Carl van Schaik | 6d7b2ff | 2018-07-06 22:00:55 +1000 | [diff] [blame] | 1 | /* |
| 2 | * devio.c - cdev I/O for service devices |
| 3 | * |
| 4 | * Copyright (c) 2016 Cog Systems Pty Ltd |
| 5 | * Author: Philip Derrin <philip@cog.systems> |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or modify it |
| 8 | * under the terms and conditions of the GNU General Public License, |
| 9 | * version 2, as published by the Free Software Foundation. |
| 10 | */ |
| 11 | |
| 12 | #include <linux/version.h> |
| 13 | #include <linux/types.h> |
| 14 | #include <linux/device.h> |
| 15 | #include <linux/cdev.h> |
| 16 | #include <linux/pagemap.h> |
| 17 | #include <linux/fs.h> |
| 18 | #include <linux/sched.h> |
| 19 | #include <linux/wait.h> |
| 20 | #include <linux/list.h> |
| 21 | #include <linux/atomic.h> |
| 22 | #include <linux/module.h> |
| 23 | #include <linux/spinlock.h> |
| 24 | #include <linux/uio.h> |
| 25 | #include <linux/uaccess.h> |
| 26 | #include <linux/poll.h> |
| 27 | #include <linux/security.h> |
| 28 | #include <linux/compat.h> |
| 29 | |
| 30 | #include <vservices/types.h> |
| 31 | #include <vservices/buffer.h> |
| 32 | #include <vservices/transport.h> |
| 33 | #include <vservices/session.h> |
| 34 | #include <vservices/service.h> |
| 35 | #include <vservices/ioctl.h> |
| 36 | #include "session.h" |
| 37 | |
| 38 | #define VSERVICES_DEVICE_MAX (VS_MAX_SERVICES * VS_MAX_SESSIONS) |
| 39 | |
| 40 | struct vs_devio_priv { |
| 41 | struct kref kref; |
| 42 | bool running, reset; |
| 43 | |
| 44 | /* Receive queue */ |
| 45 | wait_queue_head_t recv_wq; |
| 46 | atomic_t notify_pending; |
| 47 | struct list_head recv_queue; |
| 48 | }; |
| 49 | |
| 50 | static void |
| 51 | vs_devio_priv_free(struct kref *kref) |
| 52 | { |
| 53 | struct vs_devio_priv *priv = container_of(kref, struct vs_devio_priv, |
| 54 | kref); |
| 55 | |
| 56 | WARN_ON(priv->running); |
| 57 | WARN_ON(!list_empty_careful(&priv->recv_queue)); |
| 58 | WARN_ON(waitqueue_active(&priv->recv_wq)); |
| 59 | |
| 60 | kfree(priv); |
| 61 | } |
| 62 | |
| 63 | static void vs_devio_priv_put(struct vs_devio_priv *priv) |
| 64 | { |
| 65 | kref_put(&priv->kref, vs_devio_priv_free); |
| 66 | } |
| 67 | |
| 68 | static int |
| 69 | vs_devio_service_probe(struct vs_service_device *service) |
| 70 | { |
| 71 | struct vs_devio_priv *priv; |
| 72 | |
| 73 | priv = kmalloc(sizeof(*priv), GFP_KERNEL); |
| 74 | if (!priv) |
| 75 | return -ENOMEM; |
| 76 | |
| 77 | kref_init(&priv->kref); |
| 78 | priv->running = false; |
| 79 | priv->reset = false; |
| 80 | init_waitqueue_head(&priv->recv_wq); |
| 81 | atomic_set(&priv->notify_pending, 0); |
| 82 | INIT_LIST_HEAD(&priv->recv_queue); |
| 83 | |
| 84 | dev_set_drvdata(&service->dev, priv); |
| 85 | |
| 86 | wake_up(&service->quota_wq); |
| 87 | |
| 88 | return 0; |
| 89 | } |
| 90 | |
| 91 | static int |
| 92 | vs_devio_service_remove(struct vs_service_device *service) |
| 93 | { |
| 94 | struct vs_devio_priv *priv = dev_get_drvdata(&service->dev); |
| 95 | |
| 96 | WARN_ON(priv->running); |
| 97 | WARN_ON(!list_empty_careful(&priv->recv_queue)); |
| 98 | WARN_ON(waitqueue_active(&priv->recv_wq)); |
| 99 | |
| 100 | vs_devio_priv_put(priv); |
| 101 | |
| 102 | return 0; |
| 103 | } |
| 104 | |
| 105 | static int |
| 106 | vs_devio_service_receive(struct vs_service_device *service, |
| 107 | struct vs_mbuf *mbuf) |
| 108 | { |
| 109 | struct vs_devio_priv *priv = dev_get_drvdata(&service->dev); |
| 110 | |
| 111 | WARN_ON(!priv->running); |
| 112 | |
| 113 | spin_lock(&priv->recv_wq.lock); |
| 114 | list_add_tail(&mbuf->queue, &priv->recv_queue); |
| 115 | wake_up_locked(&priv->recv_wq); |
| 116 | spin_unlock(&priv->recv_wq.lock); |
| 117 | |
| 118 | return 0; |
| 119 | } |
| 120 | |
| 121 | static void |
| 122 | vs_devio_service_notify(struct vs_service_device *service, u32 flags) |
| 123 | { |
| 124 | struct vs_devio_priv *priv = dev_get_drvdata(&service->dev); |
| 125 | int old, cur; |
| 126 | |
| 127 | WARN_ON(!priv->running); |
| 128 | |
| 129 | if (!flags) |
| 130 | return; |
| 131 | |
| 132 | /* open-coded atomic_or() */ |
| 133 | cur = atomic_read(&priv->notify_pending); |
| 134 | while ((old = atomic_cmpxchg(&priv->notify_pending, |
