|
| 1 | +/* |
| 2 | + * vhost transport for vsock |
| 3 | + * |
| 4 | + * Copyright (C) 2013-2015 Red Hat, Inc. |
| 5 | + * Author: Asias He <asias@redhat.com> |
| 6 | + * Stefan Hajnoczi <stefanha@redhat.com> |
| 7 | + * |
| 8 | + * This work is licensed under the terms of the GNU GPL, version 2. |
| 9 | + */ |
| 10 | +#include <linux/miscdevice.h> |
| 11 | +#include <linux/atomic.h> |
| 12 | +#include <linux/module.h> |
| 13 | +#include <linux/mutex.h> |
| 14 | +#include <linux/vmalloc.h> |
| 15 | +#include <net/sock.h> |
| 16 | +#include <linux/virtio_vsock.h> |
| 17 | +#include <linux/vhost.h> |
| 18 | + |
| 19 | +#include <net/af_vsock.h> |
| 20 | +#include "vhost.h" |
| 21 | + |
| 22 | +#define VHOST_VSOCK_DEFAULT_HOST_CID 2 |
| 23 | + |
| 24 | +enum { |
| 25 | + VHOST_VSOCK_FEATURES = VHOST_FEATURES, |
| 26 | +}; |
| 27 | + |
| 28 | +/* Used to track all the vhost_vsock instances on the system. */ |
| 29 | +static DEFINE_SPINLOCK(vhost_vsock_lock); |
| 30 | +static LIST_HEAD(vhost_vsock_list); |
| 31 | + |
| 32 | +struct vhost_vsock { |
| 33 | + struct vhost_dev dev; |
| 34 | + struct vhost_virtqueue vqs[2]; |
| 35 | + |
| 36 | + /* Link to global vhost_vsock_list, protected by vhost_vsock_lock */ |
| 37 | + struct list_head list; |
| 38 | + |
| 39 | + struct vhost_work send_pkt_work; |
| 40 | + spinlock_t send_pkt_list_lock; |
| 41 | + struct list_head send_pkt_list; /* host->guest pending packets */ |
| 42 | + |
| 43 | + atomic_t queued_replies; |
| 44 | + |
| 45 | + u32 guest_cid; |
| 46 | +}; |
| 47 | + |
| 48 | +static u32 vhost_transport_get_local_cid(void) |
| 49 | +{ |
| 50 | + return VHOST_VSOCK_DEFAULT_HOST_CID; |
| 51 | +} |
| 52 | + |
| 53 | +static struct vhost_vsock *vhost_vsock_get(u32 guest_cid) |
| 54 | +{ |
| 55 | + struct vhost_vsock *vsock; |
| 56 | + |
| 57 | + spin_lock_bh(&vhost_vsock_lock); |
| 58 | + list_for_each_entry(vsock, &vhost_vsock_list, list) { |
| 59 | + u32 other_cid = vsock->guest_cid; |
| 60 | + |
| 61 | + /* Skip instances that have no CID yet */ |
| 62 | + if (other_cid == 0) |
| 63 | + continue; |
| 64 | + |
| 65 | + if (other_cid == guest_cid) { |
| 66 | + spin_unlock_bh(&vhost_vsock_lock); |
| 67 | + return vsock; |
| 68 | + } |
| 69 | + } |
| 70 | + spin_unlock_bh(&vhost_vsock_lock); |
| 71 | + |
| 72 | + return NULL; |
| 73 | +} |
| 74 | + |
| 75 | +static void |
| 76 | +vhost_transport_do_send_pkt(struct vhost_vsock *vsock, |
| 77 | + struct vhost_virtqueue *vq) |
| 78 | +{ |
| 79 | + struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX]; |
| 80 | + bool added = false; |
| 81 | + bool restart_tx = false; |
| 82 | + |
| 83 | + mutex_lock(&vq->mutex); |
| 84 | + |
| 85 | + if (!vq->private_data) |
| 86 | + goto out; |
| 87 | + |
| 88 | + /* Avoid further vmexits, we're already processing the virtqueue */ |
| 89 | + vhost_disable_notify(&vsock->dev, vq); |
| 90 | + |
| 91 | + for (;;) { |
| 92 | + struct virtio_vsock_pkt *pkt; |
| 93 | + struct iov_iter iov_iter; |
| 94 | + unsigned out, in; |
| 95 | + size_t nbytes; |
| 96 | + size_t len; |
| 97 | + int head; |
| 98 | + |
| 99 | + spin_lock_bh(&vsock->send_pkt_list_lock); |
