| 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(&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, vq->iov, ARRAY_SIZE(vq->iov), |
| 128 | &out, &in, NULL, NULL); |
| 129 | if (head < 0) { |
| 130 | virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb); |
| 131 | break; |
| 132 | } |
| 133 | |
| 134 | if (head == vq->num) { |
| 135 | virtio_vsock_skb_queue_head(&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(&vq->iov[out], 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(&iov_iter, ITER_DEST, &vq->iov[out], in, 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(hdr, sizeof(*hdr), &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(skb, |
| 203 | offset, |
| 204 | &iov_iter, |
| 205 | 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, 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(&vsock->send_pkt_queue, skb); |
| 234 | } else { |
| 235 | if (virtio_vsock_skb_reply(skb)) { |
| 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 | |
| 247 | virtio_transport_consume_skb_sent(skb, true); |
| 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(&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 |
| 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(&vsock->queued_replies); |
| 290 | |
| 291 | virtio_vsock_skb_queue_tail(&vsock->send_pkt_queue, skb); |
| 292 | vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &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(vsk->remote_addr.svm_cid); |
| 309 | if (!vsock) |
| 310 | goto out; |
| 311 | |
| 312 | cnt = virtio_transport_purge_skbs(vsk, &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(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 | |
| 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(vq->iov, out); |
| 346 | |
| 347 | /* len contains both payload and hdr */ |
| 348 | skb = virtio_vsock_alloc_skb(len, GFP_KERNEL); |
| 349 | if (!skb) |
| 350 | return NULL; |
| 351 | |
| 352 | iov_iter_init(&iov_iter, ITER_SOURCE, vq->iov, out, len); |
| 353 | |
| 354 | hdr = virtio_vsock_hdr(skb); |
| 355 | nbytes = copy_from_iter(hdr, sizeof(*hdr), &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(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); |
| 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(&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 | .notify_set_rcvlowat = virtio_transport_notify_set_rcvlowat, |
| 453 | |
| 454 | .unsent_bytes = virtio_transport_unsent_bytes, |
| 455 | |
| 456 | .read_skb = virtio_transport_read_skb, |
| 457 | }, |
| 458 | |
| 459 | .send_pkt = vhost_transport_send_pkt, |
| 460 | }; |
| 461 | |
| 462 | static bool vhost_transport_seqpacket_allow(u32 remote_cid) |
| 463 | { |
| 464 | struct vhost_vsock *vsock; |
| 465 | bool seqpacket_allow = false; |
| 466 | |
| 467 | rcu_read_lock(); |
| 468 | vsock = vhost_vsock_get(remote_cid); |
| 469 | |
| 470 | if (vsock) |
| 471 | seqpacket_allow = vsock->seqpacket_allow; |
| 472 | |
| 473 | rcu_read_unlock(); |
| 474 | |
| 475 | return seqpacket_allow; |
| 476 | } |
| 477 | |
| 478 | static void vhost_vsock_handle_tx_kick(struct vhost_work *work) |
| 479 | { |
| 480 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