| 135 | cur, cur | flags)) != cur) |
| 136 | cur = old; |
| 137 | |
| 138 | wake_up(&priv->recv_wq); |
| 139 | } |
| 140 | |
| 141 | static void |
| 142 | vs_devio_service_start(struct vs_service_device *service) |
| 143 | { |
| 144 | struct vs_devio_priv *priv = dev_get_drvdata(&service->dev); |
| 145 | |
| 146 | if (!priv->reset) { |
| 147 | WARN_ON(priv->running); |
| 148 | priv->running = true; |
| 149 | wake_up(&service->quota_wq); |
| 150 | } |
| 151 | } |
| 152 | |
| 153 | static void |
| 154 | vs_devio_service_reset(struct vs_service_device *service) |
| 155 | { |
| 156 | struct vs_devio_priv *priv = dev_get_drvdata(&service->dev); |
| 157 | struct vs_mbuf *mbuf, *tmp; |
| 158 | |
| 159 | WARN_ON(!priv->running && !priv->reset); |
| 160 | |
| 161 | /* |
| 162 | * Mark the service as being in reset. This flag can never be cleared |
| 163 | * on an open device; the user must acknowledge the reset by closing |
| 164 | * and reopening the device. |
| 165 | */ |
| 166 | priv->reset = true; |
| 167 | priv->running = false; |
| 168 | |
| 169 | spin_lock_irq(&priv->recv_wq.lock); |
| 170 | list_for_each_entry_safe(mbuf, tmp, &priv->recv_queue, queue) |
| 171 | vs_service_free_mbuf(service, mbuf); |
| 172 | INIT_LIST_HEAD(&priv->recv_queue); |
| 173 | spin_unlock_irq(&priv->recv_wq.lock); |
| 174 | wake_up_all(&priv->recv_wq); |
| 175 | } |
| 176 | |
| 177 | /* |
| 178 | * This driver will be registered by the core server module, which must also |
| 179 | * set its bus and owner function pointers. |
| 180 | */ |
| 181 | struct vs_service_driver vs_devio_server_driver = { |
| 182 | /* No protocol, so the normal bus match will never bind this. */ |
| 183 | .protocol = NULL, |
| 184 | .is_server = true, |
| 185 | .rx_atomic = true, |
| 186 | |
| 187 | .probe = vs_devio_service_probe, |
| 188 | .remove = vs_devio_service_remove, |
| 189 | .receive = vs_devio_service_receive, |
| 190 | .notify = vs_devio_service_notify, |
| 191 | .start = vs_devio_service_start, |
| 192 | .reset = vs_devio_service_reset, |
| 193 | |
| 194 | /* |
| 195 | * Set reasonable default quotas. These can be overridden by passing |
| 196 | * nonzero values to IOCTL_VS_BIND_SERVER, which will set the |
| 197 | * service's *_quota_set fields. |
| 198 | */ |
| 199 | .in_quota_min = 1, |
| 200 | .in_quota_best = 8, |
| 201 | .out_quota_min = 1, |
| 202 | .out_quota_best = 8, |
| 203 | |
| 204 | /* Mark the notify counts as invalid; the service's will be used. */ |
| 205 | .in_notify_count = (unsigned)-1, |
| 206 | .out_notify_count = (unsigned)-1, |
| 207 | |
| 208 | .driver = { |
| 209 | .name = "vservices-server-devio", |
| 210 | .owner = NULL, /* set by core server */ |
| 211 | .bus = NULL, /* set by core server */ |
| 212 | .suppress_bind_attrs = true, /* see vs_devio_poll */ |
| 213 | }, |
| 214 | }; |
| 215 | EXPORT_SYMBOL_GPL(vs_devio_server_driver); |
| 216 | |
| 217 | static int |
| 218 | vs_devio_bind_server(struct vs_service_device *service, |
| 219 | struct vs_ioctl_bind *bind) |
| 220 | { |
| 221 | int ret = -ENODEV; |
| 222 | |
| 223 | /* Ensure the server module is loaded and the driver is registered. */ |
| 224 | if (!try_module_get(vs_devio_server_driver.driver.owner)) |
| 225 | goto fail_module_get; |
| 226 | |
| 227 | device_lock(&service->dev); |
| 228 | ret = -EBUSY; |
| 229 | if (service->dev.driver != NULL) |
| 230 | goto fail_device_unbound; |
| 231 | |
| 232 | /* Set up the quota and notify counts. */ |
| 233 | service->in_quota_set = bind->recv_quota; |
| 234 | service->out_quota_set = bind->send_quota; |
| 235 | service->notify_send_bits = bind->send_notify_bits; |
| 236 | service->notify_recv_bits = bind->recv_notify_bits; |
| 237 | |
| 238 | /* Manually probe the driver. */ |
| 239 | service->dev.driver = &vs_devio_server_driver.driver; |
| 240 | ret = service->dev.bus->probe(&service->dev); |
| 241 | if (ret < 0) |
| 242 | goto fail_probe_driver; |
| 243 | |
| 244 | ret = device_bind_driver(&service->dev); |
| 245 | if (ret < 0) |
| 246 | goto fail_bind_driver; |
| 247 | |
| 248 | /* Pass the allocated quotas back to the user. */ |
| 249 | bind->recv_quota = service->recv_quota; |
| 250 | bind->send_quota = service->send_quota; |
| 251 | bind->msg_size = vs_service_max_mbuf_size(service); |
