646 | kvfree(addr: vsock); |
647 | } |
648 | |
649 | static int vhost_vsock_dev_open(struct inode *inode, struct file *file) |
650 | { |
651 | struct vhost_virtqueue **vqs; |
652 | struct vhost_vsock *vsock; |
653 | int ret; |
654 | |
655 | /* This struct is large and allocation could fail, fall back to vmalloc |
656 | * if there is no other way. |
657 | */ |
658 | vsock = kvmalloc(size: sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
659 | if (!vsock) |
660 | return -ENOMEM; |
661 | |
662 | vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), size: sizeof(*vqs), GFP_KERNEL); |
663 | if (!vqs) { |
664 | ret = -ENOMEM; |
665 | goto out; |
666 | } |
667 | |
668 | vsock->guest_cid = 0; /* no CID assigned yet */ |
669 | |
670 | atomic_set(v: &vsock->queued_replies, i: 0); |
671 | |
672 | vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX]; |
673 | vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX]; |
674 | vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick; |
675 | vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick; |
676 | |
677 | vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs), |
678 | UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT, |
679 | VHOST_VSOCK_WEIGHT, use_worker: true, NULL); |
680 | |
681 | file->private_data = vsock; |
682 | skb_queue_head_init(list: &vsock->send_pkt_queue); |
683 | vhost_work_init(work: &vsock->send_pkt_work, fn: vhost_transport_send_pkt_work); |
684 | return 0; |
685 | |
686 | out: |
687 | vhost_vsock_free(vsock); |
688 | return ret; |
689 | } |
690 | |
691 | static void vhost_vsock_flush(struct vhost_vsock *vsock) |
692 | { |
693 | vhost_dev_flush(dev: &vsock->dev); |
694 | } |
695 | |
696 | static void vhost_vsock_reset_orphans(struct sock *sk) |
697 | { |
698 | struct vsock_sock *vsk = vsock_sk(sk); |
699 | |
700 | /* vmci_transport.c doesn't take sk_lock here either. At least we're |
701 | * under vsock_table_lock so the sock cannot disappear while we're |
702 | * executing. |
703 | */ |
704 | |
705 | /* If the peer is still valid, no need to reset connection */ |
706 | if (vhost_vsock_get(guest_cid: vsk->remote_addr.svm_cid)) |
707 | return; |
708 | |
709 | /* If the close timeout is pending, let it expire. This avoids races |
710 | * with the timeout callback. |
711 | */ |
712 | if (vsk->close_work_scheduled) |
713 | return; |
714 | |
715 | sock_set_flag(sk, flag: SOCK_DONE); |
716 | vsk->peer_shutdown = SHUTDOWN_MASK; |
717 | sk->sk_state = SS_UNCONNECTED; |
718 | sk->sk_err = ECONNRESET; |
719 | sk_error_report(sk); |
720 | } |
721 | |
722 | static int vhost_vsock_dev_release(struct inode *inode, struct file *file) |
723 | { |
724 | struct vhost_vsock *vsock = file->private_data; |
725 | |
726 | mutex_lock(&vhost_vsock_mutex); |
727 | if (vsock->guest_cid) |
728 | hash_del_rcu(node: &vsock->hash); |
729 | mutex_unlock(lock: &vhost_vsock_mutex); |
730 | |
731 | /* Wait for other CPUs to finish using vsock */ |
732 | synchronize_rcu(); |
733 | |
734 | /* Iterating over all connections for all CIDs to find orphans is |
735 | * inefficient. Room for improvement here. */ |
736 | vsock_for_each_connected_socket(transport: &vhost_transport.transport, |
737 | fn: vhost_vsock_reset_orphans); |
738 | |
739 | /* Don't check the owner, because we are in the release path, so we |
740 | * need to stop the vsock device in any case. |
741 | * vhost_vsock_stop() can not fail in this case, so we don't need to |
742 | * check the return code. |
743 | */ |
744 | vhost_vsock_stop(vsock, check_owner: false); |