| 481 | poll.work); |
| 482 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, |
| 483 | dev); |
| 484 | int head, pkts = 0, total_len = 0; |
| 485 | unsigned int out, in; |
| 486 | struct sk_buff *skb; |
| 487 | bool added = false; |
| 488 | |
| 489 | mutex_lock(&vq->mutex); |
| 490 | |
| 491 | if (!vhost_vq_get_backend(vq)) |
| 492 | goto out; |
| 493 | |
| 494 | if (!vq_meta_prefetch(vq)) |
| 495 | goto out; |
| 496 | |
| 497 | vhost_disable_notify(&vsock->dev, vq); |
| 498 | do { |
| 499 | struct virtio_vsock_hdr *hdr; |
| 500 | |
| 501 | if (!vhost_vsock_more_replies(vsock)) { |
| 502 | /* Stop tx until the device processes already |
| 503 | * pending replies. Leave tx virtqueue |
| 504 | * callbacks disabled. |
| 505 | */ |
| 506 | goto no_more_replies; |
| 507 | } |
| 508 | |
| 509 | head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), |
| 510 | &out, &in, NULL, NULL); |
| 511 | if (head < 0) |
| 512 | break; |
| 513 | |
| 514 | if (head == vq->num) { |
| 515 | if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { |
| 516 | vhost_disable_notify(&vsock->dev, vq); |
| 517 | continue; |
| 518 | } |
| 519 | break; |
| 520 | } |
| 521 | |
| 522 | skb = vhost_vsock_alloc_skb(vq, out, in); |
| 523 | if (!skb) { |
| 524 | vq_err(vq, "Faulted on pkt\n"); |
| 525 | continue; |
| 526 | } |
| 527 | |
| 528 | total_len += sizeof(*hdr) + skb->len; |
| 529 | |
| 530 | /* Deliver to monitoring devices all received packets */ |
| 531 | virtio_transport_deliver_tap_pkt(skb); |
| 532 | |
| 533 | hdr = virtio_vsock_hdr(skb); |
| 534 | |
| 535 | /* Only accept correctly addressed packets */ |
| 536 | if (le64_to_cpu(hdr->src_cid) == vsock->guest_cid && |
| 537 | le64_to_cpu(hdr->dst_cid) == |
| 538 | vhost_transport_get_local_cid()) |
| 539 | virtio_transport_recv_pkt(&vhost_transport, skb); |
| 540 | else |
| 541 | kfree_skb(skb); |
| 542 | |
| 543 | vhost_add_used(vq, head, 0); |
| 544 | added = true; |
| 545 | } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); |
| 546 | |
| 547 | no_more_replies: |
| 548 | if (added) |
| 549 | vhost_signal(&vsock->dev, vq); |
| 550 | |
| 551 | out: |
| 552 | mutex_unlock(&vq->mutex); |
| 553 | } |
| 554 | |
| 555 | static void vhost_vsock_handle_rx_kick(struct vhost_work *work) |
| 556 | { |
| 557 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
| 558 | poll.work); |
| 559 | struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, |
| 560 | dev); |
| 561 | |
| 562 | vhost_transport_do_send_pkt(vsock, vq); |
| 563 | } |
| 564 | |
| 565 | static int vhost_vsock_start(struct vhost_vsock *vsock) |
| 566 | { |
| 567 | struct vhost_virtqueue *vq; |
| 568 | size_t i; |
| 569 | int ret; |
| 570 | |
| 571 | mutex_lock(&vsock->dev.mutex); |
| 572 | |
| 573 | ret = vhost_dev_check_owner(&vsock->dev); |
| 574 | if (ret) |
| 575 | goto err; |
| 576 | |
| 577 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
| 578 | vq = &vsock->vqs[i]; |
| 579 | |
| 580 | mutex_lock(&vq->mutex); |
| 581 | |
| 582 | if (!vhost_vq_access_ok(vq)) { |
| 583 | ret = -EFAULT; |
| 584 | goto err_vq; |
| 585 | } |
| 586 | |
| 587 | if (!vhost_vq_get_backend(vq)) { |
