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 { | |
e13a6915 | 33 | VHOST_VSOCK_FEATURES = VHOST_FEATURES | |
ced7b713 AK |
34 | (1ULL << VIRTIO_F_ACCESS_PLATFORM) | |
35 | (1ULL << VIRTIO_VSOCK_F_SEQPACKET) | |
e13a6915 SG |
36 | }; |
37 | ||
38 | enum { | |
39 | VHOST_VSOCK_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2) | |
433fc58e AH |
40 | }; |
41 | ||
42 | /* Used to track all the vhost_vsock instances on the system. */ | |
6db3d8dc | 43 | static DEFINE_MUTEX(vhost_vsock_mutex); |
834e772c | 44 | static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8); |
433fc58e AH |
45 | |
46 | struct vhost_vsock { | |
47 | struct vhost_dev dev; | |
48 | struct vhost_virtqueue vqs[2]; | |
49 | ||
6db3d8dc | 50 | /* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */ |
834e772c | 51 | struct hlist_node hash; |
433fc58e AH |
52 | |
53 | struct vhost_work send_pkt_work; | |
71dc9ec9 | 54 | struct sk_buff_head send_pkt_queue; /* host->guest pending packets */ |
433fc58e AH |
55 | |
56 | atomic_t queued_replies; | |
57 | ||
58 | u32 guest_cid; | |
ced7b713 | 59 | bool seqpacket_allow; |
433fc58e AH |
60 | }; |
61 | ||
62 | static u32 vhost_transport_get_local_cid(void) | |
63 | { | |
64 | return VHOST_VSOCK_DEFAULT_HOST_CID; | |
65 | } | |
66 | ||
6db3d8dc | 67 | /* Callers that dereference the return value must hold vhost_vsock_mutex or the |
834e772c SH |
68 | * RCU read lock. |
69 | */ | |
70 | static struct vhost_vsock *vhost_vsock_get(u32 guest_cid) | |
433fc58e AH |
71 | { |
72 | struct vhost_vsock *vsock; | |
73 | ||
834e772c | 74 | hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) { |
433fc58e AH |
75 | u32 other_cid = vsock->guest_cid; |
76 | ||
77 | /* Skip instances that have no CID yet */ | |
78 | if (other_cid == 0) | |
79 | continue; | |
80 | ||
ff3c1b1a | 81 | if (other_cid == guest_cid) |
433fc58e | 82 | return vsock; |
ff3c1b1a | 83 | |
433fc58e | 84 | } |
433fc58e AH |
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]; | |
e79b431f | 94 | int pkts = 0, total_len = 0; |
433fc58e AH |
95 | bool added = false; |
96 | bool restart_tx = false; | |
97 | ||
98 | mutex_lock(&vq->mutex); | |
99 | ||
247643f8 | 100 | if (!vhost_vq_get_backend(vq)) |
433fc58e AH |
101 | goto out; |
102 | ||
e13a6915 SG |
103 | if (!vq_meta_prefetch(vq)) |
104 | goto out; | |
105 | ||
433fc58e AH |
106 | /* Avoid further vmexits, we're already processing the virtqueue */ |
107 | vhost_disable_notify(&vsock->dev, vq); | |
108 | ||
e79b431f | 109 | do { |
71dc9ec9 BE |
110 | struct virtio_vsock_hdr *hdr; |
111 | size_t iov_len, payload_len; | |
433fc58e | 112 | struct iov_iter iov_iter; |
71dc9ec9 BE |
113 | u32 flags_to_restore = 0; |
114 | struct sk_buff *skb; | |
433fc58e AH |
115 | unsigned out, in; |
116 | size_t nbytes; | |
0df7cd3c | 117 | u32 offset; |
433fc58e AH |
118 | int head; |
119 | ||
71dc9ec9 BE |
120 | skb = virtio_vsock_skb_dequeue(&vsock->send_pkt_queue); |
121 | ||
122 | if (!skb) { | |
433fc58e AH |
123 | vhost_enable_notify(&vsock->dev, vq); |
124 | break; | |
125 | } | |
126 | ||
433fc58e AH |
127 | head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), |
128 | &out, &in, NULL, NULL); | |
129 | if (head < 0) { | |
71dc9ec9 | 130 | virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb); |
433fc58e AH |
131 | break; |
132 | } | |
133 | ||
134 | if (head == vq->num) { | |
71dc9ec9 | 135 | virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb); |
433fc58e AH |
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) { | |
71dc9ec9 | 147 | kfree_skb(skb); |
433fc58e AH |
148 | vq_err(vq, "Expected 0 output buffers, got %u\n", out); |
149 | break; | |
150 | } | |
151 | ||
6dbd3e66 | 152 | iov_len = iov_length(&vq->iov[out], in); |
71dc9ec9 BE |
153 | if (iov_len < sizeof(*hdr)) { |
154 | kfree_skb(skb); | |
6dbd3e66 SG |
155 | vq_err(vq, "Buffer len [%zu] too small\n", iov_len); |
156 | break; | |
157 | } | |
