Commit | Line | Data |
---|---|---|
7a338472 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
0ea9e1d3 AH |
2 | /* |
3 | * virtio 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 | * Some of the code is take from Gerd Hoffmann <kraxel@redhat.com>'s | |
10 | * early virtio-vsock proof-of-concept bits. | |
0ea9e1d3 AH |
11 | */ |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/module.h> | |
14 | #include <linux/list.h> | |
15 | #include <linux/atomic.h> | |
16 | #include <linux/virtio.h> | |
17 | #include <linux/virtio_ids.h> | |
18 | #include <linux/virtio_config.h> | |
19 | #include <linux/virtio_vsock.h> | |
20 | #include <net/sock.h> | |
21 | #include <linux/mutex.h> | |
22 | #include <net/af_vsock.h> | |
23 | ||
24 | static struct workqueue_struct *virtio_vsock_workqueue; | |
f961134a | 25 | static struct virtio_vsock __rcu *the_virtio_vsock; |
0ea9e1d3 | 26 | static DEFINE_MUTEX(the_virtio_vsock_mutex); /* protects the_virtio_vsock */ |
8e6ed963 | 27 | static struct virtio_transport virtio_transport; /* forward declaration */ |
0ea9e1d3 AH |
28 | |
29 | struct virtio_vsock { | |
30 | struct virtio_device *vdev; | |
31 | struct virtqueue *vqs[VSOCK_VQ_MAX]; | |
32 | ||
33 | /* Virtqueue processing is deferred to a workqueue */ | |
34 | struct work_struct tx_work; | |
35 | struct work_struct rx_work; | |
36 | struct work_struct event_work; | |
37 | ||
38 | /* The following fields are protected by tx_lock. vqs[VSOCK_VQ_TX] | |
39 | * must be accessed with tx_lock held. | |
40 | */ | |
41 | struct mutex tx_lock; | |
b917507e | 42 | bool tx_run; |
0ea9e1d3 AH |
43 | |
44 | struct work_struct send_pkt_work; | |
45 | spinlock_t send_pkt_list_lock; | |
46 | struct list_head send_pkt_list; | |
47 | ||
48 | atomic_t queued_replies; | |
49 | ||
50 | /* The following fields are protected by rx_lock. vqs[VSOCK_VQ_RX] | |
51 | * must be accessed with rx_lock held. | |
52 | */ | |
53 | struct mutex rx_lock; | |
b917507e | 54 | bool rx_run; |
0ea9e1d3 AH |
55 | int rx_buf_nr; |
56 | int rx_buf_max_nr; | |
57 | ||
58 | /* The following fields are protected by event_lock. | |
59 | * vqs[VSOCK_VQ_EVENT] must be accessed with event_lock held. | |
60 | */ | |
61 | struct mutex event_lock; | |
b917507e | 62 | bool event_run; |
0ea9e1d3 AH |
63 | struct virtio_vsock_event event_list[8]; |
64 | ||
65 | u32 guest_cid; | |
53efbba1 | 66 | bool seqpacket_allow; |
0ea9e1d3 AH |
67 | }; |
68 | ||
0ea9e1d3 AH |
69 | static u32 virtio_transport_get_local_cid(void) |
70 | { | |
0deab087 SG |
71 | struct virtio_vsock *vsock; |
72 | u32 ret; | |
0ea9e1d3 | 73 | |
0deab087 SG |
74 | rcu_read_lock(); |
75 | vsock = rcu_dereference(the_virtio_vsock); | |
76 | if (!vsock) { | |
77 | ret = VMADDR_CID_ANY; | |
78 | goto out_rcu; | |
79 | } | |
22b5c0b6 | 80 | |
0deab087 SG |
81 | ret = vsock->guest_cid; |
82 | out_rcu: | |
83 | rcu_read_unlock(); | |
84 | return ret; | |
0ea9e1d3 AH |
85 | } |
86 | ||
87 | static void | |
88 | virtio_transport_send_pkt_work(struct work_struct *work) | |
89 | { | |
90 | struct virtio_vsock *vsock = | |
91 | container_of(work, struct virtio_vsock, send_pkt_work); | |
92 | struct virtqueue *vq; | |
93 | bool added = false; | |
94 | bool restart_rx = false; | |
95 | ||
96 | mutex_lock(&vsock->tx_lock); | |
97 | ||
b917507e SG |
98 | if (!vsock->tx_run) |
99 | goto out; | |
100 | ||
0ea9e1d3 AH |