| 588 | vhost_vq_set_backend(vq, vsock); |
| 589 | ret = vhost_vq_init_access(vq); |
| 590 | if (ret) |
| 591 | goto err_vq; |
| 592 | } |
| 593 | |
| 594 | mutex_unlock(&vq->mutex); |
| 595 | } |
| 596 | |
| 597 | /* Some packets may have been queued before the device was started, |
| 598 | * let's kick the send worker to send them. |
| 599 | */ |
| 600 | vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &vsock->send_pkt_work); |
| 601 | |
| 602 | mutex_unlock(&vsock->dev.mutex); |
| 603 | return 0; |
| 604 | |
| 605 | err_vq: |
| 606 | vhost_vq_set_backend(vq, NULL); |
| 607 | mutex_unlock(&vq->mutex); |
| 608 | |
| 609 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
| 610 | vq = &vsock->vqs[i]; |
| 611 | |
| 612 | mutex_lock(&vq->mutex); |
| 613 | vhost_vq_set_backend(vq, NULL); |
| 614 | mutex_unlock(&vq->mutex); |
| 615 | } |
| 616 | err: |
| 617 | mutex_unlock(&vsock->dev.mutex); |
| 618 | return ret; |
| 619 | } |
| 620 | |
| 621 | static int vhost_vsock_stop(struct vhost_vsock *vsock, bool check_owner) |
| 622 | { |
| 623 | size_t i; |
| 624 | int ret = 0; |
| 625 | |
| 626 | mutex_lock(&vsock->dev.mutex); |
| 627 | |
| 628 | if (check_owner) { |
| 629 | ret = vhost_dev_check_owner(&vsock->dev); |
| 630 | if (ret) |
| 631 | goto err; |
| 632 | } |
| 633 | |
| 634 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
| 635 | struct vhost_virtqueue *vq = &vsock->vqs[i]; |
| 636 | |
| 637 | mutex_lock(&vq->mutex); |
| 638 | vhost_vq_set_backend(vq, NULL); |
| 639 | mutex_unlock(&vq->mutex); |
| 640 | } |
| 641 | |
| 642 | err: |
| 643 | mutex_unlock(&vsock->dev.mutex); |
| 644 | return ret; |
| 645 | } |
| 646 | |
| 647 | static void vhost_vsock_free(struct vhost_vsock *vsock) |
| 648 | { |
| 649 | kvfree(vsock); |
| 650 | } |
| 651 | |
| 652 | static int vhost_vsock_dev_open(struct inode *inode, struct file *file) |
| 653 | { |
| 654 | struct vhost_virtqueue **vqs; |
| 655 | struct vhost_vsock *vsock; |
| 656 | int ret; |
| 657 | |
| 658 | /* This struct is large and allocation could fail, fall back to vmalloc |
| 659 | * if there is no other way. |
| 660 | */ |
| 661 | vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
| 662 | if (!vsock) |
| 663 | return -ENOMEM; |
| 664 | |
| 665 | vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL); |
| 666 | if (!vqs) { |
| 667 | ret = -ENOMEM; |
| 668 | goto out; |
| 669 | } |
| 670 | |
| 671 | vsock->guest_cid = 0; /* no CID assigned yet */ |
| 672 | vsock->seqpacket_allow = false; |
| 673 | |
| 674 | atomic_set(&vsock->queued_replies, 0); |
| 675 | |
| 676 | vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX]; |
| 677 | vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX]; |
| 678 | vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick; |
| 679 | vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick; |
| 680 | |
| 681 | vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs), |
| 682 | UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT, |
| 683 | VHOST_VSOCK_WEIGHT, true, NULL); |
| 684 | |
| 685 | file->private_data = vsock; |
| 686 | skb_queue_head_init(&vsock->send_pkt_queue); |