| 100 | + if (list_empty(&vsock->send_pkt_list)) { |
| 101 | + spin_unlock_bh(&vsock->send_pkt_list_lock); |
| 102 | + vhost_enable_notify(&vsock->dev, vq); |
| 103 | + break; |
| 104 | + } |
| 105 | + |
| 106 | + pkt = list_first_entry(&vsock->send_pkt_list, |
| 107 | + struct virtio_vsock_pkt, list); |
| 108 | + list_del_init(&pkt->list); |
| 109 | + spin_unlock_bh(&vsock->send_pkt_list_lock); |
| 110 | + |
| 111 | + head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), |
| 112 | + &out, &in, NULL, NULL); |
| 113 | + if (head < 0) { |
| 114 | + spin_lock_bh(&vsock->send_pkt_list_lock); |
| 115 | + list_add(&pkt->list, &vsock->send_pkt_list); |
| 116 | + spin_unlock_bh(&vsock->send_pkt_list_lock); |
| 117 | + break; |
| 118 | + } |
| 119 | + |
| 120 | + if (head == vq->num) { |
| 121 | + spin_lock_bh(&vsock->send_pkt_list_lock); |
| 122 | + list_add(&pkt->list, &vsock->send_pkt_list); |
| 123 | + spin_unlock_bh(&vsock->send_pkt_list_lock); |
| 124 | + |
| 125 | + /* We cannot finish yet if more buffers snuck in while |
| 126 | + * re-enabling notify. |
| 127 | + */ |
| 128 | + if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { |
| 129 | + vhost_disable_notify(&vsock->dev, vq); |
| 130 | + continue; |
| 131 | + } |
| 132 | + break; |
| 133 | + } |
| 134 | + |
| 135 | + if (out) { |
| 136 | + virtio_transport_free_pkt(pkt); |
| 137 | + vq_err(vq, "Expected 0 output buffers, got %u\n", out); |
| 138 | + break; |
| 139 | + } |
| 140 | + |
| 141 | + len = iov_length(&vq->iov[out], in); |
| 142 | + iov_iter_init(&iov_iter, READ, &vq->iov[out], in, len); |
| 143 | + |
| 144 | + nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); |
| 145 | + if (nbytes != sizeof(pkt->hdr)) { |
| 146 | + virtio_transport_free_pkt(pkt); |
| 147 | + vq_err(vq, "Faulted on copying pkt hdr\n"); |
| 148 | + break; |
| 149 | + } |
| 150 | + |
| 151 | + nbytes = copy_to_iter(pkt->buf, pkt->len, &iov_iter); |
| 152 | + if (nbytes != pkt->len) { |
| 153 | + virtio_transport_free_pkt(pkt); |
| 154 | + vq_err(vq, "Faulted on copying pkt buf\n"); |
| 155 | + break; |
| 156 | + } |
| 157 | + |
| 158 | + vhost_add_used(vq, head, sizeof(pkt->hdr) + pkt->len); |
| 159 | + added = true; |
| 160 | + |
| 161 | + if (pkt->reply) { |
| 162 | + int val; |
| 163 | + |
| 164 | + val = atomic_dec_return(&vsock->queued_replies); |
| 165 | + |
| 166 | + /* Do we have resources to resume tx processing? */ |
| 167 | + if (val + 1 == tx_vq->num) |
| 168 | + restart_tx = true; |
| 169 | + } |
| 170 | + |
| 171 | + virtio_transport_free_pkt(pkt); |
| 172 | + } |
| 173 | + if (added) |
| 174 | + vhost_signal(&vsock->dev, vq); |
| 175 | + |
| 176 | +out: |
| 177 | + mutex_unlock(&vq->mutex); |
| 178 | + |
| 179 | + if (restart_tx) |
| 180 | + vhost_poll_queue(&tx_vq->poll); |
| 181 | +} |
| 182 | + |
| 183 | +static void vhost_transport_send_pkt_work(struct vhost_work *work) |
| 184 | +{ |
| 185 | + struct vhost_virtqueue *vq; |
| 186 | + struct vhost_vsock *vsock; |
| 187 | + |
| 188 | + vsock = container_of(work, struct vhost_vsock, send_pkt_work); |
| 189 | + vq = &vsock->vqs[VSOCK_VQ_RX]; |