158 | ||
de4eda9d | 159 | iov_iter_init(&iov_iter, ITER_DEST, &vq->iov[out], in, iov_len); |
0df7cd3c AK |
160 | offset = VIRTIO_VSOCK_SKB_CB(skb)->offset; |
161 | payload_len = skb->len - offset; | |
71dc9ec9 | 162 | hdr = virtio_vsock_hdr(skb); |
6dbd3e66 SG |
163 | |
164 | /* If the packet is greater than the space available in the | |
165 | * buffer, we split it using multiple buffers. | |
166 | */ | |
71dc9ec9 BE |
167 | if (payload_len > iov_len - sizeof(*hdr)) { |
168 | payload_len = iov_len - sizeof(*hdr); | |
6dbd3e66 | 169 | |
ced7b713 AK |
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 | |
9af8f106 | 174 | * SOCK_SEQPACKET, we also must clear message delimeter |
1af7e555 AK |
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. | |
ced7b713 | 180 | */ |
71dc9ec9 BE |
181 | if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { |
182 | hdr->flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM); | |
1af7e555 AK |
183 | flags_to_restore |= VIRTIO_VSOCK_SEQ_EOM; |
184 | ||
71dc9ec9 BE |
185 | if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) { |
186 | hdr->flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR); | |
1af7e555 AK |
187 | flags_to_restore |= VIRTIO_VSOCK_SEQ_EOR; |
188 | } | |
ced7b713 AK |
189 | } |
190 | } | |
191 | ||
6dbd3e66 | 192 | /* Set the correct length in the header */ |
71dc9ec9 | 193 | hdr->len = cpu_to_le32(payload_len); |
433fc58e | 194 | |
71dc9ec9 BE |
195 | nbytes = copy_to_iter(hdr, sizeof(*hdr), &iov_iter); |
196 | if (nbytes != sizeof(*hdr)) { | |
197 | kfree_skb(skb); | |
433fc58e AH |
198 | vq_err(vq, "Faulted on copying pkt hdr\n"); |
199 | break; | |
200 | } | |
201 | ||
0df7cd3c AK |
202 | if (skb_copy_datagram_iter(skb, |
203 | offset, | |
204 | &iov_iter, | |
205 | payload_len)) { | |
71dc9ec9 | 206 | kfree_skb(skb); |
433fc58e AH |
207 | vq_err(vq, "Faulted on copying pkt buf\n"); |
208 | break; | |
209 | } | |
210 | ||
107bc076 SG |
211 | /* Deliver to monitoring devices all packets that we |
212 | * will transmit. | |
82dfb540 | 213 | */ |
71dc9ec9 | 214 | virtio_transport_deliver_tap_pkt(skb); |
82dfb540 | 215 | |
71dc9ec9 | 216 | vhost_add_used(vq, head, sizeof(*hdr) + payload_len); |
107bc076 SG |
217 | added = true; |
218 | ||
0df7cd3c | 219 | VIRTIO_VSOCK_SKB_CB(skb)->offset += payload_len; |
6dbd3e66 SG |
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 | */ | |
0df7cd3c | 225 | if (VIRTIO_VSOCK_SKB_CB(skb)->offset < skb->len) { |
71dc9ec9 | 226 | hdr->flags |= cpu_to_le32(flags_to_restore); |
ced7b713 | 227 | |
71dc9ec9 | 228 | /* We are queueing the same skb to handle |
a78d1639 SG |
229 | * the remaining bytes, and we want to deliver it |
230 | * to monitoring devices in the next iteration. | |
231 | */ | |
71dc9ec9 BE |
232 | virtio_vsock_skb_clear_tap_delivered(skb); |
233 | virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb); | |
6dbd3e66 | 234 | } else { |
71dc9ec9 | 235 | if (virtio_vsock_skb_reply(skb)) { |
6dbd3e66 SG |
236 | int val; |
237 | ||
238 | val = atomic_dec_return(&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 | ||
71dc9ec9 | 247 | consume_skb(skb); |
6dbd3e66 | 248 | } |
e79b431f | 249 | } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); |
433fc58e AH |
250 | if (added) |
251 | vhost_signal(&vsock->dev, vq); | |
252 | ||
253 | out: | |
254 | mutex_unlock(&vq->mutex); | |
255 | ||
256 | if (restart_tx) | |
257 | vhost_poll_queue(&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 | |
71dc9ec9 | 272 | vhost_transport_send_pkt(struct sk_buff *skb) |
433fc58e | 273 | { |
71dc9ec9 | 274 | struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); |
433fc58e | 275 | struct vhost_vsock *vsock; |
71dc9ec9 | 276 | int len = skb->len; |
433fc58e | 277 | |
834e772c SH |
278 | rcu_read_lock(); |
279 | ||
433fc58e | 280 | /* Find the vhost_vsock according to guest context id */ |
71dc9ec9 | 281 | vsock = vhost_vsock_get(le64_to_cpu(hdr->dst_cid)); |