| 687 | vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work); |
| 688 | return 0; |
| 689 | |
| 690 | out: |
| 691 | vhost_vsock_free(vsock); |
| 692 | return ret; |
| 693 | } |
| 694 | |
| 695 | static void vhost_vsock_flush(struct vhost_vsock *vsock) |
| 696 | { |
| 697 | vhost_dev_flush(&vsock->dev); |
| 698 | } |
| 699 | |
| 700 | static void vhost_vsock_reset_orphans(struct sock *sk) |
| 701 | { |
| 702 | struct vsock_sock *vsk = vsock_sk(sk); |
| 703 | |
| 704 | /* vmci_transport.c doesn't take sk_lock here either. At least we're |
| 705 | * under vsock_table_lock so the sock cannot disappear while we're |
| 706 | * executing. |
| 707 | */ |
| 708 | |
| 709 | /* If the peer is still valid, no need to reset connection */ |
| 710 | if (vhost_vsock_get(vsk->remote_addr.svm_cid)) |
| 711 | return; |
| 712 | |
| 713 | /* If the close timeout is pending, let it expire. This avoids races |
| 714 | * with the timeout callback. |
| 715 | */ |
| 716 | if (vsk->close_work_scheduled) |
| 717 | return; |
| 718 | |
| 719 | sock_set_flag(sk, SOCK_DONE); |
| 720 | vsk->peer_shutdown = SHUTDOWN_MASK; |
| 721 | sk->sk_state = SS_UNCONNECTED; |
| 722 | sk->sk_err = ECONNRESET; |
| 723 | sk_error_report(sk); |
| 724 | } |
| 725 | |
| 726 | static int vhost_vsock_dev_release(struct inode *inode, struct file *file) |
| 727 | { |
| 728 | struct vhost_vsock *vsock = file->private_data; |
| 729 | |
| 730 | mutex_lock(&vhost_vsock_mutex); |
| 731 | if (vsock->guest_cid) |
| 732 | hash_del_rcu(&vsock->hash); |
| 733 | mutex_unlock(&vhost_vsock_mutex); |
| 734 | |
| 735 | /* Wait for other CPUs to finish using vsock */ |
| 736 | synchronize_rcu(); |
| 737 | |
| 738 | /* Iterating over all connections for all CIDs to find orphans is |
| 739 | * inefficient. Room for improvement here. */ |
| 740 | vsock_for_each_connected_socket(&vhost_transport.transport, |
| 741 | vhost_vsock_reset_orphans); |
| 742 | |
| 743 | /* Don't check the owner, because we are in the release path, so we |
| 744 | * need to stop the vsock device in any case. |
| 745 | * vhost_vsock_stop() can not fail in this case, so we don't need to |
| 746 | * check the return code. |
| 747 | */ |
| 748 | vhost_vsock_stop(vsock, false); |
| 749 | vhost_vsock_flush(vsock); |
| 750 | vhost_dev_stop(&vsock->dev); |
| 751 | |
| 752 | virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue); |
| 753 | |
| 754 | vhost_dev_cleanup(&vsock->dev); |
| 755 | kfree(vsock->dev.vqs); |
| 756 | vhost_vsock_free(vsock); |
| 757 | return 0; |
| 758 | } |
| 759 | |
| 760 | static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid) |
| 761 | { |
| 762 | struct vhost_vsock *other; |
| 763 | |
| 764 | /* Refuse reserved CIDs */ |
| 765 | if (guest_cid <= VMADDR_CID_HOST || |
| 766 | guest_cid == U32_MAX) |
| 767 | return -EINVAL; |
| 768 | |
| 769 | /* 64-bit CIDs are not yet supported */ |
| 770 | if (guest_cid > U32_MAX) |
| 771 | return -EINVAL; |
| 772 | |
| 773 | /* Refuse if CID is assigned to the guest->host transport (i.e. nested |
| 774 | * VM), to make the loopback work. |
| 775 | */ |
| 776 | if (vsock_find_cid(guest_cid)) |
| 777 | return -EADDRINUSE; |
| 778 | |
