Commit | Line | Data |
---|---|---|
cba3345c AW |
1 | /* |
2 | * VFIO core | |
3 | * | |
4 | * Copyright (C) 2012 Red Hat, Inc. All rights reserved. | |
5 | * Author: Alex Williamson <alex.williamson@redhat.com> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | * | |
11 | * Derived from original vfio: | |
12 | * Copyright 2010 Cisco Systems, Inc. All rights reserved. | |
13 | * Author: Tom Lyon, pugs@cisco.com | |
14 | */ | |
15 | ||
16 | #include <linux/cdev.h> | |
17 | #include <linux/compat.h> | |
18 | #include <linux/device.h> | |
19 | #include <linux/file.h> | |
20 | #include <linux/anon_inodes.h> | |
21 | #include <linux/fs.h> | |
22 | #include <linux/idr.h> | |
23 | #include <linux/iommu.h> | |
24 | #include <linux/list.h> | |
d1099901 | 25 | #include <linux/miscdevice.h> |
cba3345c AW |
26 | #include <linux/module.h> |
27 | #include <linux/mutex.h> | |
5f096b14 | 28 | #include <linux/pci.h> |
9587f44a | 29 | #include <linux/rwsem.h> |
cba3345c AW |
30 | #include <linux/sched.h> |
31 | #include <linux/slab.h> | |
664e9386 | 32 | #include <linux/stat.h> |
cba3345c AW |
33 | #include <linux/string.h> |
34 | #include <linux/uaccess.h> | |
35 | #include <linux/vfio.h> | |
36 | #include <linux/wait.h> | |
37 | ||
38 | #define DRIVER_VERSION "0.3" | |
39 | #define DRIVER_AUTHOR "Alex Williamson <alex.williamson@redhat.com>" | |
40 | #define DRIVER_DESC "VFIO - User Level meta-driver" | |
41 | ||
42 | static struct vfio { | |
43 | struct class *class; | |
44 | struct list_head iommu_drivers_list; | |
45 | struct mutex iommu_drivers_lock; | |
46 | struct list_head group_list; | |
47 | struct idr group_idr; | |
48 | struct mutex group_lock; | |
49 | struct cdev group_cdev; | |
d1099901 | 50 | dev_t group_devt; |
cba3345c AW |
51 | wait_queue_head_t release_q; |
52 | } vfio; | |
53 | ||
54 | struct vfio_iommu_driver { | |
55 | const struct vfio_iommu_driver_ops *ops; | |
56 | struct list_head vfio_next; | |
57 | }; | |
58 | ||
59 | struct vfio_container { | |
60 | struct kref kref; | |
61 | struct list_head group_list; | |
9587f44a | 62 | struct rw_semaphore group_lock; |
cba3345c AW |
63 | struct vfio_iommu_driver *iommu_driver; |
64 | void *iommu_data; | |
03a76b60 | 65 | bool noiommu; |
cba3345c AW |
66 | }; |
67 | ||
60720a0f AW |
68 | struct vfio_unbound_dev { |
69 | struct device *dev; | |
70 | struct list_head unbound_next; | |
71 | }; | |
72 | ||
cba3345c AW |
73 | struct vfio_group { |
74 | struct kref kref; | |
75 | int minor; | |
76 | atomic_t container_users; | |
77 | struct iommu_group *iommu_group; | |
78 | struct vfio_container *container; | |
79 | struct list_head device_list; | |
80 | struct mutex device_lock; | |
81 | struct device *dev; | |
82 | struct notifier_block nb; | |
83 | struct list_head vfio_next; | |
84 | struct list_head container_next; | |
60720a0f AW |
85 | struct list_head unbound_list; |
86 | struct mutex unbound_lock; | |
6d6768c6 | 87 | atomic_t opened; |
03a76b60 | 88 | bool noiommu; |
cba3345c AW |
89 | }; |
90 | ||
91 | struct vfio_device { | |
92 | struct kref kref; | |
93 | struct device *dev; | |
94 | const struct vfio_device_ops *ops; | |
95 | struct vfio_group *group; | |
96 | struct list_head group_next; | |
97 | void *device_data; | |
98 | }; | |
99 | ||
03a76b60 AW |
100 | #ifdef CONFIG_VFIO_NOIOMMU |
101 | static bool noiommu __read_mostly; | |
102 | module_param_named(enable_unsafe_noiommu_mode, | |
103 | noiommu, bool, S_IRUGO | S_IWUSR); | |
104 | MODULE_PARM_DESC(enable_unsafe_noiommu_mode, "Enable UNSAFE, no-IOMMU mode. This mode provides no device isolation, no DMA translation, no host kernel protection, cannot be used for device assignment to virtual machines, requires RAWIO permissions, and will taint the kernel. If you do not know what this is for, step away. (default: false)"); | |
105 | #endif | |
106 | ||
107 | /* | |
108 | * vfio_iommu_group_{get,put} are only intended for VFIO bus driver probe | |