| 252 | |
| 253 | device_unlock(&service->dev); |
| 254 | module_put(vs_devio_server_driver.driver.owner); |
| 255 | |
| 256 | return 0; |
| 257 | |
| 258 | fail_bind_driver: |
| 259 | ret = service->dev.bus->remove(&service->dev); |
| 260 | fail_probe_driver: |
| 261 | service->dev.driver = NULL; |
| 262 | fail_device_unbound: |
| 263 | device_unlock(&service->dev); |
| 264 | module_put(vs_devio_server_driver.driver.owner); |
| 265 | fail_module_get: |
| 266 | return ret; |
| 267 | } |
| 268 | |
| 269 | /* |
| 270 | * This driver will be registered by the core client module, which must also |
| 271 | * set its bus and owner pointers. |
| 272 | */ |
| 273 | struct vs_service_driver vs_devio_client_driver = { |
| 274 | /* No protocol, so the normal bus match will never bind this. */ |
| 275 | .protocol = NULL, |
| 276 | .is_server = false, |
| 277 | .rx_atomic = true, |
| 278 | |
| 279 | .probe = vs_devio_service_probe, |
| 280 | .remove = vs_devio_service_remove, |
| 281 | .receive = vs_devio_service_receive, |
| 282 | .notify = vs_devio_service_notify, |
| 283 | .start = vs_devio_service_start, |
| 284 | .reset = vs_devio_service_reset, |
| 285 | |
| 286 | .driver = { |
| 287 | .name = "vservices-client-devio", |
| 288 | .owner = NULL, /* set by core client */ |
| 289 | .bus = NULL, /* set by core client */ |
| 290 | .suppress_bind_attrs = true, /* see vs_devio_poll */ |
| 291 | }, |
| 292 | }; |
| 293 | EXPORT_SYMBOL_GPL(vs_devio_client_driver); |
| 294 | |
| 295 | static int |
| 296 | vs_devio_bind_client(struct vs_service_device *service, |
| 297 | struct vs_ioctl_bind *bind) |
| 298 | { |
| 299 | int ret = -ENODEV; |
| 300 | |
| 301 | /* Ensure the client module is loaded and the driver is registered. */ |
| 302 | if (!try_module_get(vs_devio_client_driver.driver.owner)) |
| 303 | goto fail_module_get; |
| 304 | |
| 305 | device_lock(&service->dev); |
| 306 | ret = -EBUSY; |
| 307 | if (service->dev.driver != NULL) |
| 308 | goto fail_device_unbound; |
| 309 | |
| 310 | /* Manually probe the driver. */ |
| 311 | service->dev.driver = &vs_devio_client_driver.driver; |
| 312 | ret = service->dev.bus->probe(&service->dev); |
| 313 | if (ret < 0) |
| 314 | goto fail_probe_driver; |
| 315 | |
| 316 | ret = device_bind_driver(&service->dev); |
| 317 | if (ret < 0) |
| 318 | goto fail_bind_driver; |
| 319 | |
| 320 | /* Pass the allocated quotas back to the user. */ |
| 321 | bind->recv_quota = service->recv_quota; |
| 322 | bind->send_quota = service->send_quota; |
| 323 | bind->msg_size = vs_service_max_mbuf_size(service); |
| 324 | bind->send_notify_bits = service->notify_send_bits; |
| 325 | bind->recv_notify_bits = service->notify_recv_bits; |
| 326 | |
| 327 | device_unlock(&service->dev); |
| 328 | module_put(vs_devio_client_driver.driver.owner); |
| 329 | |
| 330 | return 0; |
| 331 | |
| 332 | fail_bind_driver: |
| 333 | ret = service->dev.bus->remove(&service->dev); |
| 334 | fail_probe_driver: |
| 335 | service->dev.driver = NULL; |
| 336 | fail_device_unbound: |
| 337 | device_unlock(&service->dev); |
| 338 | module_put(vs_devio_client_driver.driver.owner); |
| 339 | fail_module_get: |
| 340 | return ret; |
| 341 | } |
| 342 | |
| 343 | static struct vs_devio_priv * |
| 344 | vs_devio_priv_get_from_service(struct vs_service_device *service) |
| 345 | { |
| 346 | struct vs_devio_priv *priv = NULL; |
| 347 | struct device_driver *drv; |
| 348 | |
| 349 | if (!service) |
| 350 | return NULL; |
| 351 | |
| 352 | device_lock(&service->dev); |
| 353 | drv = service->dev.driver; |
| 354 | |
| 355 | if ((drv == &vs_devio_client_driver.driver) || |
| 356 | (drv == &vs_devio_server_driver.driver)) { |
| 357 | vs_service_state_lock(service); |
| 358 | priv = dev_get_drvdata(&service->dev); |
| 359 | if (priv) |
| 360 | kref_get(&priv->kref); |
| 361 | vs_service_state_unlock(service); |
| 362 | } |
| 363 | |
| 364 | device_unlock(&service->dev); |
| 365 | |
| 366 | return priv; |
| 367 | } |
| 368 | |
| 369 | static int |
| 370 | vs_devio_open(struct inode *inode, struct file *file) |
| 371 | { |
| 372 | struct vs_service_device *service; |
| 373 | |
| 374 | if (imajor(inode) != vservices_cdev_major) |
| 375 | return -ENODEV; |
| 376 | |