| 779 | /* Refuse if CID is already in use */ |
| 780 | mutex_lock(&vhost_vsock_mutex); |
| 781 | other = vhost_vsock_get(guest_cid); |
| 782 | if (other && other != vsock) { |
| 783 | mutex_unlock(&vhost_vsock_mutex); |
| 784 | return -EADDRINUSE; |
| 785 | } |
| 786 | |
| 787 | if (vsock->guest_cid) |
| 788 | hash_del_rcu(&vsock->hash); |
| 789 | |
| 790 | vsock->guest_cid = guest_cid; |
| 791 | hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid); |
| 792 | mutex_unlock(&vhost_vsock_mutex); |
| 793 | |
| 794 | return 0; |
| 795 | } |
| 796 | |
| 797 | static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features) |
| 798 | { |
| 799 | struct vhost_virtqueue *vq; |
| 800 | int i; |
| 801 | |
| 802 | if (features & ~VHOST_VSOCK_FEATURES) |
| 803 | return -EOPNOTSUPP; |
| 804 | |
| 805 | mutex_lock(&vsock->dev.mutex); |
| 806 | if ((features & (1 << VHOST_F_LOG_ALL)) && |
| 807 | !vhost_log_access_ok(&vsock->dev)) { |
| 808 | goto err; |
| 809 | } |
| 810 | |
| 811 | if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) { |
| 812 | if (vhost_init_device_iotlb(&vsock->dev)) |
| 813 | goto err; |
| 814 | } |
| 815 | |
| 816 | vsock->seqpacket_allow = features & (1ULL << VIRTIO_VSOCK_F_SEQPACKET); |
| 817 | |
| 818 | for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
| 819 | vq = &vsock->vqs[i]; |
| 820 | mutex_lock(&vq->mutex); |
| 821 | vq->acked_features = features; |
| 822 | mutex_unlock(&vq->mutex); |
| 823 | } |
| 824 | mutex_unlock(&vsock->dev.mutex); |
| 825 | return 0; |
| 826 | |
| 827 | err: |
| 828 | mutex_unlock(&vsock->dev.mutex); |
| 829 | return -EFAULT; |
| 830 | } |
| 831 | |
| 832 | static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl, |
| 833 | unsigned long arg) |
| 834 | { |
| 835 | struct vhost_vsock *vsock = f->private_data; |
| 836 | void __user *argp = (void __user *)arg; |
| 837 | u64 guest_cid; |
| 838 | u64 features; |
| 839 | int start; |
| 840 | int r; |
| 841 | |
| 842 | switch (ioctl) { |
| 843 | case VHOST_VSOCK_SET_GUEST_CID: |
| 844 | if (copy_from_user(&guest_cid, argp, sizeof(guest_cid))) |
| 845 | return -EFAULT; |
| 846 | return vhost_vsock_set_cid(vsock, guest_cid); |
| 847 | case VHOST_VSOCK_SET_RUNNING: |
| 848 | if (copy_from_user(&start, argp, sizeof(start))) |
| 849 | return -EFAULT; |
| 850 | if (start) |
| 851 | return vhost_vsock_start(vsock); |
| 852 | else |
| 853 | return vhost_vsock_stop(vsock, true); |
| 854 | case VHOST_GET_FEATURES: |
| 855 | features = VHOST_VSOCK_FEATURES; |
| 856 | if (copy_to_user(argp, &features, sizeof(features))) |
| 857 | return -EFAULT; |
| 858 | return 0; |
| 859 | case VHOST_SET_FEATURES: |
| 860 | if (copy_from_user(&features, argp, sizeof(features))) |
| 861 | return -EFAULT; |
| 862 | return vhost_vsock_set_features(vsock, features); |
| 863 | case VHOST_GET_BACKEND_FEATURES: |
| 864 | features = VHOST_VSOCK_BACKEND_FEATURES; |
| 865 | if (copy_to_user(argp, &features, sizeof(features))) |
| 866 | return -EFAULT; |
| 867 | return 0; |
| 868 | case VHOST_SET_BACKEND_FEATURES: |
| 869 | if (copy_from_user(&features, argp, sizeof(features))) |
| 870 | return -EFAULT; |