101 | vq = vsock->vqs[VSOCK_VQ_TX]; |
102 | ||
0ea9e1d3 AH |
103 | for (;;) { |
104 | struct virtio_vsock_pkt *pkt; | |
105 | struct scatterlist hdr, buf, *sgs[2]; | |
106 | int ret, in_sg = 0, out_sg = 0; | |
107 | bool reply; | |
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); | |
0ea9e1d3 AH |
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 | ||
82dfb540 GG |
120 | virtio_transport_deliver_tap_pkt(pkt); |
121 | ||
0ea9e1d3 AH |
122 | reply = pkt->reply; |
123 | ||
124 | sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr)); | |
125 | sgs[out_sg++] = &hdr; | |
126 | if (pkt->buf) { | |
127 | sg_init_one(&buf, pkt->buf, pkt->len); | |
128 | sgs[out_sg++] = &buf; | |
129 | } | |
130 | ||
131 | ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, pkt, GFP_KERNEL); | |
21bc54fc GG |
132 | /* Usually this means that there is no more space available in |
133 | * the vq | |
134 | */ | |
0ea9e1d3 AH |
135 | if (ret < 0) { |
136 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
137 | list_add(&pkt->list, &vsock->send_pkt_list); | |
138 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
0ea9e1d3 AH |
139 | break; |
140 | } | |
141 | ||
142 | if (reply) { | |
143 | struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX]; | |
144 | int val; | |
145 | ||
146 | val = atomic_dec_return(&vsock->queued_replies); | |
147 | ||
148 | /* Do we now have resources to resume rx processing? */ | |
149 | if (val + 1 == virtqueue_get_vring_size(rx_vq)) | |
150 | restart_rx = true; | |
151 | } | |
152 | ||
153 | added = true; | |
154 | } | |
155 | ||
156 | if (added) | |
157 | virtqueue_kick(vq); | |
158 | ||
b917507e | 159 | out: |
0ea9e1d3 AH |
160 | mutex_unlock(&vsock->tx_lock); |
161 | ||
162 | if (restart_rx) | |
163 | queue_work(virtio_vsock_workqueue, &vsock->rx_work); | |
164 | } | |
165 | ||
166 | static int | |
167 | virtio_transport_send_pkt(struct virtio_vsock_pkt *pkt) | |
168 | { | |
169 | struct virtio_vsock *vsock; | |
170 | int len = pkt->len; | |
171 | ||
0deab087 SG |
172 | rcu_read_lock(); |
173 | vsock = rcu_dereference(the_virtio_vsock); | |
0ea9e1d3 AH |
174 | if (!vsock) { |
175 | virtio_transport_free_pkt(pkt); | |
0deab087 SG |
176 | len = -ENODEV; |
177 | goto out_rcu; | |
0ea9e1d3 AH |
178 | } |
179 | ||
0deab087 | 180 | if (le64_to_cpu(pkt->hdr.dst_cid) == vsock->guest_cid) { |
bf5432b1 SG |
181 | virtio_transport_free_pkt(pkt); |
182 | len = -ENODEV; | |
0deab087 SG |
183 | goto out_rcu; |
184 | } | |
b9116823 | 185 | |
0ea9e1d3 AH |
186 | if (pkt->reply) |
187 | atomic_inc(&vsock->queued_replies); | |
188 | ||
189 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
190 | list_add_tail(&pkt->list, &vsock->send_pkt_list); | |
191 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
192 | ||
193 | queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work); | |
0deab087 SG |
194 | |
195 | out_rcu: | |
196 | rcu_read_unlock(); | |
0ea9e1d3 AH |
197 | return len; |
198 | } | |
199 | ||
073b4f2c PT |
200 | static int |
201 | virtio_transport_cancel_pkt(struct vsock_sock *vsk) | |
202 | { | |
203 | struct virtio_vsock *vsock; | |
204 | struct virtio_vsock_pkt *pkt, *n; | |
0deab087 | 205 | int cnt = 0, ret; |
073b4f2c PT |
206 | LIST_HEAD(freeme); |
207 | ||
0deab087 SG |
208 | rcu_read_lock(); |
209 | vsock = rcu_dereference(the_virtio_vsock); | |