109 | * and remove functions, any use cases other than acquiring the first | |
110 | * reference for the purpose of calling vfio_add_group_dev() or removing | |
111 | * that symmetric reference after vfio_del_group_dev() should use the raw | |
112 | * iommu_group_{get,put} functions. In particular, vfio_iommu_group_put() | |
113 | * removes the device from the dummy group and cannot be nested. | |
114 | */ | |
115 | struct iommu_group *vfio_iommu_group_get(struct device *dev) | |
116 | { | |
117 | struct iommu_group *group; | |
118 | int __maybe_unused ret; | |
119 | ||
120 | group = iommu_group_get(dev); | |
121 | ||
122 | #ifdef CONFIG_VFIO_NOIOMMU | |
123 | /* | |
124 | * With noiommu enabled, an IOMMU group will be created for a device | |
125 | * that doesn't already have one and doesn't have an iommu_ops on their | |
16ab8a5c AW |
126 | * bus. We set iommudata simply to be able to identify these groups |
127 | * as special use and for reclamation later. | |
03a76b60 AW |
128 | */ |
129 | if (group || !noiommu || iommu_present(dev->bus)) | |
130 | return group; | |
131 | ||
132 | group = iommu_group_alloc(); | |
133 | if (IS_ERR(group)) | |
134 | return NULL; | |
135 | ||
136 | iommu_group_set_name(group, "vfio-noiommu"); | |
16ab8a5c | 137 | iommu_group_set_iommudata(group, &noiommu, NULL); |
03a76b60 AW |
138 | ret = iommu_group_add_device(group, dev); |
139 | iommu_group_put(group); | |
140 | if (ret) | |
141 | return NULL; | |
142 | ||
143 | /* | |
144 | * Where to taint? At this point we've added an IOMMU group for a | |
145 | * device that is not backed by iommu_ops, therefore any iommu_ | |
146 | * callback using iommu_ops can legitimately Oops. So, while we may | |
147 | * be about to give a DMA capable device to a user without IOMMU | |
148 | * protection, which is clearly taint-worthy, let's go ahead and do | |
149 | * it here. | |
150 | */ | |
151 | add_taint(TAINT_USER, LOCKDEP_STILL_OK); | |
152 | dev_warn(dev, "Adding kernel taint for vfio-noiommu group on device\n"); | |
153 | #endif | |
154 | ||
155 | return group; | |
156 | } | |
157 | EXPORT_SYMBOL_GPL(vfio_iommu_group_get); | |
158 | ||
159 | void vfio_iommu_group_put(struct iommu_group *group, struct device *dev) | |
160 | { | |
161 | #ifdef CONFIG_VFIO_NOIOMMU | |
16ab8a5c | 162 | if (iommu_group_get_iommudata(group) == &noiommu) |
03a76b60 AW |
163 | iommu_group_remove_device(dev); |
164 | #endif | |
165 | ||
166 | iommu_group_put(group); | |
167 | } | |
168 | EXPORT_SYMBOL_GPL(vfio_iommu_group_put); | |
169 | ||
170 | #ifdef CONFIG_VFIO_NOIOMMU | |
171 | static void *vfio_noiommu_open(unsigned long arg) | |
172 | { | |
173 | if (arg != VFIO_NOIOMMU_IOMMU) | |
174 | return ERR_PTR(-EINVAL); | |
175 | if (!capable(CAP_SYS_RAWIO)) | |
176 | return ERR_PTR(-EPERM); | |
177 | ||
178 | return NULL; | |
179 | } | |
180 | ||
181 | static void vfio_noiommu_release(void *iommu_data) | |
182 | { | |
183 | } | |
184 | ||
185 | static long vfio_noiommu_ioctl(void *iommu_data, | |
186 | unsigned int cmd, unsigned long arg) | |
187 | { | |
188 | if (cmd == VFIO_CHECK_EXTENSION) | |
189 | return noiommu && (arg == VFIO_NOIOMMU_IOMMU) ? 1 : 0; | |
190 | ||
191 | return -ENOTTY; | |
192 | } | |
193 | ||
03a76b60 AW |
194 | static int vfio_noiommu_attach_group(void *iommu_data, |
195 | struct iommu_group *iommu_group) | |
196 | { | |
16ab8a5c | 197 | return iommu_group_get_iommudata(iommu_group) == &noiommu ? 0 : -EINVAL; |
03a76b60 AW |
198 | } |
199 | ||
200 | static void vfio_noiommu_detach_group(void *iommu_data, | |
201 | struct iommu_group *iommu_group) | |
202 | { | |
203 | } | |
204 | ||
205 | static const struct vfio_iommu_driver_ops vfio_noiommu_ops = { | |
206 | .name = "vfio-noiommu", | |
207 | .owner = THIS_MODULE, | |
208 | .open = vfio_noiommu_open, | |
209 | .release = vfio_noiommu_release, | |
210 | .ioctl = vfio_noiommu_ioctl, | |