433fc58e | 282 | if (!vsock) { |
834e772c | 283 | rcu_read_unlock(); |
71dc9ec9 | 284 | kfree_skb(skb); |
433fc58e AH |
285 | return -ENODEV; |
286 | } | |
287 | ||
71dc9ec9 | 288 | if (virtio_vsock_skb_reply(skb)) |
433fc58e AH |
289 | atomic_inc(&vsock->queued_replies); |
290 | ||
71dc9ec9 | 291 | virtio_vsock_skb_queue_tail(&vsock->send_pkt_queue, skb); |
9e09d0ec | 292 | vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &vsock->send_pkt_work); |
834e772c SH |
293 | |
294 | rcu_read_unlock(); | |
433fc58e AH |
295 | return len; |
296 | } | |
297 | ||
16320f36 PT |
298 | static int |
299 | vhost_transport_cancel_pkt(struct vsock_sock *vsk) | |
300 | { | |
301 | struct vhost_vsock *vsock; | |
16320f36 | 302 | int cnt = 0; |
834e772c | 303 | int ret = -ENODEV; |
16320f36 | 304 | |
834e772c SH |
305 | rcu_read_lock(); |
306 | ||
16320f36 PT |
307 | /* Find the vhost_vsock according to guest context id */ |
308 | vsock = vhost_vsock_get(vsk->remote_addr.svm_cid); | |
309 | if (!vsock) | |
834e772c | 310 | goto out; |
16320f36 | 311 | |
71dc9ec9 | 312 | cnt = virtio_transport_purge_skbs(vsk, &vsock->send_pkt_queue); |
16320f36 PT |
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(cnt, &vsock->queued_replies); | |
319 | if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num) | |
320 | vhost_poll_queue(&tx_vq->poll); | |
321 | } | |
322 | ||
834e772c SH |
323 | ret = 0; |
324 | out: | |
325 | rcu_read_unlock(); | |
326 | return ret; | |
16320f36 PT |
327 | } |
328 | ||
71dc9ec9 BE |
329 | static struct sk_buff * |
330 | vhost_vsock_alloc_skb(struct vhost_virtqueue *vq, | |
433fc58e AH |
331 | unsigned int out, unsigned int in) |
332 | { | |
71dc9ec9 | 333 | struct virtio_vsock_hdr *hdr; |
433fc58e | 334 | struct iov_iter iov_iter; |
71dc9ec9 BE |
335 | struct sk_buff *skb; |
336 | size_t payload_len; | |
433fc58e AH |
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 | ||
71dc9ec9 BE |
345 | len = iov_length(vq->iov, out); |
346 | ||
347 | /* len contains both payload and hdr */ | |
348 | skb = virtio_vsock_alloc_skb(len, GFP_KERNEL); | |
349 | if (!skb) | |
433fc58e AH |
350 | return NULL; |
351 | ||
de4eda9d | 352 | iov_iter_init(&iov_iter, ITER_SOURCE, vq->iov, out, len); |
433fc58e | 353 | |
71dc9ec9 BE |
354 | hdr = virtio_vsock_hdr(skb); |
355 | nbytes = copy_from_iter(hdr, sizeof(*hdr), &iov_iter); | |
356 | if (nbytes != sizeof(*hdr)) { | |
433fc58e | 357 | vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n", |
71dc9ec9 BE |
358 | sizeof(*hdr), nbytes); |
359 | kfree_skb(skb); | |
433fc58e AH |
360 | return NULL; |
361 | } | |
362 | ||
71dc9ec9 | 363 | payload_len = le32_to_cpu(hdr->len); |
433fc58e AH |
364 | |
365 | /* No payload */ | |
71dc9ec9 BE |
366 | if (!payload_len) |
367 | return skb; | |
433fc58e | 368 | |
71dc9ec9 BE |
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); | |
433fc58e AH |
373 | return NULL; |
374 | } | |
375 | ||
71dc9ec9 | 376 | virtio_vsock_skb_rx_put(skb); |
433fc58e | 377 | |
71dc9ec9 BE |
378 | nbytes = copy_from_iter(skb->data, payload_len, &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); | |
433fc58e AH |
383 | return NULL; |
384 | } | |
385 | ||
71dc9ec9 | 386 | return skb; |
433fc58e AH |
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(&vsock->queued_replies); | |
397 | ||
398 | return val < vq->num; | |
399 | } | |
400 | ||
ced7b713 AK |
401 | static bool vhost_transport_seqpacket_allow(u32 remote_cid); |
402 | ||
4c7246dc SG |
403 | static struct virtio_transport vhost_transport = { |
404 | .transport = { | |
6a2c0962 SG |
405 | .module = THIS_MODULE, |
406 | ||
4c7246dc SG |
407 | .get_local_cid = vhost_transport_get_local_cid, |
408 | ||
409 | .init = virtio_transport_do_socket_init, | |
410 | .destruct = virtio_transport_destruct, | |
411 | .release = virtio_transport_release, | |
412 | .connect = virtio_transport_connect, | |