| 190 | + |
| 191 | + vhost_transport_do_send_pkt(vsock, vq); |
| 192 | +} |
| 193 | + |
| 194 | +static int |
| 195 | +vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt) |
| 196 | +{ |
| 197 | + struct vhost_vsock *vsock; |
| 198 | + struct vhost_virtqueue *vq; |
| 199 | + int len = pkt->len; |
| 200 | + |
| 201 | + /* Find the vhost_vsock according to guest context id */ |
| 202 | + vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid)); |
| 203 | + if (!vsock) { |
| 204 | + virtio_transport_free_pkt(pkt); |
| 205 | + return -ENODEV; |
| 206 | + } |
| 207 | + |
| 208 | + vq = &vsock->vqs[VSOCK_VQ_RX]; |
| 209 | + |
| 210 | + if (pkt->reply) |
| 211 | + atomic_inc(&vsock->queued_replies); |
| 212 | + |
| 213 | + spin_lock_bh(&vsock->send_pkt_list_lock); |
| 214 | + list_add_tail(&pkt->list, &vsock->send_pkt_list); |
| 215 | + spin_unlock_bh(&vsock->send_pkt_list_lock); |
| 216 | + |
| 217 | + vhost_work_queue(&vsock->dev, &vsock->send_pkt_work); |
| 218 | + return len; |
| 219 | +} |
| 220 | + |
| 221 | +static struct virtio_vsock_pkt * |
| 222 | +vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq, |
| 223 | + unsigned int out, unsigned int in) |
| 224 | +{ |
| 225 | + struct virtio_vsock_pkt *pkt; |
| 226 | + struct iov_iter iov_iter; |
| 227 | + size_t nbytes; |
| 228 | + size_t len; |
| 229 | + |
| 230 | + if (in != 0) { |
| 231 | + vq_err(vq, "Expected 0 input buffers, got %u\n", in); |
| 232 | + return NULL; |
| 233 | + } |
| 234 | + |
| 235 | + pkt = kzalloc(sizeof(*pkt), GFP_KERNEL); |
| 236 | + if (!pkt) |
| 237 | + return NULL; |
| 238 | + |
| 239 | + len = iov_length(vq->iov, out); |
| 240 | + iov_iter_init(&iov_iter, WRITE, vq->iov, out, len); |
| 241 | + |
| 242 | + nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); |
| 243 | + if (nbytes != sizeof(pkt->hdr)) { |
| 244 | + vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n", |
| 245 | + sizeof(pkt->hdr), nbytes); |
| 246 | + kfree(pkt); |
| 247 | + return NULL; |
| 248 | + } |
| 249 | + |
| 250 | + if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM) |
| 251 | + pkt->len = le32_to_cpu(pkt->hdr.len); |
| 252 | + |
| 253 | + /* No payload */ |
| 254 | + if (!pkt->len) |
| 255 | + return pkt; |
| 256 | + |
| 257 | + /* The pkt is too big */ |
| 258 | + if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) { |
| 259 | + kfree(pkt); |
| 260 | + return NULL; |
| 261 | + } |
| 262 | + |
| 263 | + pkt->buf = kmalloc(pkt->len, GFP_KERNEL); |
| 264 | + if (!pkt->buf) { |
| 265 | + kfree(pkt); |
| 266 | + return NULL; |
| 267 | + } |
| 268 | + |
| 269 | + nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter); |
| 270 | + if (nbytes != pkt->len) { |
| 271 | + vq_err(vq, "Expected %u byte payload, got %zu bytes\n", |
| 272 | + pkt->len, nbytes); |
| 273 | + virtio_transport_free_pkt(pkt); |
| 274 | + return NULL; |
| 275 | + } |
| 276 | + |
| 277 | + return pkt; |
| 278 | +} |
| 279 | + |
| 280 | +/* Is there space left for replies to rx packets? */ |
| 281 | +static bool vhost_vsock_more_replies(struct vhost_vsock *vsock) |
| 282 | +{ |
| 283 | + struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX]; |