| 377 | service = vs_service_lookup_by_devt(inode->i_rdev); |
| 378 | if (!service) |
| 379 | return -ENODEV; |
| 380 | |
| 381 | file->private_data = service; |
| 382 | |
| 383 | return 0; |
| 384 | } |
| 385 | |
| 386 | static int |
| 387 | vs_devio_release(struct inode *inode, struct file *file) |
| 388 | { |
| 389 | struct vs_service_device *service = file->private_data; |
| 390 | |
| 391 | if (service) { |
| 392 | struct vs_devio_priv *priv = |
| 393 | vs_devio_priv_get_from_service(service); |
| 394 | |
| 395 | if (priv) { |
| 396 | device_release_driver(&service->dev); |
| 397 | vs_devio_priv_put(priv); |
| 398 | } |
| 399 | |
| 400 | file->private_data = NULL; |
| 401 | vs_put_service(service); |
| 402 | } |
| 403 | |
| 404 | return 0; |
| 405 | } |
| 406 | |
| 407 | static struct iovec * |
| 408 | vs_devio_check_iov(struct vs_ioctl_iovec *io, bool is_send, ssize_t *total) |
| 409 | { |
| 410 | struct iovec *iov; |
| 411 | unsigned i; |
| 412 | int ret; |
| 413 | |
| 414 | if (io->iovcnt > UIO_MAXIOV) |
| 415 | return ERR_PTR(-EINVAL); |
| 416 | |
| 417 | iov = kmalloc(sizeof(*iov) * io->iovcnt, GFP_KERNEL); |
| 418 | if (!iov) |
| 419 | return ERR_PTR(-ENOMEM); |
| 420 | |
| 421 | if (copy_from_user(iov, io->iov, sizeof(*iov) * io->iovcnt)) { |
| 422 | ret = -EFAULT; |
| 423 | goto fail; |
| 424 | } |
| 425 | |
| 426 | *total = 0; |
| 427 | for (i = 0; i < io->iovcnt; i++) { |
| 428 | ssize_t iov_len = (ssize_t)iov[i].iov_len; |
| 429 | |
| 430 | if (iov_len > MAX_RW_COUNT - *total) { |
| 431 | ret = -EINVAL; |
| 432 | goto fail; |
| 433 | } |
| 434 | |
| 435 | if (!access_ok(is_send ? VERIFY_READ : VERIFY_WRITE, |
| 436 | iov[i].iov_base, iov_len)) { |
| 437 | ret = -EFAULT; |
| 438 | goto fail; |
| 439 | } |
| 440 | |
| 441 | *total += iov_len; |
| 442 | } |
| 443 | |
| 444 | return iov; |
| 445 | |
| 446 | fail: |
| 447 | kfree(iov); |
| 448 | return ERR_PTR(ret); |
| 449 | } |
| 450 | |
| 451 | static ssize_t |
| 452 | vs_devio_send(struct vs_service_device *service, struct iovec *iov, |
| 453 | size_t iovcnt, ssize_t to_send, bool nonblocking) |
| 454 | { |
| 455 | struct vs_mbuf *mbuf = NULL; |
| 456 | struct vs_devio_priv *priv; |
| 457 | unsigned i; |
| 458 | ssize_t offset = 0; |
| 459 | ssize_t ret; |
| 460 | DEFINE_WAIT(wait); |
| 461 | |
| 462 | priv = vs_devio_priv_get_from_service(service); |
| 463 | ret = -ENODEV; |
| 464 | if (!priv) |
| 465 | goto fail_priv_get; |
| 466 | |
| 467 | vs_service_state_lock(service); |
| 468 | |
| 469 | /* |
| 470 | * Waiting alloc. We must open-code this because there is no real |
| 471 | * state structure or base state. |
| 472 | */ |
| 473 | ret = 0; |
| 474 | while (!vs_service_send_mbufs_available(service)) { |
| 475 | if (nonblocking) { |
| 476 | ret = -EAGAIN; |
| 477 | break; |
| 478 | } |
| 479 | |
| 480 | if (signal_pending(current)) { |
| 481 | ret = -ERESTARTSYS; |
| 482 | break; |
| 483 | } |
| 484 | |
| 485 | prepare_to_wait_exclusive(&service->quota_wq, &wait, |
| 486 | TASK_INTERRUPTIBLE); |
| 487 | |
| 488 | vs_service_state_unlock(service); |
| 489 | schedule(); |
| 490 | vs_service_state_lock(service); |
| 491 | |
| 492 | if (priv->reset) { |
| 493 | ret = -ECONNRESET; |
| 494 | break; |
| 495 | } |
| 496 | |
| 497 | if (!priv->running) { |
| 498 | ret = -ENOTCONN; |
| 499 | break; |
| 500 | } |
| 501 | } |
| 502 | finish_wait(&service->quota_wq, &wait); |
| 503 | |
| 504 | if (ret) |
| 505 | goto fail_alloc; |
| 506 | |
| 507 | mbuf = vs_service_alloc_mbuf(service, to_send, GFP_KERNEL); |
| 508 | if (IS_ERR(mbuf)) { |
| 509 | ret = PTR_ERR(mbuf); |
| 510 | goto fail_alloc; |
| 511 | } |
| 512 | |
| 513 | /* Ready to send; copy data into the mbuf. */ |
| 514 | ret = -EFAULT; |
| 515 | for (i = 0; i < iovcnt; i++) { |
| 516 | if (copy_from_user(mbuf->data + offset, iov[i].iov_base, |
| 517 | iov[i].iov_len)) |
| 518 | goto fail_copy; |
| 519 | offset += iov[i].iov_len; |
| 520 | } |
| 521 | mbuf->size = to_send; |
| 522 | |
| 523 | /* Send the message. */ |
| 524 | ret = vs_service_send(service, mbuf); |
| 525 | if (ret < 0) |
| 526 | goto fail_send; |
| 527 | |
| 528 | /* Wake the next waiter, if there's more quota available. */ |
| 529 | if (waitqueue_active(&service->quota_wq) && |