073b4f2c | 210 | if (!vsock) { |
0deab087 SG |
211 | ret = -ENODEV; |
212 | goto out_rcu; | |
073b4f2c PT |
213 | } |
214 | ||
215 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
216 | list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) { | |
217 | if (pkt->vsk != vsk) | |
218 | continue; | |
219 | list_move(&pkt->list, &freeme); | |
220 | } | |
221 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
222 | ||
223 | list_for_each_entry_safe(pkt, n, &freeme, list) { | |
224 | if (pkt->reply) | |
225 | cnt++; | |
226 | list_del(&pkt->list); | |
227 | virtio_transport_free_pkt(pkt); | |
228 | } | |
229 | ||
230 | if (cnt) { | |
231 | struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX]; | |
232 | int new_cnt; | |
233 | ||
234 | new_cnt = atomic_sub_return(cnt, &vsock->queued_replies); | |
235 | if (new_cnt + cnt >= virtqueue_get_vring_size(rx_vq) && | |
236 | new_cnt < virtqueue_get_vring_size(rx_vq)) | |
237 | queue_work(virtio_vsock_workqueue, &vsock->rx_work); | |
238 | } | |
239 | ||
0deab087 SG |
240 | ret = 0; |
241 | ||
242 | out_rcu: | |
243 | rcu_read_unlock(); | |
244 | return ret; | |
073b4f2c PT |
245 | } |
246 | ||
0ea9e1d3 AH |
247 | static void virtio_vsock_rx_fill(struct virtio_vsock *vsock) |
248 | { | |
249 | int buf_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE; | |
250 | struct virtio_vsock_pkt *pkt; | |
251 | struct scatterlist hdr, buf, *sgs[2]; | |
252 | struct virtqueue *vq; | |
253 | int ret; | |
254 | ||
255 | vq = vsock->vqs[VSOCK_VQ_RX]; | |
256 | ||
257 | do { | |
258 | pkt = kzalloc(sizeof(*pkt), GFP_KERNEL); | |
259 | if (!pkt) | |
260 | break; | |
261 | ||
262 | pkt->buf = kmalloc(buf_len, GFP_KERNEL); | |
263 | if (!pkt->buf) { | |
264 | virtio_transport_free_pkt(pkt); | |
265 | break; | |
266 | } | |
267 | ||
473c7391 | 268 | pkt->buf_len = buf_len; |
0ea9e1d3 AH |
269 | pkt->len = buf_len; |
270 | ||
271 | sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr)); | |
272 | sgs[0] = &hdr; | |
273 | ||
274 | sg_init_one(&buf, pkt->buf, buf_len); | |
275 | sgs[1] = &buf; | |
276 | ret = virtqueue_add_sgs(vq, sgs, 0, 2, pkt, GFP_KERNEL); | |
277 | if (ret) { | |
278 | virtio_transport_free_pkt(pkt); | |
279 | break; | |
280 | } | |
281 | vsock->rx_buf_nr++; | |
282 | } while (vq->num_free); | |
283 | if (vsock->rx_buf_nr > vsock->rx_buf_max_nr) | |
284 | vsock->rx_buf_max_nr = vsock->rx_buf_nr; | |
285 | virtqueue_kick(vq); | |
286 | } | |
287 | ||
288 | static void virtio_transport_tx_work(struct work_struct *work) | |
289 | { | |
290 | struct virtio_vsock *vsock = | |
291 | container_of(work, struct virtio_vsock, tx_work); | |
292 | struct virtqueue *vq; | |
293 | bool added = false; | |
294 | ||
295 | vq = vsock->vqs[VSOCK_VQ_TX]; | |
296 | mutex_lock(&vsock->tx_lock); | |
b917507e SG |
297 | |
298 | if (!vsock->tx_run) | |
299 | goto out; | |
300 | ||
0ea9e1d3 AH |
301 | do { |
302 | struct virtio_vsock_pkt *pkt; | |
303 | unsigned int len; | |
304 | ||
305 | virtqueue_disable_cb(vq); | |
306 | while ((pkt = virtqueue_get_buf(vq, &len)) != NULL) { | |
307 | virtio_transport_free_pkt(pkt); | |
308 | added = true; | |
309 | } | |
310 | } while (!virtqueue_enable_cb(vq)); | |
b917507e SG |
311 | |
312 | out: | |
0ea9e1d3 AH |
313 | mutex_unlock(&vsock->tx_lock); |
314 | ||
315 | if (added) | |
316 | queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work); | |