| 284 | + int val; |
| 285 | + |
| 286 | + smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */ |
| 287 | + val = atomic_read(&vsock->queued_replies); |
| 288 | + |
| 289 | + return val < vq->num; |
| 290 | +} |
| 291 | + |
| 292 | +static void vhost_vsock_handle_tx_kick(struct vhost_work *work) |
| 293 | +{ |
| 294 | + struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
| 295 | + poll.work); |
| 296 | + struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, |
| 297 | + dev); |
| 298 | + struct virtio_vsock_pkt *pkt; |
| 299 | + int head; |
| 300 | + unsigned int out, in; |
| 301 | + bool added = false; |
| 302 | + |
| 303 | + mutex_lock(&vq->mutex); |
| 304 | + |
| 305 | + if (!vq->private_data) |
| 306 | + goto out; |
| 307 | + |
| 308 | + vhost_disable_notify(&vsock->dev, vq); |
| 309 | + for (;;) { |
| 310 | + if (!vhost_vsock_more_replies(vsock)) { |
| 311 | + /* Stop tx until the device processes already |
| 312 | + * pending replies. Leave tx virtqueue |
| 313 | + * callbacks disabled. |
| 314 | + */ |
| 315 | + goto no_more_replies; |
| 316 | + } |
| 317 | + |
| 318 | + head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), |
| 319 | + &out, &in, NULL, NULL); |
| 320 | + if (head < 0) |
| 321 | + break; |
| 322 | + |
| 323 | + if (head == vq->num) { |
| 324 | + if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { |
| 325 | + vhost_disable_notify(&vsock->dev, vq); |
| 326 | + continue; |
| 327 | + } |
| 328 | + break; |
| 329 | + } |
| 330 | + |
| 331 | + pkt = vhost_vsock_alloc_pkt(vq, out, in); |
| 332 | + if (!pkt) { |
| 333 | + vq_err(vq, "Faulted on pkt\n"); |
| 334 | + continue; |
| 335 | + } |
| 336 | + |
| 337 | + /* Only accept correctly addressed packets */ |
| 338 | + if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid) |
| 339 | + virtio_transport_recv_pkt(pkt); |
| 340 | + else |
| 341 | + virtio_transport_free_pkt(pkt); |
| 342 | + |
| 343 | + vhost_add_used(vq, head, sizeof(pkt->hdr) + pkt->len); |
| 344 | + added = true; |
| 345 | + } |
| 346 | + |
| 347 | +no_more_replies: |
| 348 | + if (added) |
| 349 | + vhost_signal(&vsock->dev, vq); |
| 350 | + |
| 351 | +out: |
| 352 | + mutex_unlock(&vq->mutex); |
| 353 | +} |
| 354 | + |
| 355 | +static void vhost_vsock_handle_rx_kick(struct vhost_work *work) |
| 356 | +{ |
| 357 | + struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
| 358 | + poll.work); |
| 359 | + struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, |
| 360 | + dev); |
| 361 | + |
| 362 | + vhost_transport_do_send_pkt(vsock, vq); |
| 363 | +} |
| 364 | + |
| 365 | +static int vhost_vsock_start(struct vhost_vsock *vsock) |
| 366 | +{ |
| 367 | + size_t i; |
| 368 | + int ret; |
| 369 | + |
| 370 | + mutex_lock(&vsock->dev.mutex); |
| 371 | + |
| 372 | + ret = vhost_dev_check_owner(&vsock->dev); |
| 373 | + if (ret) |
| 374 | + goto err; |
| 375 | + |
| 376 | + for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
| 377 | + struct vhost_virtqueue *vq = &vsock->vqs[i]; |
| 378 | + |
| 379 | + mutex_lock(&vq->mutex); |
| 380 | + |
| 381 | + if (!vhost_vq_access_ok(vq)) { |