| 871 | if (features & ~VHOST_VSOCK_BACKEND_FEATURES) |
| 872 | return -EOPNOTSUPP; |
| 873 | vhost_set_backend_features(&vsock->dev, features); |
| 874 | return 0; |
| 875 | default: |
| 876 | mutex_lock(&vsock->dev.mutex); |
| 877 | r = vhost_dev_ioctl(&vsock->dev, ioctl, argp); |
| 878 | if (r == -ENOIOCTLCMD) |
| 879 | r = vhost_vring_ioctl(&vsock->dev, ioctl, argp); |
| 880 | else |
| 881 | vhost_vsock_flush(vsock); |
| 882 | mutex_unlock(&vsock->dev.mutex); |
| 883 | return r; |
| 884 | } |
| 885 | } |
| 886 | |
| 887 | static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to) |
| 888 | { |
| 889 | struct file *file = iocb->ki_filp; |
| 890 | struct vhost_vsock *vsock = file->private_data; |
| 891 | struct vhost_dev *dev = &vsock->dev; |
| 892 | int noblock = file->f_flags & O_NONBLOCK; |
| 893 | |
| 894 | return vhost_chr_read_iter(dev, to, noblock); |
| 895 | } |
| 896 | |
| 897 | static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb, |
| 898 | struct iov_iter *from) |
| 899 | { |
| 900 | struct file *file = iocb->ki_filp; |
| 901 | struct vhost_vsock *vsock = file->private_data; |
| 902 | struct vhost_dev *dev = &vsock->dev; |
| 903 | |
| 904 | return vhost_chr_write_iter(dev, from); |
| 905 | } |
| 906 | |
| 907 | static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait) |
| 908 | { |
| 909 | struct vhost_vsock *vsock = file->private_data; |
| 910 | struct vhost_dev *dev = &vsock->dev; |
| 911 | |
| 912 | return vhost_chr_poll(file, dev, wait); |
| 913 | } |
| 914 | |
| 915 | static const struct file_operations vhost_vsock_fops = { |
| 916 | .owner = THIS_MODULE, |
| 917 | .open = vhost_vsock_dev_open, |
| 918 | .release = vhost_vsock_dev_release, |
| 919 | .llseek = noop_llseek, |
| 920 | .unlocked_ioctl = vhost_vsock_dev_ioctl, |
| 921 | .compat_ioctl = compat_ptr_ioctl, |
| 922 | .read_iter = vhost_vsock_chr_read_iter, |
| 923 | .write_iter = vhost_vsock_chr_write_iter, |
| 924 | .poll = vhost_vsock_chr_poll, |
| 925 | }; |
| 926 | |
| 927 | static struct miscdevice vhost_vsock_misc = { |
| 928 | .minor = VHOST_VSOCK_MINOR, |
| 929 | .name = "vhost-vsock", |
| 930 | .fops = &vhost_vsock_fops, |
| 931 | }; |
| 932 | |
| 933 | static int __init vhost_vsock_init(void) |
| 934 | { |
| 935 | int ret; |
| 936 | |
| 937 | ret = vsock_core_register(&vhost_transport.transport, |
| 938 | VSOCK_TRANSPORT_F_H2G); |
| 939 | if (ret < 0) |
| 940 | return ret; |
| 941 | |
| 942 | ret = misc_register(&vhost_vsock_misc); |
| 943 | if (ret) { |
| 944 | vsock_core_unregister(&vhost_transport.transport); |
| 945 | return ret; |
| 946 | } |
| 947 | |
| 948 | return 0; |
| 949 | }; |
| 950 | |
| 951 | static void __exit vhost_vsock_exit(void) |
| 952 | { |
| 953 | misc_deregister(&vhost_vsock_misc); |
| 954 | vsock_core_unregister(&vhost_transport.transport); |
| 955 | }; |
| 956 | |
| 957 | module_init(vhost_vsock_init); |
| 958 | module_exit(vhost_vsock_exit); |
| 959 | MODULE_LICENSE("GPL v2"); |
| 960 | MODULE_AUTHOR("Asias He"); |
| 961 | MODULE_DESCRIPTION("vhost transport for vsock "); |
| 962 | MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR); |
| 963 | MODULE_ALIAS("devname:vhost-vsock"); |