413 | .shutdown = virtio_transport_shutdown, | |
414 | .cancel_pkt = vhost_transport_cancel_pkt, | |
415 | ||
416 | .dgram_enqueue = virtio_transport_dgram_enqueue, | |
417 | .dgram_dequeue = virtio_transport_dgram_dequeue, | |
418 | .dgram_bind = virtio_transport_dgram_bind, | |
419 | .dgram_allow = virtio_transport_dgram_allow, | |
420 | ||
421 | .stream_enqueue = virtio_transport_stream_enqueue, | |
422 | .stream_dequeue = virtio_transport_stream_dequeue, | |
423 | .stream_has_data = virtio_transport_stream_has_data, | |
424 | .stream_has_space = virtio_transport_stream_has_space, | |
425 | .stream_rcvhiwat = virtio_transport_stream_rcvhiwat, | |
426 | .stream_is_active = virtio_transport_stream_is_active, | |
427 | .stream_allow = virtio_transport_stream_allow, | |
428 | ||
ced7b713 AK |
429 | .seqpacket_dequeue = virtio_transport_seqpacket_dequeue, |
430 | .seqpacket_enqueue = virtio_transport_seqpacket_enqueue, | |
431 | .seqpacket_allow = vhost_transport_seqpacket_allow, | |
432 | .seqpacket_has_data = virtio_transport_seqpacket_has_data, | |
433 | ||
4c7246dc SG |
434 | .notify_poll_in = virtio_transport_notify_poll_in, |
435 | .notify_poll_out = virtio_transport_notify_poll_out, | |
436 | .notify_recv_init = virtio_transport_notify_recv_init, | |
437 | .notify_recv_pre_block = virtio_transport_notify_recv_pre_block, | |
438 | .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue, | |
439 | .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue, | |
440 | .notify_send_init = virtio_transport_notify_send_init, | |
441 | .notify_send_pre_block = virtio_transport_notify_send_pre_block, | |
442 | .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue, | |
443 | .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue, | |
b9f2b0ff | 444 | .notify_buffer_size = virtio_transport_notify_buffer_size, |
4c7246dc | 445 | |
634f1a71 | 446 | .read_skb = virtio_transport_read_skb, |
4c7246dc SG |
447 | }, |
448 | ||
449 | .send_pkt = vhost_transport_send_pkt, | |
450 | }; | |
451 | ||
ced7b713 AK |
452 | static bool vhost_transport_seqpacket_allow(u32 remote_cid) |
453 | { | |
454 | struct vhost_vsock *vsock; | |
455 | bool seqpacket_allow = false; | |
456 | ||
457 | rcu_read_lock(); | |
458 | vsock = vhost_vsock_get(remote_cid); | |
459 | ||
460 | if (vsock) | |
461 | seqpacket_allow = vsock->seqpacket_allow; | |
462 | ||
463 | rcu_read_unlock(); | |
464 | ||
465 | return seqpacket_allow; | |
466 | } | |
467 | ||
433fc58e AH |
468 | static void vhost_vsock_handle_tx_kick(struct vhost_work *work) |
469 | { | |
470 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
471 | poll.work); | |
472 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
473 | dev); | |
e79b431f | 474 | int head, pkts = 0, total_len = 0; |
433fc58e | 475 | unsigned int out, in; |
71dc9ec9 | 476 | struct sk_buff *skb; |
433fc58e AH |
477 | bool added = false; |
478 | ||
479 | mutex_lock(&vq->mutex); | |
480 | ||
247643f8 | 481 | if (!vhost_vq_get_backend(vq)) |
433fc58e AH |
482 | goto out; |
483 | ||
e13a6915 SG |
484 | if (!vq_meta_prefetch(vq)) |
485 | goto out; | |
486 | ||
433fc58e | 487 | vhost_disable_notify(&vsock->dev, vq); |
e79b431f | 488 | do { |
71dc9ec9 BE |
489 | struct virtio_vsock_hdr *hdr; |
490 | ||
433fc58e AH |
491 | if (!vhost_vsock_more_replies(vsock)) { |
492 | /* Stop tx until the device processes already | |
493 | * pending replies. Leave tx virtqueue | |
494 | * callbacks disabled. | |
495 | */ | |
496 | goto no_more_replies; | |
497 | } | |
498 | ||
499 | head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), | |
500 | &out, &in, NULL, NULL); | |
501 | if (head < 0) | |
502 | break; | |
503 | ||
504 | if (head == vq->num) { | |
505 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { | |
506 | vhost_disable_notify(&vsock->dev, vq); | |
507 | continue; | |
508 | } | |
509 | break; | |
510 | } | |
511 | ||
71dc9ec9 BE |