317 | } | |
318 | ||
319 | /* Is there space left for replies to rx packets? */ | |
320 | static bool virtio_transport_more_replies(struct virtio_vsock *vsock) | |
321 | { | |
322 | struct virtqueue *vq = vsock->vqs[VSOCK_VQ_RX]; | |
323 | int val; | |
324 | ||
325 | smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */ | |
326 | val = atomic_read(&vsock->queued_replies); | |
327 | ||
328 | return val < virtqueue_get_vring_size(vq); | |
329 | } | |
330 | ||
0ea9e1d3 AH |
331 | /* event_lock must be held */ |
332 | static int virtio_vsock_event_fill_one(struct virtio_vsock *vsock, | |
333 | struct virtio_vsock_event *event) | |
334 | { | |
335 | struct scatterlist sg; | |
336 | struct virtqueue *vq; | |
337 | ||
338 | vq = vsock->vqs[VSOCK_VQ_EVENT]; | |
339 | ||
340 | sg_init_one(&sg, event, sizeof(*event)); | |
341 | ||
342 | return virtqueue_add_inbuf(vq, &sg, 1, event, GFP_KERNEL); | |
343 | } | |
344 | ||
345 | /* event_lock must be held */ | |
346 | static void virtio_vsock_event_fill(struct virtio_vsock *vsock) | |
347 | { | |
348 | size_t i; | |
349 | ||
350 | for (i = 0; i < ARRAY_SIZE(vsock->event_list); i++) { | |
351 | struct virtio_vsock_event *event = &vsock->event_list[i]; | |
352 | ||
353 | virtio_vsock_event_fill_one(vsock, event); | |
354 | } | |
355 | ||
356 | virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]); | |
357 | } | |
358 | ||
359 | static void virtio_vsock_reset_sock(struct sock *sk) | |
360 | { | |
49b0b6ff LM |
361 | /* vmci_transport.c doesn't take sk_lock here either. At least we're |
362 | * under vsock_table_lock so the sock cannot disappear while we're | |
363 | * executing. | |
364 | */ | |
365 | ||
3b4477d2 | 366 | sk->sk_state = TCP_CLOSE; |
0ea9e1d3 | 367 | sk->sk_err = ECONNRESET; |
e3ae2365 | 368 | sk_error_report(sk); |
0ea9e1d3 AH |
369 | } |
370 | ||
371 | static void virtio_vsock_update_guest_cid(struct virtio_vsock *vsock) | |
372 | { | |
373 | struct virtio_device *vdev = vsock->vdev; | |
6c7efafd | 374 | __le64 guest_cid; |
0ea9e1d3 AH |
375 | |
376 | vdev->config->get(vdev, offsetof(struct virtio_vsock_config, guest_cid), | |
377 | &guest_cid, sizeof(guest_cid)); | |
378 | vsock->guest_cid = le64_to_cpu(guest_cid); | |
379 | } | |
380 | ||
381 | /* event_lock must be held */ | |
382 | static void virtio_vsock_event_handle(struct virtio_vsock *vsock, | |
383 | struct virtio_vsock_event *event) | |
384 | { | |
385 | switch (le32_to_cpu(event->id)) { | |
386 | case VIRTIO_VSOCK_EVENT_TRANSPORT_RESET: | |
387 | virtio_vsock_update_guest_cid(vsock); | |
8e6ed963 JP |
388 | vsock_for_each_connected_socket(&virtio_transport.transport, |
389 | virtio_vsock_reset_sock); | |
0ea9e1d3 AH |
390 | break; |
391 | } | |
392 | } | |
393 | ||
394 | static void virtio_transport_event_work(struct work_struct *work) | |
395 | { | |
396 | struct virtio_vsock *vsock = | |
397 | container_of(work, struct virtio_vsock, event_work); | |
398 | struct virtqueue *vq; | |
399 | ||
400 | vq = vsock->vqs[VSOCK_VQ_EVENT]; | |
401 | ||
402 | mutex_lock(&vsock->event_lock); | |
403 | ||
b917507e SG |
404 | if (!vsock->event_run) |
405 | goto out; | |
406 | ||
0ea9e1d3 AH |
407 | do { |
408 | struct virtio_vsock_event *event; | |
409 | unsigned int len; | |
410 | ||
411 | virtqueue_disable_cb(vq); | |
412 | while ((event = virtqueue_get_buf(vq, &len)) != NULL) { | |