| 530 | vs_service_send_mbufs_available(service) > 0) |
| 531 | wake_up(&service->quota_wq); |
| 532 | |
| 533 | vs_service_state_unlock(service); |
| 534 | vs_devio_priv_put(priv); |
| 535 | |
| 536 | return to_send; |
| 537 | |
| 538 | fail_send: |
| 539 | fail_copy: |
| 540 | vs_service_free_mbuf(service, mbuf); |
| 541 | wake_up(&service->quota_wq); |
| 542 | fail_alloc: |
| 543 | vs_service_state_unlock(service); |
| 544 | vs_devio_priv_put(priv); |
| 545 | fail_priv_get: |
| 546 | return ret; |
| 547 | } |
| 548 | |
| 549 | static ssize_t |
| 550 | vs_devio_recv(struct vs_service_device *service, struct iovec *iov, |
| 551 | size_t iovcnt, u32 *notify_bits, ssize_t recv_space, |
| 552 | bool nonblocking) |
| 553 | { |
| 554 | struct vs_mbuf *mbuf = NULL; |
| 555 | struct vs_devio_priv *priv; |
| 556 | unsigned i; |
| 557 | ssize_t offset = 0; |
| 558 | ssize_t ret; |
| 559 | DEFINE_WAIT(wait); |
| 560 | |
| 561 | priv = vs_devio_priv_get_from_service(service); |
| 562 | ret = -ENODEV; |
| 563 | if (!priv) |
| 564 | goto fail_priv_get; |
| 565 | |
| 566 | /* Take the recv_wq lock, which also protects recv_queue. */ |
| 567 | spin_lock_irq(&priv->recv_wq.lock); |
| 568 | |
| 569 | /* Wait for a message, notification, or reset. */ |
| 570 | ret = wait_event_interruptible_exclusive_locked_irq(priv->recv_wq, |
| 571 | !list_empty(&priv->recv_queue) || priv->reset || |
| 572 | atomic_read(&priv->notify_pending) || nonblocking); |
| 573 | |
| 574 | if (priv->reset) |
| 575 | ret = -ECONNRESET; /* Service reset */ |
| 576 | else if (!ret && list_empty(&priv->recv_queue)) |
| 577 | ret = -EAGAIN; /* Nonblocking, or notification */ |
| 578 | |
| 579 | if (ret < 0) { |
| 580 | spin_unlock_irq(&priv->recv_wq.lock); |
| 581 | goto no_mbuf; |
| 582 | } |
| 583 | |
| 584 | /* Take the first mbuf from the list, and check its size. */ |
| 585 | mbuf = list_first_entry(&priv->recv_queue, struct vs_mbuf, queue); |
| 586 | if (mbuf->size > recv_space) { |
| 587 | spin_unlock_irq(&priv->recv_wq.lock); |
| 588 | ret = -EMSGSIZE; |
| 589 | goto fail_msg_size; |
| 590 | } |
| 591 | list_del_init(&mbuf->queue); |
| 592 | |
| 593 | spin_unlock_irq(&priv->recv_wq.lock); |
| 594 | |
| 595 | /* Copy to user. */ |
| 596 | ret = -EFAULT; |
| 597 | for (i = 0; (mbuf->size > offset) && (i < iovcnt); i++) { |
| 598 | size_t len = min(mbuf->size - offset, iov[i].iov_len); |
| 599 | if (copy_to_user(iov[i].iov_base, mbuf->data + offset, len)) |
| 600 | goto fail_copy; |
| 601 | offset += len; |
| 602 | } |
| 603 | ret = offset; |
| 604 | |
| 605 | no_mbuf: |
| 606 | /* |
| 607 | * Read and clear the pending notification bits. If any notifications |
| 608 | * are received, don't return an error, even if we failed to receive a |
| 609 | * message. |
| 610 | */ |
| 611 | *notify_bits = atomic_xchg(&priv->notify_pending, 0); |
| 612 | if ((ret < 0) && *notify_bits) |
| 613 | ret = 0; |
| 614 | |
| 615 | fail_copy: |
| 616 | if (mbuf) |
| 617 | vs_service_free_mbuf(service, mbuf); |
| 618 | fail_msg_size: |
| 619 | vs_devio_priv_put(priv); |
| 620 | fail_priv_get: |
| 621 | return ret; |
| 622 | } |
| 623 | |
| 624 | static int |
| 625 | vs_devio_check_perms(struct file *file, unsigned flags) |
| 626 | { |
| 627 | if ((flags & MAY_READ) & !(file->f_mode & FMODE_READ)) |
| 628 | return -EBADF; |
| 629 | |
| 630 | if ((flags & MAY_WRITE) & !(file->f_mode & FMODE_WRITE)) |
| 631 | return -EBADF; |
| 632 | |
| 633 | return security_file_permission(file, flags); |
| 634 | } |
| 635 | |
| 636 | static long |
| 637 | vs_devio_ioctl(struct file *file, unsigned int cmd, unsigned long arg) |
| 638 | { |
| 639 | void __user *ptr = (void __user *)arg; |
| 640 | struct vs_service_device *service = file->private_data; |
| 641 | struct vs_ioctl_bind bind; |
| 642 | struct vs_ioctl_iovec io; |
| 643 | u32 flags; |
| 644 | long ret; |
| 645 | ssize_t iov_total; |
| 646 | struct iovec *iov; |
| 647 | |
| 648 | if (!service) |
| 649 | return -ENODEV; |
| 650 | |
| 651 | switch (cmd) { |
| 652 | case IOCTL_VS_RESET_SERVICE: |
| 653 | ret = vs_devio_check_perms(file, MAY_WRITE); |
| 654 | if (ret < 0) |
| 655 | break; |
| 656 | ret = vs_service_reset(service, service); |
| 657 | break; |
| 658 | case IOCTL_VS_GET_NAME: |