745 | vhost_vsock_flush(vsock); |
746 | vhost_dev_stop(&vsock->dev); |
747 | |
748 | virtio_vsock_skb_queue_purge(list: &vsock->send_pkt_queue); |
749 | |
750 | vhost_dev_cleanup(&vsock->dev); |
751 | kfree(objp: vsock->dev.vqs); |
752 | vhost_vsock_free(vsock); |
753 | return 0; |
754 | } |
755 | |
756 | static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid) |
757 | { |
758 | struct vhost_vsock *other; |
759 | |
760 | /* Refuse reserved CIDs */ |
761 | if (guest_cid <= VMADDR_CID_HOST || |
762 | guest_cid == U32_MAX) |
763 | return -EINVAL; |
764 | |
765 | /* 64-bit CIDs are not yet supported */ |
766 | if (guest_cid > U32_MAX) |
767 | return -EINVAL; |
768 | |
769 | /* Refuse if CID is assigned to the guest->host transport (i.e. nested |
770 | * VM), to make the loopback work. |
771 | */ |
772 | if (vsock_find_cid(cid: guest_cid)) |
773 | return -EADDRINUSE; |
774 | |
775 | /* Refuse if CID is already in use */ |
776 | mutex_lock(&vhost_vsock_mutex); |
777 | other = vhost_vsock_get(guest_cid); |
778 | if (other && other != vsock) { |
779 | mutex_unlock(lock: &vhost_vsock_mutex); |
780 | return -EADDRINUSE; |
781 | } |
782 | |
783 | if (vsock->guest_cid) |
784 | hash_del_rcu(node: &vsock->hash); |
785 | |
786 | vsock->guest_cid = guest_cid; |
787 | hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid); |
788 | mutex_unlock(lock: &vhost_vsock_mutex); |
789 | |
790 | return 0; |
791 | } |
792 | |
793 | static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features) |
794 | { |
795 | struct vhost_virtqueue *vq; |
796 | int i; |
797 | |
798 | if (features & ~VHOST_VSOCK_FEATURES) |
799 | return -EOPNOTSUPP; |
800 | |
801 | mutex_lock(&vsock->dev.mutex); |
802 | if ((features & (1 << VHOST_F_LOG_ALL)) && |
803 | !vhost_log_access_ok(&vsock->dev)) { |
804 | goto err; |
805 | } |
806 | |
807 | if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) { |
808 | if (vhost_init_device_iotlb(d: &vsock->dev)) |
809 | goto err; |
810 | } |
811 | |
812 | if (features & (1ULL << VIRTIO_VSOCK_F_SEQPACKET)) |
813 | vsock->seqpacket_allow = true; |
814 | |
815 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
816 | vq = &vsock->vqs[i]; |
817 | mutex_lock(&vq->mutex); |
818 | vq->acked_features = features; |
819 | mutex_unlock(lock: &vq->mutex); |
820 | } |
821 | mutex_unlock(lock: &vsock->dev.mutex); |
822 | return 0; |
823 | |
824 | err: |
825 | mutex_unlock(lock: &vsock->dev.mutex); |
826 | return -EFAULT; |
827 | } |
828 | |
829 | static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl, |
830 | unsigned long arg) |
831 | { |
832 | struct vhost_vsock *vsock = f->private_data; |
833 | void __user *argp = (void __user *)arg; |
834 | u64 guest_cid; |
835 | u64 features; |
836 | int start; |
837 | int r; |
838 | |
839 | switch (ioctl) { |
840 | case VHOST_VSOCK_SET_GUEST_CID: |
841 | if (copy_from_user(to: &guest_cid, from: argp, n: sizeof(guest_cid))) |
842 | return -EFAULT; |
843 | return vhost_vsock_set_cid(vsock, guest_cid); |
844 | case VHOST_VSOCK_SET_RUNNING: |
845 | if (copy_from_user(to: &start, from: argp, n: sizeof(start))) |
846 | return -EFAULT; |
847 | if (start) |
848 | return vhost_vsock_start(vsock); |
849 | else |
850 | return vhost_vsock_stop(vsock, check_owner: true); |
851 | case VHOST_GET_FEATURES: |
852 | features = VHOST_VSOCK_FEATURES; |
853 | if (copy_to_user(to: argp, from: &features, n: sizeof(features))) |
854 | return -EFAULT; |
855 | return 0; |
856 | case VHOST_SET_FEATURES: |