512 | skb = vhost_vsock_alloc_skb(vq, out, in); |
513 | if (!skb) { | |
433fc58e AH |
514 | vq_err(vq, "Faulted on pkt\n"); |
515 | continue; | |
516 | } | |
517 | ||
71dc9ec9 | 518 | total_len += sizeof(*hdr) + skb->len; |
3fda5d6e | 519 | |
82dfb540 | 520 | /* Deliver to monitoring devices all received packets */ |
71dc9ec9 BE |
521 | virtio_transport_deliver_tap_pkt(skb); |
522 | ||
523 | hdr = virtio_vsock_hdr(skb); | |
82dfb540 | 524 | |
433fc58e | 525 | /* Only accept correctly addressed packets */ |
71dc9ec9 BE |
526 | if (le64_to_cpu(hdr->src_cid) == vsock->guest_cid && |
527 | le64_to_cpu(hdr->dst_cid) == | |
8a3cc29c | 528 | vhost_transport_get_local_cid()) |
71dc9ec9 | 529 | virtio_transport_recv_pkt(&vhost_transport, skb); |
433fc58e | 530 | else |
71dc9ec9 | 531 | kfree_skb(skb); |
433fc58e | 532 | |
49d8c5ff | 533 | vhost_add_used(vq, head, 0); |
433fc58e | 534 | added = true; |
e79b431f | 535 | } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); |
433fc58e AH |
536 | |
537 | no_more_replies: | |
538 | if (added) | |
539 | vhost_signal(&vsock->dev, vq); | |
540 | ||
541 | out: | |
542 | mutex_unlock(&vq->mutex); | |
543 | } | |
544 | ||
545 | static void vhost_vsock_handle_rx_kick(struct vhost_work *work) | |
546 | { | |
547 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
548 | poll.work); | |
549 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, | |
550 | dev); | |
551 | ||
552 | vhost_transport_do_send_pkt(vsock, vq); | |
553 | } | |
554 | ||
555 | static int vhost_vsock_start(struct vhost_vsock *vsock) | |
556 | { | |
0516ffd8 | 557 | struct vhost_virtqueue *vq; |
433fc58e AH |
558 | size_t i; |
559 | int ret; | |
560 | ||
561 | mutex_lock(&vsock->dev.mutex); | |
562 | ||
563 | ret = vhost_dev_check_owner(&vsock->dev); | |
564 | if (ret) | |
565 | goto err; | |
566 | ||
567 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
0516ffd8 | 568 | vq = &vsock->vqs[i]; |
433fc58e AH |
569 | |
570 | mutex_lock(&vq->mutex); | |
571 | ||
572 | if (!vhost_vq_access_ok(vq)) { | |
573 | ret = -EFAULT; | |
433fc58e AH |
574 | goto err_vq; |
575 | } | |
576 | ||
247643f8 EP |
577 | if (!vhost_vq_get_backend(vq)) { |
578 | vhost_vq_set_backend(vq, vsock); | |
0516ffd8 SH |
579 | ret = vhost_vq_init_access(vq); |
580 | if (ret) | |
581 | goto err_vq; | |
433fc58e AH |
582 | } |
583 | ||
584 | mutex_unlock(&vq->mutex); | |
585 | } | |
586 | ||
0b841030 JH |
587 | /* Some packets may have been queued before the device was started, |
588 | * let's kick the send worker to send them. | |
589 | */ | |
9e09d0ec | 590 | vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &vsock->send_pkt_work); |
0b841030 | 591 | |
433fc58e AH |
592 | mutex_unlock(&vsock->dev.mutex); |
593 | return 0; | |
594 | ||
595 | err_vq: | |
247643f8 | 596 | vhost_vq_set_backend(vq, NULL); |
0516ffd8 SH |
597 | mutex_unlock(&vq->mutex); |
598 | ||
433fc58e | 599 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
0516ffd8 | 600 | vq = &vsock->vqs[i]; |
433fc58e AH |
601 | |
602 | mutex_lock(&vq->mutex); | |
247643f8 | 603 | vhost_vq_set_backend(vq, NULL); |
433fc58e AH |
604 | mutex_unlock(&vq->mutex); |
605 | } | |
606 | err: | |
607 | mutex_unlock(&vsock->dev.mutex); | |
608 | return ret; | |
609 | } | |
610 | ||
a58da53f | 611 | static int vhost_vsock_stop(struct vhost_vsock *vsock, bool check_owner) |
433fc58e AH |
612 | { |
613 | size_t i; | |
a58da53f | 614 | int ret = 0; |
433fc58e AH |
615 | |
616 | mutex_lock(&vsock->dev.mutex); | |
617 | ||
a58da53f SG |
618 | if (check_owner) { |
619 | ret = vhost_dev_check_owner(&vsock->dev); | |
620 | if (ret) | |
621 | goto err; | |
622 | } | |
433fc58e AH |
623 | |
624 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { | |
625 | struct vhost_virtqueue *vq = &vsock->vqs[i]; | |
626 | ||
627 | mutex_lock(&vq->mutex); | |
247643f8 | 628 | vhost_vq_set_backend(vq, NULL); |
433fc58e AH |
629 | mutex_unlock(&vq->mutex); |
630 | } | |