211 | .attach_group = vfio_noiommu_attach_group, | |
212 | .detach_group = vfio_noiommu_detach_group, | |
213 | }; | |
214 | #endif | |
215 | ||
216 | ||
cba3345c AW |
217 | /** |
218 | * IOMMU driver registration | |
219 | */ | |
220 | int vfio_register_iommu_driver(const struct vfio_iommu_driver_ops *ops) | |
221 | { | |
222 | struct vfio_iommu_driver *driver, *tmp; | |
223 | ||
224 | driver = kzalloc(sizeof(*driver), GFP_KERNEL); | |
225 | if (!driver) | |
226 | return -ENOMEM; | |
227 | ||
228 | driver->ops = ops; | |
229 | ||
230 | mutex_lock(&vfio.iommu_drivers_lock); | |
231 | ||
232 | /* Check for duplicates */ | |
233 | list_for_each_entry(tmp, &vfio.iommu_drivers_list, vfio_next) { | |
234 | if (tmp->ops == ops) { | |
235 | mutex_unlock(&vfio.iommu_drivers_lock); | |
236 | kfree(driver); | |
237 | return -EINVAL; | |
238 | } | |
239 | } | |
240 | ||
241 | list_add(&driver->vfio_next, &vfio.iommu_drivers_list); | |
242 | ||
243 | mutex_unlock(&vfio.iommu_drivers_lock); | |
244 | ||
245 | return 0; | |
246 | } | |
247 | EXPORT_SYMBOL_GPL(vfio_register_iommu_driver); | |
248 | ||
249 | void vfio_unregister_iommu_driver(const struct vfio_iommu_driver_ops *ops) | |
250 | { | |
251 | struct vfio_iommu_driver *driver; | |
252 | ||
253 | mutex_lock(&vfio.iommu_drivers_lock); | |
254 | list_for_each_entry(driver, &vfio.iommu_drivers_list, vfio_next) { | |
255 | if (driver->ops == ops) { | |
256 | list_del(&driver->vfio_next); | |
257 | mutex_unlock(&vfio.iommu_drivers_lock); | |
258 | kfree(driver); | |
259 | return; | |
260 | } | |
261 | } | |
262 | mutex_unlock(&vfio.iommu_drivers_lock); | |
263 | } | |
264 | EXPORT_SYMBOL_GPL(vfio_unregister_iommu_driver); | |
265 | ||
266 | /** | |
267 | * Group minor allocation/free - both called with vfio.group_lock held | |
268 | */ | |
269 | static int vfio_alloc_group_minor(struct vfio_group *group) | |
270 | { | |
d1099901 | 271 | return idr_alloc(&vfio.group_idr, group, 0, MINORMASK + 1, GFP_KERNEL); |
cba3345c AW |
272 | } |
273 | ||
274 | static void vfio_free_group_minor(int minor) | |
275 | { | |
276 | idr_remove(&vfio.group_idr, minor); | |
277 | } | |
278 | ||
279 | static int vfio_iommu_group_notifier(struct notifier_block *nb, | |
280 | unsigned long action, void *data); | |
281 | static void vfio_group_get(struct vfio_group *group); | |
282 | ||
283 | /** | |
284 | * Container objects - containers are created when /dev/vfio/vfio is | |
285 | * opened, but their lifecycle extends until the last user is done, so | |
286 | * it's freed via kref. Must support container/group/device being | |
287 | * closed in any order. | |
288 | */ | |
289 | static void vfio_container_get(struct vfio_container *container) | |
290 | { | |
291 | kref_get(&container->kref); | |
292 | } | |
293 | ||
294 | static void vfio_container_release(struct kref *kref) | |
295 | { | |
296 | struct vfio_container *container; | |
297 | container = container_of(kref, struct vfio_container, kref); | |
298 | ||
299 | kfree(container); | |
300 | } | |
301 | ||
302 | static void vfio_container_put(struct vfio_container *container) | |
303 | { | |
304 | kref_put(&container->kref, vfio_container_release); | |
305 | } | |
306 | ||
9df7b25a JL |
307 | static void vfio_group_unlock_and_free(struct vfio_group *group) |
308 | { | |
309 | mutex_unlock(&vfio.group_lock); | |
310 | /* | |
311 | * Unregister outside of lock. A spurious callback is harmless now | |
312 | * that the group is no longer in vfio.group_list. | |
313 | */ | |
314 | iommu_group_unregister_notifier(group->iommu_group, &group->nb); | |
315 | kfree(group); | |
316 | } | |
317 | ||
cba3345c AW |
318 | /** |
319 | * Group objects - create, release, get, put, search | |
320 | */ | |
16ab8a5c | 321 | static struct vfio_group *vfio_create_group(struct iommu_group *iommu_group) |
cba3345c AW |
322 | { |
323 | struct vfio_group *group, *tmp; | |