| 659 | ret = vs_devio_check_perms(file, MAY_READ); |
| 660 | if (ret < 0) |
| 661 | break; |
| 662 | if (service->name != NULL) { |
| 663 | size_t len = strnlen(service->name, |
| 664 | _IOC_SIZE(IOCTL_VS_GET_NAME) - 1); |
| 665 | if (copy_to_user(ptr, service->name, len + 1)) |
| 666 | ret = -EFAULT; |
| 667 | } else { |
| 668 | ret = -EINVAL; |
| 669 | } |
| 670 | break; |
| 671 | case IOCTL_VS_GET_PROTOCOL: |
| 672 | ret = vs_devio_check_perms(file, MAY_READ); |
| 673 | if (ret < 0) |
| 674 | break; |
| 675 | if (service->protocol != NULL) { |
| 676 | size_t len = strnlen(service->protocol, |
| 677 | _IOC_SIZE(IOCTL_VS_GET_PROTOCOL) - 1); |
| 678 | if (copy_to_user(ptr, service->protocol, len + 1)) |
| 679 | ret = -EFAULT; |
| 680 | } else { |
| 681 | ret = -EINVAL; |
| 682 | } |
| 683 | break; |
| 684 | case IOCTL_VS_BIND_CLIENT: |
| 685 | ret = vs_devio_check_perms(file, MAY_EXEC); |
| 686 | if (ret < 0) |
| 687 | break; |
| 688 | ret = vs_devio_bind_client(service, &bind); |
| 689 | if (!ret && copy_to_user(ptr, &bind, sizeof(bind))) |
| 690 | ret = -EFAULT; |
| 691 | break; |
| 692 | case IOCTL_VS_BIND_SERVER: |
| 693 | ret = vs_devio_check_perms(file, MAY_EXEC); |
| 694 | if (ret < 0) |
| 695 | break; |
| 696 | if (copy_from_user(&bind, ptr, sizeof(bind))) { |
| 697 | ret = -EFAULT; |
| 698 | break; |
| 699 | } |
| 700 | ret = vs_devio_bind_server(service, &bind); |
| 701 | if (!ret && copy_to_user(ptr, &bind, sizeof(bind))) |
| 702 | ret = -EFAULT; |
| 703 | break; |
| 704 | case IOCTL_VS_NOTIFY: |
| 705 | ret = vs_devio_check_perms(file, MAY_WRITE); |
| 706 | if (ret < 0) |
| 707 | break; |
| 708 | if (copy_from_user(&flags, ptr, sizeof(flags))) { |
| 709 | ret = -EFAULT; |
| 710 | break; |
| 711 | } |
| 712 | ret = vs_service_notify(service, flags); |
| 713 | break; |
| 714 | case IOCTL_VS_SEND: |
| 715 | ret = vs_devio_check_perms(file, MAY_WRITE); |
| 716 | if (ret < 0) |
| 717 | break; |
| 718 | if (copy_from_user(&io, ptr, sizeof(io))) { |
| 719 | ret = -EFAULT; |
| 720 | break; |
| 721 | } |
| 722 | |
| 723 | iov = vs_devio_check_iov(&io, true, &iov_total); |
| 724 | if (IS_ERR(iov)) { |
| 725 | ret = PTR_ERR(iov); |
| 726 | break; |
| 727 | } |
| 728 | |
| 729 | ret = vs_devio_send(service, iov, io.iovcnt, iov_total, |
| 730 | file->f_flags & O_NONBLOCK); |
| 731 | kfree(iov); |
| 732 | break; |
| 733 | case IOCTL_VS_RECV: |
| 734 | ret = vs_devio_check_perms(file, MAY_READ); |
| 735 | if (ret < 0) |
| 736 | break; |
| 737 | if (copy_from_user(&io, ptr, sizeof(io))) { |
| 738 | ret = -EFAULT; |
| 739 | break; |
| 740 | } |
| 741 | |
| 742 | iov = vs_devio_check_iov(&io, true, &iov_total); |
| 743 | if (IS_ERR(iov)) { |
| 744 | ret = PTR_ERR(iov); |
| 745 | break; |
| 746 | } |
| 747 | |
| 748 | ret = vs_devio_recv(service, iov, io.iovcnt, |
| 749 | &io.notify_bits, iov_total, |
| 750 | file->f_flags & O_NONBLOCK); |
| 751 | kfree(iov); |
| 752 | |
| 753 | if (ret >= 0) { |
| 754 | u32 __user *notify_bits_ptr = ptr + offsetof( |
| 755 | struct vs_ioctl_iovec, notify_bits); |
| 756 | if (copy_to_user(notify_bits_ptr, &io.notify_bits, |
| 757 | sizeof(io.notify_bits))) |
| 758 | ret = -EFAULT; |
| 759 | } |
| 760 | break; |
| 761 | default: |
| 762 | dev_dbg(&service->dev, "Unknown ioctl %#x, arg: %lx\n", cmd, |
| 763 | arg); |
| 764 | ret = -ENOSYS; |
| 765 | break; |
| 766 | } |
| 767 | |
| 768 | return ret; |
| 769 | } |
| 770 | |
| 771 | #ifdef CONFIG_COMPAT |
| 772 | |
| 773 | struct vs_compat_ioctl_bind { |
| 774 | __u32 send_quota; |
| 775 | __u32 recv_quota; |
| 776 | __u32 send_notify_bits; |
| 777 | __u32 recv_notify_bits; |
| 778 | compat_size_t msg_size; |
| 779 | }; |
| 780 | |
| 781 | #define compat_ioctl_bind_conv(dest, src) ({ \ |
| 782 | dest.send_quota = src.send_quota; \ |
| 783 | dest.recv_quota = src.recv_quota; \ |
| 784 | dest.send_notify_bits = src.send_notify_bits; \ |
| 785 | dest.recv_notify_bits = src.recv_notify_bits; \ |
| 786 | dest.msg_size = (compat_size_t)src.msg_size; \ |
| 787 | }) |
| 788 | |
| 789 | #define COMPAT_IOCTL_VS_BIND_CLIENT _IOR('4', 3, struct vs_compat_ioctl_bind) |
| 790 | #define COMPAT_IOCTL_VS_BIND_SERVER _IOWR('4', 4, struct vs_compat_ioctl_bind) |
| 791 | |
| 792 | struct vs_compat_ioctl_iovec { |