| 382 | + ret = -EFAULT; |
| 383 | + mutex_unlock(&vq->mutex); |
| 384 | + goto err_vq; |
| 385 | + } |
| 386 | + |
| 387 | + if (!vq->private_data) { |
| 388 | + vq->private_data = vsock; |
| 389 | + vhost_vq_init_access(vq); |
| 390 | + } |
| 391 | + |
| 392 | + mutex_unlock(&vq->mutex); |
| 393 | + } |
| 394 | + |
| 395 | + mutex_unlock(&vsock->dev.mutex); |
| 396 | + return 0; |
| 397 | + |
| 398 | +err_vq: |
| 399 | + for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
| 400 | + struct vhost_virtqueue *vq = &vsock->vqs[i]; |
| 401 | + |
| 402 | + mutex_lock(&vq->mutex); |
| 403 | + vq->private_data = NULL; |
| 404 | + mutex_unlock(&vq->mutex); |
| 405 | + } |
| 406 | +err: |
| 407 | + mutex_unlock(&vsock->dev.mutex); |
| 408 | + return ret; |
| 409 | +} |
| 410 | + |
| 411 | +static int vhost_vsock_stop(struct vhost_vsock *vsock) |
| 412 | +{ |
| 413 | + size_t i; |
| 414 | + int ret; |
| 415 | + |
| 416 | + mutex_lock(&vsock->dev.mutex); |
| 417 | + |
| 418 | + ret = vhost_dev_check_owner(&vsock->dev); |
| 419 | + if (ret) |
| 420 | + goto err; |
| 421 | + |
| 422 | + for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
| 423 | + struct vhost_virtqueue *vq = &vsock->vqs[i]; |
| 424 | + |
| 425 | + mutex_lock(&vq->mutex); |
| 426 | + vq->private_data = NULL; |
| 427 | + mutex_unlock(&vq->mutex); |
| 428 | + } |
| 429 | + |
| 430 | +err: |
| 431 | + mutex_unlock(&vsock->dev.mutex); |
| 432 | + return ret; |
| 433 | +} |
| 434 | + |
| 435 | +static void vhost_vsock_free(struct vhost_vsock *vsock) |
| 436 | +{ |
| 437 | + if (is_vmalloc_addr(vsock)) |
| 438 | + vfree(vsock); |
| 439 | + else |
| 440 | + kfree(vsock); |
| 441 | +} |
| 442 | + |
| 443 | +static int vhost_vsock_dev_open(struct inode *inode, struct file *file) |
| 444 | +{ |
| 445 | + struct vhost_virtqueue **vqs; |
| 446 | + struct vhost_vsock *vsock; |
| 447 | + int ret; |
| 448 | + |
| 449 | + /* This struct is large and allocation could fail, fall back to vmalloc |
| 450 | + * if there is no other way. |
| 451 | + */ |
| 452 | + vsock = kzalloc(sizeof(*vsock), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
| 453 | + if (!vsock) { |
| 454 | + vsock = vmalloc(sizeof(*vsock)); |
| 455 | + if (!vsock) |
| 456 | + return -ENOMEM; |
| 457 | + } |
| 458 | + |
| 459 | + vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL); |
| 460 | + if (!vqs) { |
| 461 | + ret = -ENOMEM; |
| 462 | + goto out; |
| 463 | + } |
| 464 | + |
| 465 | + atomic_set(&vsock->queued_replies, 0); |
| 466 | + |
| 467 | + vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX]; |
| 468 | + vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX]; |
| 469 | + vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick; |
| 470 | + vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick; |
| 471 | + |
| 472 | + vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs)); |
| 473 | + |
| 474 | + file->private_data = vsock; |
| 475 | + spin_lock_init(&vsock->send_pkt_list_lock); |
| 476 | + INIT_LIST_HEAD(&vsock->send_pkt_list); |
| 477 | + vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work); |
| 478 | + |