324 | struct device *dev; | |
325 | int ret, minor; | |
326 | ||
327 | group = kzalloc(sizeof(*group), GFP_KERNEL); | |
328 | if (!group) | |
329 | return ERR_PTR(-ENOMEM); | |
330 | ||
331 | kref_init(&group->kref); | |
332 | INIT_LIST_HEAD(&group->device_list); | |
333 | mutex_init(&group->device_lock); | |
60720a0f AW |
334 | INIT_LIST_HEAD(&group->unbound_list); |
335 | mutex_init(&group->unbound_lock); | |
cba3345c | 336 | atomic_set(&group->container_users, 0); |
6d6768c6 | 337 | atomic_set(&group->opened, 0); |
cba3345c | 338 | group->iommu_group = iommu_group; |
16ab8a5c AW |
339 | #ifdef CONFIG_VFIO_NOIOMMU |
340 | group->noiommu = (iommu_group_get_iommudata(iommu_group) == &noiommu); | |
341 | #endif | |
cba3345c AW |
342 | |
343 | group->nb.notifier_call = vfio_iommu_group_notifier; | |
344 | ||
345 | /* | |
346 | * blocking notifiers acquire a rwsem around registering and hold | |
347 | * it around callback. Therefore, need to register outside of | |
348 | * vfio.group_lock to avoid A-B/B-A contention. Our callback won't | |
349 | * do anything unless it can find the group in vfio.group_list, so | |
350 | * no harm in registering early. | |
351 | */ | |
352 | ret = iommu_group_register_notifier(iommu_group, &group->nb); | |
353 | if (ret) { | |
354 | kfree(group); | |
355 | return ERR_PTR(ret); | |
356 | } | |
357 | ||
358 | mutex_lock(&vfio.group_lock); | |
359 | ||
cba3345c AW |
360 | /* Did we race creating this group? */ |
361 | list_for_each_entry(tmp, &vfio.group_list, vfio_next) { | |
362 | if (tmp->iommu_group == iommu_group) { | |
363 | vfio_group_get(tmp); | |
9df7b25a | 364 | vfio_group_unlock_and_free(group); |
cba3345c AW |
365 | return tmp; |
366 | } | |
367 | } | |
368 | ||
2f51bf4b ZL |
369 | minor = vfio_alloc_group_minor(group); |
370 | if (minor < 0) { | |
371 | vfio_group_unlock_and_free(group); | |
372 | return ERR_PTR(minor); | |
373 | } | |
374 | ||
d1099901 AW |
375 | dev = device_create(vfio.class, NULL, |
376 | MKDEV(MAJOR(vfio.group_devt), minor), | |
03a76b60 AW |
377 | group, "%s%d", group->noiommu ? "noiommu-" : "", |
378 | iommu_group_id(iommu_group)); | |
cba3345c AW |
379 | if (IS_ERR(dev)) { |
380 | vfio_free_group_minor(minor); | |
9df7b25a | 381 | vfio_group_unlock_and_free(group); |
cba3345c AW |
382 | return (struct vfio_group *)dev; /* ERR_PTR */ |
383 | } | |
384 | ||
385 | group->minor = minor; | |
386 | group->dev = dev; | |
387 | ||
388 | list_add(&group->vfio_next, &vfio.group_list); | |
389 | ||
390 | mutex_unlock(&vfio.group_lock); | |
391 | ||
392 | return group; | |
393 | } | |
394 | ||
6d2cd3ce | 395 | /* called with vfio.group_lock held */ |
cba3345c AW |
396 | static void vfio_group_release(struct kref *kref) |
397 | { | |
398 | struct vfio_group *group = container_of(kref, struct vfio_group, kref); | |
60720a0f | 399 | struct vfio_unbound_dev *unbound, *tmp; |
4a68810d | 400 | struct iommu_group *iommu_group = group->iommu_group; |
cba3345c AW |
401 | |
402 | WARN_ON(!list_empty(&group->device_list)); | |
403 | ||
60720a0f AW |
404 | list_for_each_entry_safe(unbound, tmp, |
405 | &group->unbound_list, unbound_next) { | |
406 | list_del(&unbound->unbound_next); | |
407 | kfree(unbound); | |
408 | } | |
409 | ||
d1099901 | 410 | device_destroy(vfio.class, MKDEV(MAJOR(vfio.group_devt), group->minor)); |
cba3345c AW |
411 | list_del(&group->vfio_next); |
412 | vfio_free_group_minor(group->minor); | |
9df7b25a | 413 | vfio_group_unlock_and_free(group); |
4a68810d | 414 | iommu_group_put(iommu_group); |
cba3345c AW |
415 | } |
416 | ||
417 | static void vfio_group_put(struct vfio_group *group) | |
418 | { | |
6d2cd3ce | 419 | kref_put_mutex(&group->kref, vfio_group_release, &vfio.group_lock); |
cba3345c AW |
420 | } |
421 | ||
422 | /* Assume group_lock or group reference is held */ | |
423 | static void vfio_group_get(struct vfio_group *group) | |
424 | { | |