| 793 | union { |
| 794 | __u32 iovcnt; /* input */ |
| 795 | __u32 notify_bits; /* output (recv only) */ |
| 796 | }; |
| 797 | compat_uptr_t iov; |
| 798 | }; |
| 799 | |
| 800 | #define COMPAT_IOCTL_VS_SEND \ |
| 801 | _IOW('4', 6, struct vs_compat_ioctl_iovec) |
| 802 | #define COMPAT_IOCTL_VS_RECV \ |
| 803 | _IOWR('4', 7, struct vs_compat_ioctl_iovec) |
| 804 | |
| 805 | static struct iovec * |
| 806 | vs_devio_check_compat_iov(struct vs_compat_ioctl_iovec *c_io, |
| 807 | bool is_send, ssize_t *total) |
| 808 | { |
| 809 | struct iovec *iov; |
| 810 | struct compat_iovec *c_iov; |
| 811 | |
| 812 | unsigned i; |
| 813 | int ret; |
| 814 | |
| 815 | if (c_io->iovcnt > UIO_MAXIOV) |
| 816 | return ERR_PTR(-EINVAL); |
| 817 | |
| 818 | c_iov = kzalloc(sizeof(*c_iov) * c_io->iovcnt, GFP_KERNEL); |
| 819 | if (!c_iov) |
| 820 | return ERR_PTR(-ENOMEM); |
| 821 | |
| 822 | iov = kzalloc(sizeof(*iov) * c_io->iovcnt, GFP_KERNEL); |
| 823 | if (!iov) { |
| 824 | kfree(c_iov); |
| 825 | return ERR_PTR(-ENOMEM); |
| 826 | } |
| 827 | |
| 828 | if (copy_from_user(c_iov, (struct compat_iovec __user *) |
| 829 | compat_ptr(c_io->iov), sizeof(*c_iov) * c_io->iovcnt)) { |
| 830 | ret = -EFAULT; |
| 831 | goto fail; |
| 832 | } |
| 833 | |
| 834 | *total = 0; |
| 835 | for (i = 0; i < c_io->iovcnt; i++) { |
| 836 | ssize_t iov_len; |
| 837 | iov[i].iov_base = compat_ptr (c_iov[i].iov_base); |
| 838 | iov[i].iov_len = (compat_size_t) c_iov[i].iov_len; |
| 839 | |
| 840 | iov_len = (ssize_t)iov[i].iov_len; |
| 841 | |
| 842 | if (iov_len > MAX_RW_COUNT - *total) { |
| 843 | ret = -EINVAL; |
| 844 | goto fail; |
| 845 | } |
| 846 | |
| 847 | if (!access_ok(is_send ? VERIFY_READ : VERIFY_WRITE, |
| 848 | iov[i].iov_base, iov_len)) { |
| 849 | ret = -EFAULT; |
| 850 | goto fail; |
| 851 | } |
| 852 | |
| 853 | *total += iov_len; |
| 854 | } |
| 855 | |
| 856 | kfree (c_iov); |
| 857 | return iov; |
| 858 | |
| 859 | fail: |
| 860 | kfree(c_iov); |
| 861 | kfree(iov); |
| 862 | return ERR_PTR(ret); |
| 863 | } |
| 864 | |
| 865 | static long |
| 866 | vs_devio_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg) |
| 867 | { |
| 868 | void __user *ptr = (void __user *)arg; |
| 869 | struct vs_service_device *service = file->private_data; |
| 870 | struct vs_ioctl_bind bind; |
| 871 | struct vs_compat_ioctl_bind compat_bind; |
| 872 | struct vs_compat_ioctl_iovec compat_io; |
| 873 | long ret; |
| 874 | ssize_t iov_total; |
| 875 | struct iovec *iov; |
| 876 | |
| 877 | if (!service) |
| 878 | return -ENODEV; |
| 879 | |
| 880 | switch (cmd) { |
| 881 | case IOCTL_VS_RESET_SERVICE: |
| 882 | case IOCTL_VS_GET_NAME: |
| 883 | case IOCTL_VS_GET_PROTOCOL: |
| 884 | return vs_devio_ioctl (file, cmd, arg); |
| 885 | case COMPAT_IOCTL_VS_SEND: |
| 886 | ret = vs_devio_check_perms(file, MAY_WRITE); |
| 887 | if (ret < 0) |
| 888 | break; |
| 889 | if (copy_from_user(&compat_io, ptr, sizeof(compat_io))) { |
| 890 | ret = -EFAULT; |
| 891 | break; |
| 892 | } |
| 893 | |
| 894 | iov = vs_devio_check_compat_iov(&compat_io, true, &iov_total); |
| 895 | if (IS_ERR(iov)) { |
| 896 | ret = PTR_ERR(iov); |
| 897 | break; |
| 898 | } |
| 899 | |
| 900 | ret = vs_devio_send(service, iov, compat_io.iovcnt, iov_total, |
| 901 | file->f_flags & O_NONBLOCK); |
| 902 | kfree(iov); |
| 903 | |
| 904 | break; |
| 905 | case COMPAT_IOCTL_VS_RECV: |
| 906 | ret = vs_devio_check_perms(file, MAY_READ); |
| 907 | if (ret < 0) |
| 908 | break; |
| 909 | if (copy_from_user(&compat_io, ptr, sizeof(compat_io))) { |
| 910 | ret = -EFAULT; |
| 911 | break; |
| 912 | } |
| 913 | |
| 914 | iov = vs_devio_check_compat_iov(&compat_io, true, &iov_total); |
| 915 | if (IS_ERR(iov)) { |
| 916 | ret = PTR_ERR(iov); |
| 917 | break; |
| 918 | } |
| 919 | |
| 920 | ret = vs_devio_recv(service, iov, compat_io.iovcnt, |
| 921 | &compat_io.notify_bits, iov_total, |
| 922 | file->f_flags & O_NONBLOCK); |
| 923 | kfree(iov); |
| 924 | |
| 925 | if (ret >= 0) { |
| 926 | u32 __user *notify_bits_ptr = ptr + offsetof( |
| 927 | struct vs_compat_ioctl_iovec, notify_bits); |
| 928 | if (copy_to_user(notify_bits_ptr, &compat_io.notify_bits, |
| 929 | sizeof(compat_io.notify_bits))) |
| 930 | ret = -EFAULT; |
| 931 | } |