413 | if (len == sizeof(*event)) | |
414 | virtio_vsock_event_handle(vsock, event); | |
415 | ||
416 | virtio_vsock_event_fill_one(vsock, event); | |
417 | } | |
418 | } while (!virtqueue_enable_cb(vq)); | |
419 | ||
420 | virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]); | |
b917507e | 421 | out: |
0ea9e1d3 AH |
422 | mutex_unlock(&vsock->event_lock); |
423 | } | |
424 | ||
425 | static void virtio_vsock_event_done(struct virtqueue *vq) | |
426 | { | |
427 | struct virtio_vsock *vsock = vq->vdev->priv; | |
428 | ||
429 | if (!vsock) | |
430 | return; | |
431 | queue_work(virtio_vsock_workqueue, &vsock->event_work); | |
432 | } | |
433 | ||
434 | static void virtio_vsock_tx_done(struct virtqueue *vq) | |
435 | { | |
436 | struct virtio_vsock *vsock = vq->vdev->priv; | |
437 | ||
438 | if (!vsock) | |
439 | return; | |
440 | queue_work(virtio_vsock_workqueue, &vsock->tx_work); | |
441 | } | |
442 | ||
443 | static void virtio_vsock_rx_done(struct virtqueue *vq) | |
444 | { | |
445 | struct virtio_vsock *vsock = vq->vdev->priv; | |
446 | ||
447 | if (!vsock) | |
448 | return; | |
449 | queue_work(virtio_vsock_workqueue, &vsock->rx_work); | |
450 | } | |
451 | ||
53efbba1 AK |
452 | static bool virtio_transport_seqpacket_allow(u32 remote_cid); |
453 | ||
0ea9e1d3 AH |
454 | static struct virtio_transport virtio_transport = { |
455 | .transport = { | |
6a2c0962 SG |
456 | .module = THIS_MODULE, |
457 | ||
0ea9e1d3 AH |
458 | .get_local_cid = virtio_transport_get_local_cid, |
459 | ||
460 | .init = virtio_transport_do_socket_init, | |
461 | .destruct = virtio_transport_destruct, | |
462 | .release = virtio_transport_release, | |
463 | .connect = virtio_transport_connect, | |
464 | .shutdown = virtio_transport_shutdown, | |
073b4f2c | 465 | .cancel_pkt = virtio_transport_cancel_pkt, |
0ea9e1d3 AH |
466 | |
467 | .dgram_bind = virtio_transport_dgram_bind, | |
468 | .dgram_dequeue = virtio_transport_dgram_dequeue, | |
469 | .dgram_enqueue = virtio_transport_dgram_enqueue, | |
470 | .dgram_allow = virtio_transport_dgram_allow, | |
471 | ||
472 | .stream_dequeue = virtio_transport_stream_dequeue, | |
473 | .stream_enqueue = virtio_transport_stream_enqueue, | |
474 | .stream_has_data = virtio_transport_stream_has_data, | |
475 | .stream_has_space = virtio_transport_stream_has_space, | |
476 | .stream_rcvhiwat = virtio_transport_stream_rcvhiwat, | |
477 | .stream_is_active = virtio_transport_stream_is_active, | |
478 | .stream_allow = virtio_transport_stream_allow, | |
479 | ||
53efbba1 AK |
480 | .seqpacket_dequeue = virtio_transport_seqpacket_dequeue, |
481 | .seqpacket_enqueue = virtio_transport_seqpacket_enqueue, | |
482 | .seqpacket_allow = virtio_transport_seqpacket_allow, | |
483 | .seqpacket_has_data = virtio_transport_seqpacket_has_data, | |
484 | ||
0ea9e1d3 AH |
485 | .notify_poll_in = virtio_transport_notify_poll_in, |
486 | .notify_poll_out = virtio_transport_notify_poll_out, | |
487 | .notify_recv_init = virtio_transport_notify_recv_init, | |
488 | .notify_recv_pre_block = virtio_transport_notify_recv_pre_block, | |
489 | .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue, | |
490 | .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue, | |
491 | .notify_send_init = virtio_transport_notify_send_init, | |
492 | .notify_send_pre_block = virtio_transport_notify_send_pre_block, | |