| 479 | + spin_lock_bh(&vhost_vsock_lock); |
| 480 | + list_add_tail(&vsock->list, &vhost_vsock_list); |
| 481 | + spin_unlock_bh(&vhost_vsock_lock); |
| 482 | + return 0; |
| 483 | + |
| 484 | +out: |
| 485 | + vhost_vsock_free(vsock); |
| 486 | + return ret; |
| 487 | +} |
| 488 | + |
| 489 | +static void vhost_vsock_flush(struct vhost_vsock *vsock) |
| 490 | +{ |
| 491 | + int i; |
| 492 | + |
| 493 | + for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) |
| 494 | + if (vsock->vqs[i].handle_kick) |
| 495 | + vhost_poll_flush(&vsock->vqs[i].poll); |
| 496 | + vhost_work_flush(&vsock->dev, &vsock->send_pkt_work); |
| 497 | +} |
| 498 | + |
| 499 | +static void vhost_vsock_reset_orphans(struct sock *sk) |
| 500 | +{ |
| 501 | + struct vsock_sock *vsk = vsock_sk(sk); |
| 502 | + |
| 503 | + /* vmci_transport.c doesn't take sk_lock here either. At least we're |
| 504 | + * under vsock_table_lock so the sock cannot disappear while we're |
| 505 | + * executing. |
| 506 | + */ |
| 507 | + |
| 508 | + if (!vhost_vsock_get(vsk->local_addr.svm_cid)) { |
| 509 | + sock_set_flag(sk, SOCK_DONE); |
| 510 | + vsk->peer_shutdown = SHUTDOWN_MASK; |
| 511 | + sk->sk_state = SS_UNCONNECTED; |
| 512 | + sk->sk_err = ECONNRESET; |
| 513 | + sk->sk_error_report(sk); |
| 514 | + } |
| 515 | +} |
| 516 | + |
| 517 | +static int vhost_vsock_dev_release(struct inode *inode, struct file *file) |
| 518 | +{ |
| 519 | + struct vhost_vsock *vsock = file->private_data; |
| 520 | + |
| 521 | + spin_lock_bh(&vhost_vsock_lock); |
| 522 | + list_del(&vsock->list); |
| 523 | + spin_unlock_bh(&vhost_vsock_lock); |
| 524 | + |
| 525 | + /* Iterating over all connections for all CIDs to find orphans is |
| 526 | + * inefficient. Room for improvement here. */ |
| 527 | + vsock_for_each_connected_socket(vhost_vsock_reset_orphans); |
| 528 | + |
| 529 | + vhost_vsock_stop(vsock); |
| 530 | + vhost_vsock_flush(vsock); |
| 531 | + vhost_dev_stop(&vsock->dev); |
| 532 | + |
| 533 | + spin_lock_bh(&vsock->send_pkt_list_lock); |
| 534 | + while (!list_empty(&vsock->send_pkt_list)) { |
| 535 | + struct virtio_vsock_pkt *pkt; |
| 536 | + |
| 537 | + pkt = list_first_entry(&vsock->send_pkt_list, |
| 538 | + struct virtio_vsock_pkt, list); |
| 539 | + list_del_init(&pkt->list); |
| 540 | + virtio_transport_free_pkt(pkt); |
| 541 | + } |
| 542 | + spin_unlock_bh(&vsock->send_pkt_list_lock); |
| 543 | + |
| 544 | + vhost_dev_cleanup(&vsock->dev, false); |
| 545 | + kfree(vsock->dev.vqs); |
| 546 | + vhost_vsock_free(vsock); |
| 547 | + return 0; |
| 548 | +} |
| 549 | + |
| 550 | +static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid) |
| 551 | +{ |
| 552 | + struct vhost_vsock *other; |
| 553 | + |
| 554 | + /* Refuse reserved CIDs */ |
| 555 | + if (guest_cid <= VMADDR_CID_HOST || |
| 556 | + guest_cid == U32_MAX) |
| 557 | + return -EINVAL; |
| 558 | + |
| 559 | + /* 64-bit CIDs are not yet supported */ |
| 560 | + if (guest_cid > U32_MAX) |
| 561 | + return -EINVAL; |
| 562 | + |
| 563 | + /* Refuse if CID is already in use */ |
| 564 | + other = vhost_vsock_get(guest_cid); |