631 | ||
632 | err: | |
633 | mutex_unlock(&vsock->dev.mutex); | |
634 | return ret; | |
635 | } | |
636 | ||
637 | static void vhost_vsock_free(struct vhost_vsock *vsock) | |
638 | { | |
b226acab | 639 | kvfree(vsock); |
433fc58e AH |
640 | } |
641 | ||
642 | static int vhost_vsock_dev_open(struct inode *inode, struct file *file) | |
643 | { | |
644 | struct vhost_virtqueue **vqs; | |
645 | struct vhost_vsock *vsock; | |
646 | int ret; | |
647 | ||
648 | /* This struct is large and allocation could fail, fall back to vmalloc | |
649 | * if there is no other way. | |
650 | */ | |
dcda9b04 | 651 | vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
6c5ab651 MH |
652 | if (!vsock) |
653 | return -ENOMEM; | |
433fc58e AH |
654 | |
655 | vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL); | |
656 | if (!vqs) { | |
657 | ret = -ENOMEM; | |
658 | goto out; | |
659 | } | |
660 | ||
a72b69dc SH |
661 | vsock->guest_cid = 0; /* no CID assigned yet */ |
662 | ||
433fc58e AH |
663 | atomic_set(&vsock->queued_replies, 0); |
664 | ||
665 | vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX]; | |
666 | vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX]; | |
667 | vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick; | |
668 | vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick; | |
669 | ||
e82b9b07 JW |
670 | vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs), |
671 | UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT, | |
01fcb1cb | 672 | VHOST_VSOCK_WEIGHT, true, NULL); |
433fc58e AH |
673 | |
674 | file->private_data = vsock; | |
71dc9ec9 | 675 | skb_queue_head_init(&vsock->send_pkt_queue); |
433fc58e | 676 | vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work); |
433fc58e AH |
677 | return 0; |
678 | ||
679 | out: | |
680 | vhost_vsock_free(vsock); | |
681 | return ret; | |
682 | } | |
683 | ||
684 | static void vhost_vsock_flush(struct vhost_vsock *vsock) | |
685 | { | |
b2ffa407 | 686 | vhost_dev_flush(&vsock->dev); |
433fc58e AH |
687 | } |
688 | ||
689 | static void vhost_vsock_reset_orphans(struct sock *sk) | |
690 | { | |
691 | struct vsock_sock *vsk = vsock_sk(sk); | |
692 | ||
693 | /* vmci_transport.c doesn't take sk_lock here either. At least we're | |
694 | * under vsock_table_lock so the sock cannot disappear while we're | |
695 | * executing. | |
696 | */ | |
697 | ||
c38f57da SH |
698 | /* If the peer is still valid, no need to reset connection */ |
699 | if (vhost_vsock_get(vsk->remote_addr.svm_cid)) | |
700 | return; | |
701 | ||
702 | /* If the close timeout is pending, let it expire. This avoids races | |
703 | * with the timeout callback. | |
704 | */ | |
705 | if (vsk->close_work_scheduled) | |
706 | return; | |
707 | ||
708 | sock_set_flag(sk, SOCK_DONE); | |
709 | vsk->peer_shutdown = SHUTDOWN_MASK; | |
710 | sk->sk_state = SS_UNCONNECTED; | |
711 | sk->sk_err = ECONNRESET; | |
e3ae2365 | 712 | sk_error_report(sk); |
433fc58e AH |
713 | } |
714 | ||
715 | static int vhost_vsock_dev_release(struct inode *inode, struct file *file) | |
716 | { | |
717 | struct vhost_vsock *vsock = file->private_data; | |
718 | ||
6db3d8dc | 719 | mutex_lock(&vhost_vsock_mutex); |
834e772c SH |
720 | if (vsock->guest_cid) |
721 | hash_del_rcu(&vsock->hash); | |
6db3d8dc | 722 | mutex_unlock(&vhost_vsock_mutex); |
433fc58e | 723 | |
834e772c SH |
724 | /* Wait for other CPUs to finish using vsock */ |
725 | synchronize_rcu(); | |
726 | ||
433fc58e AH |
727 | /* Iterating over all connections for all CIDs to find orphans is |
728 | * inefficient. Room for improvement here. */ | |
8e6ed963 JP |
729 | vsock_for_each_connected_socket(&vhost_transport.transport, |
730 | vhost_vsock_reset_orphans); | |
433fc58e | 731 | |
a58da53f SG |
732 | /* Don't check the owner, because we are in the release path, so we |
733 | * need to stop the vsock device in any case. | |
734 | * vhost_vsock_stop() can not fail in this case, so we don't need to | |