493 | .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue, | |
494 | .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue, | |
b9f2b0ff | 495 | .notify_buffer_size = virtio_transport_notify_buffer_size, |
0ea9e1d3 AH |
496 | }, |
497 | ||
498 | .send_pkt = virtio_transport_send_pkt, | |
499 | }; | |
500 | ||
53efbba1 AK |
501 | static bool virtio_transport_seqpacket_allow(u32 remote_cid) |
502 | { | |
503 | struct virtio_vsock *vsock; | |
504 | bool seqpacket_allow; | |
505 | ||
64295f0d | 506 | seqpacket_allow = false; |
53efbba1 AK |
507 | rcu_read_lock(); |
508 | vsock = rcu_dereference(the_virtio_vsock); | |
64295f0d ED |
509 | if (vsock) |
510 | seqpacket_allow = vsock->seqpacket_allow; | |
53efbba1 AK |
511 | rcu_read_unlock(); |
512 | ||
513 | return seqpacket_allow; | |
514 | } | |
515 | ||
4c7246dc SG |
516 | static void virtio_transport_rx_work(struct work_struct *work) |
517 | { | |
518 | struct virtio_vsock *vsock = | |
519 | container_of(work, struct virtio_vsock, rx_work); | |
520 | struct virtqueue *vq; | |
521 | ||
522 | vq = vsock->vqs[VSOCK_VQ_RX]; | |
523 | ||
524 | mutex_lock(&vsock->rx_lock); | |
525 | ||
526 | if (!vsock->rx_run) | |
527 | goto out; | |
528 | ||
529 | do { | |
530 | virtqueue_disable_cb(vq); | |
531 | for (;;) { | |
532 | struct virtio_vsock_pkt *pkt; | |
533 | unsigned int len; | |
534 | ||
535 | if (!virtio_transport_more_replies(vsock)) { | |
536 | /* Stop rx until the device processes already | |
537 | * pending replies. Leave rx virtqueue | |
538 | * callbacks disabled. | |
539 | */ | |
540 | goto out; | |
541 | } | |
542 | ||
543 | pkt = virtqueue_get_buf(vq, &len); | |
544 | if (!pkt) { | |
545 | break; | |
546 | } | |
547 | ||
548 | vsock->rx_buf_nr--; | |
549 | ||
550 | /* Drop short/long packets */ | |
551 | if (unlikely(len < sizeof(pkt->hdr) || | |
552 | len > sizeof(pkt->hdr) + pkt->len)) { | |
553 | virtio_transport_free_pkt(pkt); | |
554 | continue; | |
555 | } | |
556 | ||
557 | pkt->len = len - sizeof(pkt->hdr); | |
558 | virtio_transport_deliver_tap_pkt(pkt); | |
559 | virtio_transport_recv_pkt(&virtio_transport, pkt); | |
560 | } | |
561 | } while (!virtqueue_enable_cb(vq)); | |
562 | ||
563 | out: | |
564 | if (vsock->rx_buf_nr < vsock->rx_buf_max_nr / 2) | |
565 | virtio_vsock_rx_fill(vsock); | |
566 | mutex_unlock(&vsock->rx_lock); | |
567 | } | |
568 | ||
0ea9e1d3 AH |
569 | static int virtio_vsock_probe(struct virtio_device *vdev) |
570 | { | |
571 | vq_callback_t *callbacks[] = { | |
572 | virtio_vsock_rx_done, | |
573 | virtio_vsock_tx_done, | |
574 | virtio_vsock_event_done, | |
575 | }; | |
576 | static const char * const names[] = { | |
577 | "rx", | |
578 | "tx", | |
579 | "event", | |
580 | }; | |
581 | struct virtio_vsock *vsock = NULL; | |
582 | int ret; | |
583 | ||
584 | ret = mutex_lock_interruptible(&the_virtio_vsock_mutex); | |
585 | if (ret) | |
586 | return ret; | |
587 | ||
588 | /* Only one virtio-vsock device per guest is supported */ | |
0deab087 SG |
589 | if (rcu_dereference_protected(the_virtio_vsock, |
590 | lockdep_is_held(&the_virtio_vsock_mutex))) { | |
0ea9e1d3 AH |
591 | ret = -EBUSY; |
592 | goto out; | |
593 | } | |
594 | ||
595 | vsock = kzalloc(sizeof(*vsock), GFP_KERNEL); | |
596 | if (!vsock) { | |
597 | ret = -ENOMEM; | |
598 | goto out; | |
599 | } | |
600 | ||
601 | vsock->vdev = vdev; | |