857 | if (copy_from_user(to: &features, from: argp, n: sizeof(features))) |
858 | return -EFAULT; |
859 | return vhost_vsock_set_features(vsock, features); |
860 | case VHOST_GET_BACKEND_FEATURES: |
861 | features = VHOST_VSOCK_BACKEND_FEATURES; |
862 | if (copy_to_user(to: argp, from: &features, n: sizeof(features))) |
863 | return -EFAULT; |
864 | return 0; |
865 | case VHOST_SET_BACKEND_FEATURES: |
866 | if (copy_from_user(to: &features, from: argp, n: sizeof(features))) |
867 | return -EFAULT; |
868 | if (features & ~VHOST_VSOCK_BACKEND_FEATURES) |
869 | return -EOPNOTSUPP; |
870 | vhost_set_backend_features(dev: &vsock->dev, features); |
871 | return 0; |
872 | default: |
873 | mutex_lock(&vsock->dev.mutex); |
874 | r = vhost_dev_ioctl(&vsock->dev, ioctl, argp); |
875 | if (r == -ENOIOCTLCMD) |
876 | r = vhost_vring_ioctl(d: &vsock->dev, ioctl, argp); |
877 | else |
878 | vhost_vsock_flush(vsock); |
879 | mutex_unlock(lock: &vsock->dev.mutex); |
880 | return r; |
881 | } |
882 | } |
883 | |
884 | static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to) |
885 | { |
886 | struct file *file = iocb->ki_filp; |
887 | struct vhost_vsock *vsock = file->private_data; |
888 | struct vhost_dev *dev = &vsock->dev; |
889 | int noblock = file->f_flags & O_NONBLOCK; |
890 | |
891 | return vhost_chr_read_iter(dev, to, noblock); |
892 | } |
893 | |
894 | static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb, |
895 | struct iov_iter *from) |
896 | { |
897 | struct file *file = iocb->ki_filp; |
898 | struct vhost_vsock *vsock = file->private_data; |
899 | struct vhost_dev *dev = &vsock->dev; |
900 | |
901 | return vhost_chr_write_iter(dev, from); |
902 | } |
903 | |
904 | static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait) |
905 | { |
906 | struct vhost_vsock *vsock = file->private_data; |
907 | struct vhost_dev *dev = &vsock->dev; |
908 | |
909 | return vhost_chr_poll(file, dev, wait); |
910 | } |
911 | |
912 | static const struct file_operations vhost_vsock_fops = { |
913 | .owner = THIS_MODULE, |
914 | .open = vhost_vsock_dev_open, |
915 | .release = vhost_vsock_dev_release, |
916 | .llseek = noop_llseek, |
917 | .unlocked_ioctl = vhost_vsock_dev_ioctl, |
918 | .compat_ioctl = compat_ptr_ioctl, |
919 | .read_iter = vhost_vsock_chr_read_iter, |
920 | .write_iter = vhost_vsock_chr_write_iter, |
921 | .poll = vhost_vsock_chr_poll, |
922 | }; |
923 | |
924 | static struct miscdevice vhost_vsock_misc = { |
925 | .minor = VHOST_VSOCK_MINOR, |
926 | .name = "vhost-vsock" , |
927 | .fops = &vhost_vsock_fops, |
928 | }; |
929 | |
930 | static int __init vhost_vsock_init(void) |
931 | { |
932 | int ret; |
933 | |
934 | ret = vsock_core_register(t: &vhost_transport.transport, |
935 | VSOCK_TRANSPORT_F_H2G); |
936 | if (ret < 0) |
937 | return ret; |
938 | |
939 | ret = misc_register(misc: &vhost_vsock_misc); |
940 | if (ret) { |
941 | vsock_core_unregister(t: &vhost_transport.transport); |
942 | return ret; |
943 | } |
944 | |
945 | return 0; |
946 | }; |
947 | |
948 | static void __exit vhost_vsock_exit(void) |
949 | { |
950 | misc_deregister(misc: &vhost_vsock_misc); |
951 | vsock_core_unregister(t: &vhost_transport.transport); |
952 | }; |
953 | |
954 | module_init(vhost_vsock_init); |
955 | module_exit(vhost_vsock_exit); |
956 | MODULE_LICENSE("GPL v2" ); |
957 | MODULE_AUTHOR("Asias He" ); |
958 | MODULE_DESCRIPTION("vhost transport for vsock " ); |
959 | MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR); |
960 | MODULE_ALIAS("devname:vhost-vsock" ); |
961 | |