425 | kref_get(&group->kref); | |
426 | } | |
427 | ||
428 | /* | |
429 | * Not really a try as we will sleep for mutex, but we need to make | |
430 | * sure the group pointer is valid under lock and get a reference. | |
431 | */ | |
432 | static struct vfio_group *vfio_group_try_get(struct vfio_group *group) | |
433 | { | |
434 | struct vfio_group *target = group; | |
435 | ||
436 | mutex_lock(&vfio.group_lock); | |
437 | list_for_each_entry(group, &vfio.group_list, vfio_next) { | |
438 | if (group == target) { | |
439 | vfio_group_get(group); | |
440 | mutex_unlock(&vfio.group_lock); | |
441 | return group; | |
442 | } | |
443 | } | |
444 | mutex_unlock(&vfio.group_lock); | |
445 | ||
446 | return NULL; | |
447 | } | |
448 | ||
449 | static | |
450 | struct vfio_group *vfio_group_get_from_iommu(struct iommu_group *iommu_group) | |
451 | { | |
452 | struct vfio_group *group; | |
453 | ||
454 | mutex_lock(&vfio.group_lock); | |
455 | list_for_each_entry(group, &vfio.group_list, vfio_next) { | |
456 | if (group->iommu_group == iommu_group) { | |
457 | vfio_group_get(group); | |
458 | mutex_unlock(&vfio.group_lock); | |
459 | return group; | |
460 | } | |
461 | } | |
462 | mutex_unlock(&vfio.group_lock); | |
463 | ||
464 | return NULL; | |
465 | } | |
466 | ||
467 | static struct vfio_group *vfio_group_get_from_minor(int minor) | |
468 | { | |
469 | struct vfio_group *group; | |
470 | ||
471 | mutex_lock(&vfio.group_lock); | |
472 | group = idr_find(&vfio.group_idr, minor); | |
473 | if (!group) { | |
474 | mutex_unlock(&vfio.group_lock); | |
475 | return NULL; | |
476 | } | |
477 | vfio_group_get(group); | |
478 | mutex_unlock(&vfio.group_lock); | |
479 | ||
480 | return group; | |
481 | } | |
482 | ||
483 | /** | |
484 | * Device objects - create, release, get, put, search | |
485 | */ | |
486 | static | |
487 | struct vfio_device *vfio_group_create_device(struct vfio_group *group, | |
488 | struct device *dev, | |
489 | const struct vfio_device_ops *ops, | |
490 | void *device_data) | |
491 | { | |
492 | struct vfio_device *device; | |
cba3345c AW |
493 | |
494 | device = kzalloc(sizeof(*device), GFP_KERNEL); | |
495 | if (!device) | |
496 | return ERR_PTR(-ENOMEM); | |
497 | ||
498 | kref_init(&device->kref); | |
499 | device->dev = dev; | |
500 | device->group = group; | |
501 | device->ops = ops; | |
502 | device->device_data = device_data; | |
8283b491 | 503 | dev_set_drvdata(dev, device); |
cba3345c AW |
504 | |
505 | /* No need to get group_lock, caller has group reference */ | |
506 | vfio_group_get(group); | |
507 | ||
508 | mutex_lock(&group->device_lock); | |
509 | list_add(&device->group_next, &group->device_list); | |
510 | mutex_unlock(&group->device_lock); | |
511 | ||
512 | return device; | |
513 | } | |
514 | ||
515 | static void vfio_device_release(struct kref *kref) | |
516 | { | |
517 | struct vfio_device *device = container_of(kref, | |
518 | struct vfio_device, kref); | |
519 | struct vfio_group *group = device->group; | |
520 | ||
cba3345c AW |
521 | list_del(&device->group_next); |
522 | mutex_unlock(&group->device_lock); | |
523 | ||
524 | dev_set_drvdata(device->dev, NULL); | |
525 | ||
526 | kfree(device); | |
527 | ||
528 | /* vfio_del_group_dev may be waiting for this device */ | |
529 | wake_up(&vfio.release_q); | |
530 | } | |
531 | ||
532 | /* Device reference always implies a group reference */ | |
44f50716 | 533 | void vfio_device_put(struct vfio_device *device) |
cba3345c | 534 | { |
934ad4c2 | 535 | struct vfio_group *group = device->group; |
90b1253e | 536 | kref_put_mutex(&device->kref, vfio_device_release, &group->device_lock); |
934ad4c2 | 537 | vfio_group_put(group); |
cba3345c | 538 | } |
44f50716 | 539 | EXPORT_SYMBOL_GPL(vfio_device_put); |
cba3345c AW |
540 | |
541 | static void vfio_device_get(struct vfio_device *device) | |
542 | { | |
543 | vfio_group_get(device->group); | |