| 932 | break; |
| 933 | case COMPAT_IOCTL_VS_BIND_CLIENT: |
| 934 | ret = vs_devio_check_perms(file, MAY_EXEC); |
| 935 | if (ret < 0) |
| 936 | break; |
| 937 | ret = vs_devio_bind_client(service, &bind); |
| 938 | compat_ioctl_bind_conv(compat_bind, bind); |
| 939 | if (!ret && copy_to_user(ptr, &compat_bind, |
| 940 | sizeof(compat_bind))) |
| 941 | ret = -EFAULT; |
| 942 | break; |
| 943 | case COMPAT_IOCTL_VS_BIND_SERVER: |
| 944 | ret = vs_devio_check_perms(file, MAY_EXEC); |
| 945 | if (ret < 0) |
| 946 | break; |
| 947 | if (copy_from_user(&compat_bind, ptr, sizeof(compat_bind))) { |
| 948 | ret = -EFAULT; |
| 949 | break; |
| 950 | } |
| 951 | compat_ioctl_bind_conv(bind, compat_bind); |
| 952 | ret = vs_devio_bind_server(service, &bind); |
| 953 | compat_ioctl_bind_conv(compat_bind, bind); |
| 954 | if (!ret && copy_to_user(ptr, &compat_bind, |
| 955 | sizeof(compat_bind))) |
| 956 | ret = -EFAULT; |
| 957 | break; |
| 958 | default: |
| 959 | dev_dbg(&service->dev, "Unknown ioctl %#x, arg: %lx\n", cmd, |
| 960 | arg); |
| 961 | ret = -ENOSYS; |
| 962 | break; |
| 963 | } |
| 964 | |
| 965 | return ret; |
| 966 | } |
| 967 | |
| 968 | #endif /* CONFIG_COMPAT */ |
| 969 | |
| 970 | static unsigned int |
| 971 | vs_devio_poll(struct file *file, struct poll_table_struct *wait) |
| 972 | { |
| 973 | struct vs_service_device *service = file->private_data; |
| 974 | struct vs_devio_priv *priv = vs_devio_priv_get_from_service(service); |
| 975 | unsigned int flags = 0; |
| 976 | |
| 977 | poll_wait(file, &service->quota_wq, wait); |
| 978 | |
| 979 | if (priv) { |
| 980 | /* |
| 981 | * Note: there is no way for us to ensure that all poll |
| 982 | * waiters on a given workqueue have gone away, other than to |
| 983 | * actually close the file. So, this poll_wait() is only safe |
| 984 | * if we never release our claim on the service before the |
| 985 | * file is closed. |
| 986 | * |
| 987 | * We try to guarantee this by only unbinding the devio driver |
| 988 | * on close, and setting suppress_bind_attrs in the driver so |
| 989 | * root can't unbind us with sysfs. |
| 990 | */ |
| 991 | poll_wait(file, &priv->recv_wq, wait); |
| 992 | |
| 993 | if (priv->reset) { |
| 994 | /* Service reset; raise poll error. */ |
| 995 | flags |= POLLERR | POLLHUP; |
| 996 | } else if (priv->running) { |
| 997 | if (!list_empty_careful(&priv->recv_queue)) |
| 998 | flags |= POLLRDNORM | POLLIN; |
| 999 | if (atomic_read(&priv->notify_pending)) |
| 1000 | flags |= POLLRDNORM | POLLIN; |
| 1001 | if (vs_service_send_mbufs_available(service) > 0) |
| 1002 | flags |= POLLWRNORM | POLLOUT; |
| 1003 | } |
| 1004 | |
| 1005 | vs_devio_priv_put(priv); |
| 1006 | } else { |
| 1007 | /* No driver attached. Return error flags. */ |
| 1008 | flags |= POLLERR | POLLHUP; |
| 1009 | } |
| 1010 | |
| 1011 | return flags; |
| 1012 | } |
| 1013 | |
| 1014 | static const struct file_operations vs_fops = { |
| 1015 | .owner = THIS_MODULE, |
| 1016 | .open = vs_devio_open, |
| 1017 | .release = vs_devio_release, |
| 1018 | .unlocked_ioctl = vs_devio_ioctl, |
| 1019 | #ifdef CONFIG_COMPAT |
| 1020 | .compat_ioctl = vs_devio_compat_ioctl, |
| 1021 | #endif |
| 1022 | .poll = vs_devio_poll, |
| 1023 | }; |
| 1024 | |
| 1025 | int vservices_cdev_major; |
| 1026 | static struct cdev vs_cdev; |
| 1027 | |
| 1028 | int __init |
| 1029 | vs_devio_init(void) |
| 1030 | { |
| 1031 | dev_t dev; |
| 1032 | int r; |
| 1033 | |
| 1034 | r = alloc_chrdev_region(&dev, 0, VSERVICES_DEVICE_MAX, |
| 1035 | "vs_service"); |
| 1036 | if (r < 0) |
| 1037 | goto fail_alloc_chrdev; |
| 1038 | vservices_cdev_major = MAJOR(dev); |
| 1039 | |
| 1040 | cdev_init(&vs_cdev, &vs_fops); |
| 1041 | r = cdev_add(&vs_cdev, dev, VSERVICES_DEVICE_MAX); |
| 1042 | if (r < 0) |
| 1043 | goto fail_cdev_add; |
| 1044 | |
| 1045 | return 0; |
| 1046 | |
| 1047 | fail_cdev_add: |
| 1048 | unregister_chrdev_region(dev, VSERVICES_DEVICE_MAX); |
| 1049 | fail_alloc_chrdev: |
| 1050 | return r; |
| 1051 | } |
| 1052 | |
| 1053 | void __exit |
| 1054 | vs_devio_exit(void) |
| 1055 | { |
| 1056 | cdev_del(&vs_cdev); |
| 1057 | unregister_chrdev_region(MKDEV(vservices_cdev_major, 0), |
| 1058 | VSERVICES_DEVICE_MAX); |
| 1059 | } |