| 565 | + if (other && other != vsock) |
| 566 | + return -EADDRINUSE; |
| 567 | + |
| 568 | + spin_lock_bh(&vhost_vsock_lock); |
| 569 | + vsock->guest_cid = guest_cid; |
| 570 | + spin_unlock_bh(&vhost_vsock_lock); |
| 571 | + |
| 572 | + return 0; |
| 573 | +} |
| 574 | + |
| 575 | +static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features) |
| 576 | +{ |
| 577 | + struct vhost_virtqueue *vq; |
| 578 | + int i; |
| 579 | + |
| 580 | + if (features & ~VHOST_VSOCK_FEATURES) |
| 581 | + return -EOPNOTSUPP; |
| 582 | + |
| 583 | + mutex_lock(&vsock->dev.mutex); |
| 584 | + if ((features & (1 << VHOST_F_LOG_ALL)) && |
| 585 | + !vhost_log_access_ok(&vsock->dev)) { |
| 586 | + mutex_unlock(&vsock->dev.mutex); |
| 587 | + return -EFAULT; |
| 588 | + } |
| 589 | + |
| 590 | + for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) { |
| 591 | + vq = &vsock->vqs[i]; |
| 592 | + mutex_lock(&vq->mutex); |
| 593 | + vq->acked_features = features; |
| 594 | + mutex_unlock(&vq->mutex); |
| 595 | + } |
| 596 | + mutex_unlock(&vsock->dev.mutex); |
| 597 | + return 0; |
| 598 | +} |
| 599 | + |
| 600 | +static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl, |
| 601 | + unsigned long arg) |
| 602 | +{ |
| 603 | + struct vhost_vsock *vsock = f->private_data; |
| 604 | + void __user *argp = (void __user *)arg; |
| 605 | + u64 guest_cid; |
| 606 | + u64 features; |
| 607 | + int start; |
| 608 | + int r; |
| 609 | + |
| 610 | + switch (ioctl) { |
| 611 | + case VHOST_VSOCK_SET_GUEST_CID: |
| 612 | + if (copy_from_user(&guest_cid, argp, sizeof(guest_cid))) |
| 613 | + return -EFAULT; |
| 614 | + return vhost_vsock_set_cid(vsock, guest_cid); |
| 615 | + case VHOST_VSOCK_SET_RUNNING: |
| 616 | + if (copy_from_user(&start, argp, sizeof(start))) |
| 617 | + return -EFAULT; |
| 618 | + if (start) |
| 619 | + return vhost_vsock_start(vsock); |
| 620 | + else |
| 621 | + return vhost_vsock_stop(vsock); |
| 622 | + case VHOST_GET_FEATURES: |
| 623 | + features = VHOST_VSOCK_FEATURES; |
| 624 | + if (copy_to_user(argp, &features, sizeof(features))) |
| 625 | + return -EFAULT; |
| 626 | + return 0; |
| 627 | + case VHOST_SET_FEATURES: |
| 628 | + if (copy_from_user(&features, argp, sizeof(features))) |
| 629 | + return -EFAULT; |
| 630 | + return vhost_vsock_set_features(vsock, features); |
| 631 | + default: |
| 632 | + mutex_lock(&vsock->dev.mutex); |
| 633 | + r = vhost_dev_ioctl(&vsock->dev, ioctl, argp); |
| 634 | + if (r == -ENOIOCTLCMD) |
| 635 | + r = vhost_vring_ioctl(&vsock->dev, ioctl, argp); |
| 636 | + else |
| 637 | + vhost_vsock_flush(vsock); |
| 638 | + mutex_unlock(&vsock->dev.mutex); |
| 639 | + return r; |
| 640 | + } |
| 641 | +} |
| 642 | + |
| 643 | +static const struct file_operations vhost_vsock_fops = { |
| 644 | + .owner = THIS_MODULE, |
| 645 | + .open = vhost_vsock_dev_open, |
| 646 | + .release = vhost_vsock_dev_release, |
| 647 | + .llseek = noop_llseek, |
| 648 | + .unlocked_ioctl = vhost_vsock_dev_ioctl, |
| 649 | +}; |
| 650 | + |
| 651 | +static struct miscdevice vhost_vsock_misc = { |
| 652 | + .minor = MISC_DYNAMIC_MINOR, |
| 653 | + .name = "vhost-vsock", |