544 | kref_get(&device->kref); | |
545 | } | |
546 | ||
547 | static struct vfio_device *vfio_group_get_device(struct vfio_group *group, | |
548 | struct device *dev) | |
549 | { | |
550 | struct vfio_device *device; | |
551 | ||
552 | mutex_lock(&group->device_lock); | |
553 | list_for_each_entry(device, &group->device_list, group_next) { | |
554 | if (device->dev == dev) { | |
555 | vfio_device_get(device); | |
556 | mutex_unlock(&group->device_lock); | |
557 | return device; | |
558 | } | |
559 | } | |
560 | mutex_unlock(&group->device_lock); | |
561 | return NULL; | |
562 | } | |
563 | ||
564 | /* | |
5f096b14 AW |
565 | * Some drivers, like pci-stub, are only used to prevent other drivers from |
566 | * claiming a device and are therefore perfectly legitimate for a user owned | |
567 | * group. The pci-stub driver has no dependencies on DMA or the IOVA mapping | |
568 | * of the device, but it does prevent the user from having direct access to | |
569 | * the device, which is useful in some circumstances. | |
570 | * | |
571 | * We also assume that we can include PCI interconnect devices, ie. bridges. | |
572 | * IOMMU grouping on PCI necessitates that if we lack isolation on a bridge | |
573 | * then all of the downstream devices will be part of the same IOMMU group as | |
574 | * the bridge. Thus, if placing the bridge into the user owned IOVA space | |
575 | * breaks anything, it only does so for user owned devices downstream. Note | |
576 | * that error notification via MSI can be affected for platforms that handle | |
577 | * MSI within the same IOVA space as DMA. | |
cba3345c | 578 | */ |
5f096b14 | 579 | static const char * const vfio_driver_whitelist[] = { "pci-stub" }; |
cba3345c | 580 | |
5f096b14 | 581 | static bool vfio_dev_whitelisted(struct device *dev, struct device_driver *drv) |
cba3345c AW |
582 | { |
583 | int i; | |
584 | ||
5f096b14 AW |
585 | if (dev_is_pci(dev)) { |
586 | struct pci_dev *pdev = to_pci_dev(dev); | |
587 | ||
588 | if (pdev->hdr_type != PCI_HEADER_TYPE_NORMAL) | |
589 | return true; | |
590 | } | |
591 | ||
cba3345c AW |
592 | for (i = 0; i < ARRAY_SIZE(vfio_driver_whitelist); i++) { |
593 | if (!strcmp(drv->name, vfio_driver_whitelist[i])) | |
594 | return true; | |
595 | } | |
596 | ||
597 | return false; | |
598 | } | |
599 | ||
600 | /* | |
60720a0f AW |
601 | * A vfio group is viable for use by userspace if all devices are in |
602 | * one of the following states: | |
603 | * - driver-less | |
604 | * - bound to a vfio driver | |
605 | * - bound to a whitelisted driver | |
5f096b14 | 606 | * - a PCI interconnect device |
60720a0f AW |
607 | * |
608 | * We use two methods to determine whether a device is bound to a vfio | |
609 | * driver. The first is to test whether the device exists in the vfio | |
610 | * group. The second is to test if the device exists on the group | |
611 | * unbound_list, indicating it's in the middle of transitioning from | |
612 | * a vfio driver to driver-less. | |
cba3345c AW |
613 | */ |
614 | static int vfio_dev_viable(struct device *dev, void *data) | |
615 | { | |
616 | struct vfio_group *group = data; | |
617 | struct vfio_device *device; | |
de2b3eea | 618 | struct device_driver *drv = ACCESS_ONCE(dev->driver); |
60720a0f AW |
619 | struct vfio_unbound_dev *unbound; |
620 | int ret = -EINVAL; | |
621 | ||
622 | mutex_lock(&group->unbound_lock); | |
623 | list_for_each_entry(unbound, &group->unbound_list, unbound_next) { | |
624 | if (dev == unbound->dev) { | |
625 | ret = 0; | |
626 | break; | |
627 | } | |
628 | } | |
629 | mutex_unlock(&group->unbound_lock); | |
cba3345c | 630 | |
5f096b14 | 631 | if (!ret || !drv || vfio_dev_whitelisted(dev, drv)) |
cba3345c AW |
632 | return 0; |
633 | ||
634 | device = vfio_group_get_device(group, dev); | |
635 | if (device) { | |
636 | vfio_device_put(device); | |
637 | return 0; | |
638 | } | |
639 | ||