602 | ||
9b2bbdb2 MT |
603 | ret = virtio_find_vqs(vsock->vdev, VSOCK_VQ_MAX, |
604 | vsock->vqs, callbacks, names, | |
605 | NULL); | |
0ea9e1d3 AH |
606 | if (ret < 0) |
607 | goto out; | |
608 | ||
609 | virtio_vsock_update_guest_cid(vsock); | |
610 | ||
0ea9e1d3 AH |
611 | vsock->rx_buf_nr = 0; |
612 | vsock->rx_buf_max_nr = 0; | |
613 | atomic_set(&vsock->queued_replies, 0); | |
614 | ||
0ea9e1d3 AH |
615 | mutex_init(&vsock->tx_lock); |
616 | mutex_init(&vsock->rx_lock); | |
617 | mutex_init(&vsock->event_lock); | |
618 | spin_lock_init(&vsock->send_pkt_list_lock); | |
619 | INIT_LIST_HEAD(&vsock->send_pkt_list); | |
620 | INIT_WORK(&vsock->rx_work, virtio_transport_rx_work); | |
621 | INIT_WORK(&vsock->tx_work, virtio_transport_tx_work); | |
622 | INIT_WORK(&vsock->event_work, virtio_transport_event_work); | |
623 | INIT_WORK(&vsock->send_pkt_work, virtio_transport_send_pkt_work); | |
624 | ||
c1011c0b SG |
625 | if (virtio_has_feature(vdev, VIRTIO_VSOCK_F_SEQPACKET)) |
626 | vsock->seqpacket_allow = true; | |
627 | ||
4b5f1ad5 SG |
628 | vdev->priv = vsock; |
629 | ||
b917507e SG |
630 | mutex_lock(&vsock->tx_lock); |
631 | vsock->tx_run = true; | |
632 | mutex_unlock(&vsock->tx_lock); | |
633 | ||
0ea9e1d3 AH |
634 | mutex_lock(&vsock->rx_lock); |
635 | virtio_vsock_rx_fill(vsock); | |
b917507e | 636 | vsock->rx_run = true; |
0ea9e1d3 AH |
637 | mutex_unlock(&vsock->rx_lock); |
638 | ||
639 | mutex_lock(&vsock->event_lock); | |
640 | virtio_vsock_event_fill(vsock); | |
b917507e | 641 | vsock->event_run = true; |
0ea9e1d3 AH |
642 | mutex_unlock(&vsock->event_lock); |
643 | ||
0deab087 SG |
644 | rcu_assign_pointer(the_virtio_vsock, vsock); |
645 | ||
0ea9e1d3 | 646 | mutex_unlock(&the_virtio_vsock_mutex); |
53efbba1 | 647 | |
0ea9e1d3 AH |
648 | return 0; |
649 | ||
0ea9e1d3 AH |
650 | out: |
651 | kfree(vsock); | |
652 | mutex_unlock(&the_virtio_vsock_mutex); | |
653 | return ret; | |
654 | } | |
655 | ||
656 | static void virtio_vsock_remove(struct virtio_device *vdev) | |
657 | { | |
658 | struct virtio_vsock *vsock = vdev->priv; | |
659 | struct virtio_vsock_pkt *pkt; | |
660 | ||
0deab087 SG |
661 | mutex_lock(&the_virtio_vsock_mutex); |
662 | ||
663 | vdev->priv = NULL; | |
664 | rcu_assign_pointer(the_virtio_vsock, NULL); | |
665 | synchronize_rcu(); | |
666 | ||
85965487 | 667 | /* Reset all connected sockets when the device disappear */ |
8e6ed963 JP |
668 | vsock_for_each_connected_socket(&virtio_transport.transport, |
669 | virtio_vsock_reset_sock); | |
85965487 | 670 | |
b917507e | 671 | /* Stop all work handlers to make sure no one is accessing the device, |
d9679d00 | 672 | * so we can safely call virtio_reset_device(). |
b917507e SG |
673 | */ |
674 | mutex_lock(&vsock->rx_lock); | |
675 | vsock->rx_run = false; | |
676 | mutex_unlock(&vsock->rx_lock); | |
677 | ||
678 | mutex_lock(&vsock->tx_lock); | |
679 | vsock->tx_run = false; | |
680 | mutex_unlock(&vsock->tx_lock); | |
681 | ||
682 | mutex_lock(&vsock->event_lock); | |
683 | vsock->event_run = false; | |
684 | mutex_unlock(&vsock->event_lock); | |
685 | ||
686 | /* Flush all device writes and interrupts, device will not use any | |
687 | * more buffers. | |
688 | */ | |
d9679d00 | 689 | virtio_reset_device(vdev); |
0ea9e1d3 AH |
690 | |
691 | mutex_lock(&vsock->rx_lock); | |