735 | * check the return code. | |
736 | */ | |
737 | vhost_vsock_stop(vsock, false); | |
433fc58e AH |
738 | vhost_vsock_flush(vsock); |
739 | vhost_dev_stop(&vsock->dev); | |
740 | ||
71dc9ec9 | 741 | virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue); |
433fc58e | 742 | |
f6f93f75 | 743 | vhost_dev_cleanup(&vsock->dev); |
433fc58e AH |
744 | kfree(vsock->dev.vqs); |
745 | vhost_vsock_free(vsock); | |
746 | return 0; | |
747 | } | |
748 | ||
749 | static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid) | |
750 | { | |
751 | struct vhost_vsock *other; | |
752 | ||
753 | /* Refuse reserved CIDs */ | |
754 | if (guest_cid <= VMADDR_CID_HOST || | |
755 | guest_cid == U32_MAX) | |
756 | return -EINVAL; | |
757 | ||
758 | /* 64-bit CIDs are not yet supported */ | |
759 | if (guest_cid > U32_MAX) | |
760 | return -EINVAL; | |
761 | ||
ed8640a9 SG |
762 | /* Refuse if CID is assigned to the guest->host transport (i.e. nested |
763 | * VM), to make the loopback work. | |
764 | */ | |
765 | if (vsock_find_cid(guest_cid)) | |
766 | return -EADDRINUSE; | |
767 | ||
433fc58e | 768 | /* Refuse if CID is already in use */ |
6db3d8dc | 769 | mutex_lock(&vhost_vsock_mutex); |
834e772c | 770 | other = vhost_vsock_get(guest_cid); |
6c083c2b | 771 | if (other && other != vsock) { |
6db3d8dc | 772 | mutex_unlock(&vhost_vsock_mutex); |
6c083c2b G |
773 | return -EADDRINUSE; |
774 | } | |
834e772c SH |
775 | |
776 | if (vsock->guest_cid) | |
777 | hash_del_rcu(&vsock->hash); | |
778 | ||
433fc58e | 779 | vsock->guest_cid = guest_cid; |
7fbe078c | 780 | hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid); |
6db3d8dc | 781 | mutex_unlock(&vhost_vsock_mutex); |
433fc58e AH |
782 | |
783 | return 0; | |
784 | } | |
785 | ||
786 | static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features) | |
787 | { | |
788 | struct vhost_virtqueue *vq; | |
789 | int i; | |
790 | ||
791 | if (features & ~VHOST_VSOCK_FEATURES) | |
792 | return -EOPNOTSUPP; | |
793 | ||
794 | mutex_lock(&vsock->dev.mutex); | |
795 | if ((features & (1 << VHOST_F_LOG_ALL)) && | |
796 | !vhost_log_access_ok(&vsock->dev)) { | |
e13a6915 SG |
797 | goto err; |
798 | } | |
799 | ||
800 | if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) { | |
759aba1e | 801 | if (vhost_init_device_iotlb(&vsock->dev)) |
e13a6915 | 802 | goto err; |
433fc58e AH |
803 | } |
804 | ||
ced7b713 AK |
805 | if (features & (1ULL << VIRTIO_VSOCK_F_SEQPACKET)) |
806 | vsock->seqpacket_allow = true; | |
807 | ||
433fc58e AH |
808 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
809 | vq = &vsock->vqs[i]; | |
810 | mutex_lock(&vq->mutex); | |
811 | vq->acked_features = features; | |
812 | mutex_unlock(&vq->mutex); | |
813 | } | |
814 | mutex_unlock(&vsock->dev.mutex); | |
815 | return 0; | |
e13a6915 SG |
816 | |
817 | err: | |
818 | mutex_unlock(&vsock->dev.mutex); | |
819 | return -EFAULT; | |
433fc58e AH |
820 | } |
821 | ||
822 | static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl, | |
823 | unsigned long arg) | |
824 | { | |
825 | struct vhost_vsock *vsock = f->private_data; | |
826 | void __user *argp = (void __user *)arg; | |
827 | u64 guest_cid; | |
828 | u64 features; | |
829 | int start; | |
830 | int r; | |
831 | ||
832 | switch (ioctl) { | |
833 | case VHOST_VSOCK_SET_GUEST_CID: | |
834 | if (copy_from_user(&guest_cid, argp, sizeof(guest_cid))) | |
835 | return -EFAULT; | |
836 | return vhost_vsock_set_cid(vsock, guest_cid); | |
837 | case VHOST_VSOCK_SET_RUNNING: | |
838 | if (copy_from_user(&start, argp, sizeof(start))) | |
839 | return -EFAULT; | |
840 | if (start) | |
841 | return vhost_vsock_start(vsock); | |
842 | else | |
a58da53f | 843 | return vhost_vsock_stop(vsock, true); |
433fc58e AH |
844 | case VHOST_GET_FEATURES: |
845 | features = VHOST_VSOCK_FEATURES; | |
846 | if (copy_to_user(argp, &features, sizeof(features))) | |