60720a0f | 640 | return ret; |
cba3345c AW |
641 | } |
642 | ||
643 | /** | |
644 | * Async device support | |
645 | */ | |
646 | static int vfio_group_nb_add_dev(struct vfio_group *group, struct device *dev) | |
647 | { | |
648 | struct vfio_device *device; | |
649 | ||
650 | /* Do we already know about it? We shouldn't */ | |
651 | device = vfio_group_get_device(group, dev); | |
652 | if (WARN_ON_ONCE(device)) { | |
653 | vfio_device_put(device); | |
654 | return 0; | |
655 | } | |
656 | ||
657 | /* Nothing to do for idle groups */ | |
658 | if (!atomic_read(&group->container_users)) | |
659 | return 0; | |
660 | ||
661 | /* TODO Prevent device auto probing */ | |
049af106 | 662 | WARN(1, "Device %s added to live group %d!\n", dev_name(dev), |
cba3345c AW |
663 | iommu_group_id(group->iommu_group)); |
664 | ||
665 | return 0; | |
666 | } | |
667 | ||
cba3345c AW |
668 | static int vfio_group_nb_verify(struct vfio_group *group, struct device *dev) |
669 | { | |
670 | /* We don't care what happens when the group isn't in use */ | |
671 | if (!atomic_read(&group->container_users)) | |
672 | return 0; | |
673 | ||
674 | return vfio_dev_viable(dev, group); | |
675 | } | |
676 | ||
677 | static int vfio_iommu_group_notifier(struct notifier_block *nb, | |
678 | unsigned long action, void *data) | |
679 | { | |
680 | struct vfio_group *group = container_of(nb, struct vfio_group, nb); | |
681 | struct device *dev = data; | |
60720a0f | 682 | struct vfio_unbound_dev *unbound; |
cba3345c AW |
683 | |
684 | /* | |
c6401930 AW |
685 | * Need to go through a group_lock lookup to get a reference or we |
686 | * risk racing a group being removed. Ignore spurious notifies. | |
cba3345c AW |
687 | */ |
688 | group = vfio_group_try_get(group); | |
c6401930 | 689 | if (!group) |
cba3345c AW |
690 | return NOTIFY_OK; |
691 | ||
692 | switch (action) { | |
693 | case IOMMU_GROUP_NOTIFY_ADD_DEVICE: | |
694 | vfio_group_nb_add_dev(group, dev); | |
695 | break; | |
696 | case IOMMU_GROUP_NOTIFY_DEL_DEVICE: | |
de9c7602 AW |
697 | /* |
698 | * Nothing to do here. If the device is in use, then the | |
699 | * vfio sub-driver should block the remove callback until | |
700 | * it is unused. If the device is unused or attached to a | |
701 | * stub driver, then it should be released and we don't | |
702 | * care that it will be going away. | |
703 | */ | |
cba3345c AW |
704 | break; |
705 | case IOMMU_GROUP_NOTIFY_BIND_DRIVER: | |
706 | pr_debug("%s: Device %s, group %d binding to driver\n", | |
707 | __func__, dev_name(dev), | |
708 | iommu_group_id(group->iommu_group)); | |
709 | break; | |
710 | case IOMMU_GROUP_NOTIFY_BOUND_DRIVER: | |
711 | pr_debug("%s: Device %s, group %d bound to driver %s\n", | |
712 | __func__, dev_name(dev), | |
713 | iommu_group_id(group->iommu_group), dev->driver->name); | |
714 | BUG_ON(vfio_group_nb_verify(group, dev)); | |
715 | break; | |
716 | case IOMMU_GROUP_NOTIFY_UNBIND_DRIVER: | |
717 | pr_debug("%s: Device %s, group %d unbinding from driver %s\n", | |
718 | __func__, dev_name(dev), | |
719 | iommu_group_id(group->iommu_group), dev->driver->name); | |
720 | break; | |
721 | case IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER: | |
722 | pr_debug("%s: Device %s, group %d unbound from driver\n", | |
723 | __func__, dev_name(dev), | |
724 | iommu_group_id(group->iommu_group)); | |
725 | /* | |
726 | * XXX An unbound device in a live group is ok, but we'd | |
727 | * really like to avoid the above BUG_ON by preventing other | |
728 | * drivers from binding to it. Once that occurs, we have to | |
729 | * stop the system to maintain isolation. At a minimum, we'd | |
730 | * want a toggle to disable driver auto probe for this device. | |
731 | */ | |
60720a0f AW |
732 | |
733 | mutex_lock(&group->unbound_lock); | |
734 | list_for_each_entry(unbound, | |
735 | &group->unbound_list, unbound_next) { | |