692 | while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX]))) | |
693 | virtio_transport_free_pkt(pkt); | |
694 | mutex_unlock(&vsock->rx_lock); | |
695 | ||
696 | mutex_lock(&vsock->tx_lock); | |
697 | while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX]))) | |
698 | virtio_transport_free_pkt(pkt); | |
699 | mutex_unlock(&vsock->tx_lock); | |
700 | ||
701 | spin_lock_bh(&vsock->send_pkt_list_lock); | |
702 | while (!list_empty(&vsock->send_pkt_list)) { | |
703 | pkt = list_first_entry(&vsock->send_pkt_list, | |
704 | struct virtio_vsock_pkt, list); | |
705 | list_del(&pkt->list); | |
706 | virtio_transport_free_pkt(pkt); | |
707 | } | |
708 | spin_unlock_bh(&vsock->send_pkt_list_lock); | |
709 | ||
b917507e | 710 | /* Delete virtqueues and flush outstanding callbacks if any */ |
0ea9e1d3 AH |
711 | vdev->config->del_vqs(vdev); |
712 | ||
e226121f SG |
713 | /* Other works can be queued before 'config->del_vqs()', so we flush |
714 | * all works before to free the vsock object to avoid use after free. | |
715 | */ | |
e226121f SG |
716 | flush_work(&vsock->rx_work); |
717 | flush_work(&vsock->tx_work); | |
718 | flush_work(&vsock->event_work); | |
719 | flush_work(&vsock->send_pkt_work); | |
720 | ||
0deab087 SG |
721 | mutex_unlock(&the_virtio_vsock_mutex); |
722 | ||
0ea9e1d3 AH |
723 | kfree(vsock); |
724 | } | |
725 | ||
726 | static struct virtio_device_id id_table[] = { | |
727 | { VIRTIO_ID_VSOCK, VIRTIO_DEV_ANY_ID }, | |
728 | { 0 }, | |
729 | }; | |
730 | ||
731 | static unsigned int features[] = { | |
53efbba1 | 732 | VIRTIO_VSOCK_F_SEQPACKET |
0ea9e1d3 AH |
733 | }; |
734 | ||
735 | static struct virtio_driver virtio_vsock_driver = { | |
736 | .feature_table = features, | |
737 | .feature_table_size = ARRAY_SIZE(features), | |
738 | .driver.name = KBUILD_MODNAME, | |
739 | .driver.owner = THIS_MODULE, | |
740 | .id_table = id_table, | |
741 | .probe = virtio_vsock_probe, | |
742 | .remove = virtio_vsock_remove, | |
743 | }; | |
744 | ||
745 | static int __init virtio_vsock_init(void) | |
746 | { | |
747 | int ret; | |
748 | ||
749 | virtio_vsock_workqueue = alloc_workqueue("virtio_vsock", 0, 0); | |
750 | if (!virtio_vsock_workqueue) | |
751 | return -ENOMEM; | |
22b5c0b6 | 752 | |
c0cfa2d8 SG |
753 | ret = vsock_core_register(&virtio_transport.transport, |
754 | VSOCK_TRANSPORT_F_G2H); | |
0ea9e1d3 | 755 | if (ret) |
22b5c0b6 SG |
756 | goto out_wq; |
757 | ||
ba95e5df | 758 | ret = register_virtio_driver(&virtio_vsock_driver); |
22b5c0b6 | 759 | if (ret) |
ba95e5df | 760 | goto out_vci; |
22b5c0b6 SG |
761 | |
762 | return 0; | |
763 | ||
ba95e5df | 764 | out_vci: |
c0cfa2d8 | 765 | vsock_core_unregister(&virtio_transport.transport); |
22b5c0b6 SG |
766 | out_wq: |
767 | destroy_workqueue(virtio_vsock_workqueue); | |
0ea9e1d3 AH |
768 | return ret; |
769 | } | |
770 | ||
771 | static void __exit virtio_vsock_exit(void) | |
772 | { | |
773 | unregister_virtio_driver(&virtio_vsock_driver); | |
c0cfa2d8 | 774 | vsock_core_unregister(&virtio_transport.transport); |
0ea9e1d3 AH |
775 | destroy_workqueue(virtio_vsock_workqueue); |
776 | } | |
777 | ||
778 | module_init(virtio_vsock_init); | |
779 | module_exit(virtio_vsock_exit); | |
780 | MODULE_LICENSE("GPL v2"); | |
781 | MODULE_AUTHOR("Asias He"); | |
782 | MODULE_DESCRIPTION("virtio transport for vsock"); | |
783 | MODULE_DEVICE_TABLE(virtio, id_table); |