847 | return -EFAULT; | |
848 | return 0; | |
849 | case VHOST_SET_FEATURES: | |
850 | if (copy_from_user(&features, argp, sizeof(features))) | |
851 | return -EFAULT; | |
852 | return vhost_vsock_set_features(vsock, features); | |
e13a6915 SG |
853 | case VHOST_GET_BACKEND_FEATURES: |
854 | features = VHOST_VSOCK_BACKEND_FEATURES; | |
855 | if (copy_to_user(argp, &features, sizeof(features))) | |
856 | return -EFAULT; | |
857 | return 0; | |
858 | case VHOST_SET_BACKEND_FEATURES: | |
859 | if (copy_from_user(&features, argp, sizeof(features))) | |
860 | return -EFAULT; | |
861 | if (features & ~VHOST_VSOCK_BACKEND_FEATURES) | |
862 | return -EOPNOTSUPP; | |
863 | vhost_set_backend_features(&vsock->dev, features); | |
864 | return 0; | |
433fc58e AH |
865 | default: |
866 | mutex_lock(&vsock->dev.mutex); | |
867 | r = vhost_dev_ioctl(&vsock->dev, ioctl, argp); | |
868 | if (r == -ENOIOCTLCMD) | |
869 | r = vhost_vring_ioctl(&vsock->dev, ioctl, argp); | |
870 | else | |
871 | vhost_vsock_flush(vsock); | |
872 | mutex_unlock(&vsock->dev.mutex); | |
873 | return r; | |
874 | } | |
875 | } | |
876 | ||
e13a6915 SG |
877 | static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to) |
878 | { | |
879 | struct file *file = iocb->ki_filp; | |
880 | struct vhost_vsock *vsock = file->private_data; | |
881 | struct vhost_dev *dev = &vsock->dev; | |
882 | int noblock = file->f_flags & O_NONBLOCK; | |
883 | ||
884 | return vhost_chr_read_iter(dev, to, noblock); | |
885 | } | |
886 | ||
887 | static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb, | |
888 | struct iov_iter *from) | |
889 | { | |
890 | struct file *file = iocb->ki_filp; | |
891 | struct vhost_vsock *vsock = file->private_data; | |
892 | struct vhost_dev *dev = &vsock->dev; | |
893 | ||
894 | return vhost_chr_write_iter(dev, from); | |
895 | } | |
896 | ||
897 | static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait) | |
898 | { | |
899 | struct vhost_vsock *vsock = file->private_data; | |
900 | struct vhost_dev *dev = &vsock->dev; | |
901 | ||
902 | return vhost_chr_poll(file, dev, wait); | |
903 | } | |
904 | ||
433fc58e AH |
905 | static const struct file_operations vhost_vsock_fops = { |
906 | .owner = THIS_MODULE, | |
907 | .open = vhost_vsock_dev_open, | |
908 | .release = vhost_vsock_dev_release, | |
909 | .llseek = noop_llseek, | |
910 | .unlocked_ioctl = vhost_vsock_dev_ioctl, | |
407e9ef7 | 911 | .compat_ioctl = compat_ptr_ioctl, |
e13a6915 SG |
912 | .read_iter = vhost_vsock_chr_read_iter, |
913 | .write_iter = vhost_vsock_chr_write_iter, | |
914 | .poll = vhost_vsock_chr_poll, | |
433fc58e AH |
915 | }; |
916 | ||
917 | static struct miscdevice vhost_vsock_misc = { | |
f4660cc9 | 918 | .minor = VHOST_VSOCK_MINOR, |
433fc58e AH |
919 | .name = "vhost-vsock", |
920 | .fops = &vhost_vsock_fops, | |
921 | }; | |
922 | ||
433fc58e AH |
923 | static int __init vhost_vsock_init(void) |
924 | { | |
925 | int ret; | |
926 | ||
c0cfa2d8 SG |
927 | ret = vsock_core_register(&vhost_transport.transport, |
928 | VSOCK_TRANSPORT_F_H2G); | |
433fc58e AH |
929 | if (ret < 0) |
930 | return ret; | |
7a4efe18 YC |
931 | |
932 | ret = misc_register(&vhost_vsock_misc); | |
933 | if (ret) { | |
934 | vsock_core_unregister(&vhost_transport.transport); | |
935 | return ret; | |
936 | } | |
937 | ||
938 | return 0; | |
433fc58e AH |
939 | }; |
940 | ||
941 | static void __exit vhost_vsock_exit(void) | |
942 | { | |
943 | misc_deregister(&vhost_vsock_misc); | |
c0cfa2d8 | 944 | vsock_core_unregister(&vhost_transport.transport); |
433fc58e AH |
945 | }; |
946 | ||
947 | module_init(vhost_vsock_init); | |
948 | module_exit(vhost_vsock_exit); | |
949 | MODULE_LICENSE("GPL v2"); | |
950 | MODULE_AUTHOR("Asias He"); | |
951 | MODULE_DESCRIPTION("vhost transport for vsock "); | |
f4660cc9 SH |
952 | MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR); |
953 | MODULE_ALIAS("devname:vhost-vsock"); |