| 654 | + .fops = &vhost_vsock_fops, |
| 655 | +}; |
| 656 | + |
| 657 | +static struct virtio_transport vhost_transport = { |
| 658 | + .transport = { |
| 659 | + .get_local_cid = vhost_transport_get_local_cid, |
| 660 | + |
| 661 | + .init = virtio_transport_do_socket_init, |
| 662 | + .destruct = virtio_transport_destruct, |
| 663 | + .release = virtio_transport_release, |
| 664 | + .connect = virtio_transport_connect, |
| 665 | + .shutdown = virtio_transport_shutdown, |
| 666 | + |
| 667 | + .dgram_enqueue = virtio_transport_dgram_enqueue, |
| 668 | + .dgram_dequeue = virtio_transport_dgram_dequeue, |
| 669 | + .dgram_bind = virtio_transport_dgram_bind, |
| 670 | + .dgram_allow = virtio_transport_dgram_allow, |
| 671 | + |
| 672 | + .stream_enqueue = virtio_transport_stream_enqueue, |
| 673 | + .stream_dequeue = virtio_transport_stream_dequeue, |
| 674 | + .stream_has_data = virtio_transport_stream_has_data, |
| 675 | + .stream_has_space = virtio_transport_stream_has_space, |
| 676 | + .stream_rcvhiwat = virtio_transport_stream_rcvhiwat, |
| 677 | + .stream_is_active = virtio_transport_stream_is_active, |
| 678 | + .stream_allow = virtio_transport_stream_allow, |
| 679 | + |
| 680 | + .notify_poll_in = virtio_transport_notify_poll_in, |
| 681 | + .notify_poll_out = virtio_transport_notify_poll_out, |
| 682 | + .notify_recv_init = virtio_transport_notify_recv_init, |
| 683 | + .notify_recv_pre_block = virtio_transport_notify_recv_pre_block, |
| 684 | + .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue, |
| 685 | + .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue, |
| 686 | + .notify_send_init = virtio_transport_notify_send_init, |
| 687 | + .notify_send_pre_block = virtio_transport_notify_send_pre_block, |
| 688 | + .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue, |
| 689 | + .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue, |
| 690 | + |
| 691 | + .set_buffer_size = virtio_transport_set_buffer_size, |
| 692 | + .set_min_buffer_size = virtio_transport_set_min_buffer_size, |
| 693 | + .set_max_buffer_size = virtio_transport_set_max_buffer_size, |
| 694 | + .get_buffer_size = virtio_transport_get_buffer_size, |
| 695 | + .get_min_buffer_size = virtio_transport_get_min_buffer_size, |
| 696 | + .get_max_buffer_size = virtio_transport_get_max_buffer_size, |
| 697 | + }, |
| 698 | + |
| 699 | + .send_pkt = vhost_transport_send_pkt, |
| 700 | +}; |
| 701 | + |
| 702 | +static int __init vhost_vsock_init(void) |
| 703 | +{ |
| 704 | + int ret; |
| 705 | + |
| 706 | + ret = vsock_core_init(&vhost_transport.transport); |
| 707 | + if (ret < 0) |
| 708 | + return ret; |
| 709 | + return misc_register(&vhost_vsock_misc); |
| 710 | +}; |
| 711 | + |
| 712 | +static void __exit vhost_vsock_exit(void) |
| 713 | +{ |
| 714 | + misc_deregister(&vhost_vsock_misc); |
| 715 | + vsock_core_exit(); |
| 716 | +}; |
| 717 | + |
| 718 | +module_init(vhost_vsock_init); |
| 719 | +module_exit(vhost_vsock_exit); |
| 720 | +MODULE_LICENSE("GPL v2"); |
| 721 | +MODULE_AUTHOR("Asias He"); |
| 722 | +MODULE_DESCRIPTION("vhost transport for vsock "); |
0 commit comments