736 | if (dev == unbound->dev) { | |
737 | list_del(&unbound->unbound_next); | |
738 | kfree(unbound); | |
739 | break; | |
740 | } | |
741 | } | |
742 | mutex_unlock(&group->unbound_lock); | |
cba3345c AW |
743 | break; |
744 | } | |
745 | ||
746 | vfio_group_put(group); | |
747 | return NOTIFY_OK; | |
748 | } | |
749 | ||
750 | /** | |
751 | * VFIO driver API | |
752 | */ | |
753 | int vfio_add_group_dev(struct device *dev, | |
754 | const struct vfio_device_ops *ops, void *device_data) | |
755 | { | |
756 | struct iommu_group *iommu_group; | |
757 | struct vfio_group *group; | |
758 | struct vfio_device *device; | |
759 | ||
760 | iommu_group = iommu_group_get(dev); | |
761 | if (!iommu_group) | |
762 | return -EINVAL; | |
763 | ||
764 | group = vfio_group_get_from_iommu(iommu_group); | |
765 | if (!group) { | |
16ab8a5c | 766 | group = vfio_create_group(iommu_group); |
cba3345c AW |
767 | if (IS_ERR(group)) { |
768 | iommu_group_put(iommu_group); | |
769 | return PTR_ERR(group); | |
770 | } | |
4a68810d AW |
771 | } else { |
772 | /* | |
773 | * A found vfio_group already holds a reference to the | |
774 | * iommu_group. A created vfio_group keeps the reference. | |
775 | */ | |
776 | iommu_group_put(iommu_group); | |
cba3345c AW |
777 | } |
778 | ||
779 | device = vfio_group_get_device(group, dev); | |
780 | if (device) { | |
781 | WARN(1, "Device %s already exists on group %d\n", | |
782 | dev_name(dev), iommu_group_id(iommu_group)); | |
783 | vfio_device_put(device); | |
784 | vfio_group_put(group); | |
cba3345c AW |
785 | return -EBUSY; |
786 | } | |
787 | ||
788 | device = vfio_group_create_device(group, dev, ops, device_data); | |
789 | if (IS_ERR(device)) { | |
790 | vfio_group_put(group); | |
cba3345c AW |
791 | return PTR_ERR(device); |
792 | } | |
793 | ||
794 | /* | |
4a68810d AW |
795 | * Drop all but the vfio_device reference. The vfio_device holds |
796 | * a reference to the vfio_group, which holds a reference to the | |
797 | * iommu_group. | |
cba3345c AW |
798 | */ |
799 | vfio_group_put(group); | |
800 | ||
801 | return 0; | |
802 | } | |
803 | EXPORT_SYMBOL_GPL(vfio_add_group_dev); | |
804 | ||
44f50716 | 805 | /** |
20f30017 AW |
806 | * Get a reference to the vfio_device for a device. Even if the |
807 | * caller thinks they own the device, they could be racing with a | |
808 | * release call path, so we can't trust drvdata for the shortcut. | |
809 | * Go the long way around, from the iommu_group to the vfio_group | |
810 | * to the vfio_device. | |
44f50716 VMP |
811 | */ |
812 | struct vfio_device *vfio_device_get_from_dev(struct device *dev) | |
813 | { | |
20f30017 AW |
814 | struct iommu_group *iommu_group; |
815 | struct vfio_group *group; | |
816 | struct vfio_device *device; | |
817 | ||
818 | iommu_group = iommu_group_get(dev); | |
819 | if (!iommu_group) | |
820 | return NULL; | |
44f50716 | 821 | |
20f30017 AW |
822 | group = vfio_group_get_from_iommu(iommu_group); |
823 | iommu_group_put(iommu_group); | |
824 | if (!group) | |
825 | return NULL; | |
826 | ||
827 | device = vfio_group_get_device(group, dev); | |
828 | vfio_group_put(group); | |
44f50716 VMP |
829 | |
830 | return device; | |
831 | } | |
832 | EXPORT_SYMBOL_GPL(vfio_device_get_from_dev); | |
833 | ||
4bc94d5d AW |
834 | static struct vfio_device *vfio_device_get_from_name(struct vfio_group *group, |
835 | char *buf) | |
836 | { | |
e324fc82 | 837 | struct vfio_device *it, *device = NULL; |
4bc94d5d AW |
838 | |
839 | mutex_lock(&group->device_lock); | |
e324fc82 JR |
840 | list_for_each_entry(it, &group->device_list, group_next) { |
841 | if (!strcmp(dev_name(it->dev), buf)) { | |
842 | device = it; | |
4bc94d5d AW |
843 | vfio_device_get(device); |
844 | break; | |
845 | } | |
846 | } | |
847 | mutex_unlock(&group->device_lock); | |
848 | ||
849 | return device; | |
850 | } | |
851 | ||
44f50716 VMP |
852 | /* |
853 | * Caller must hold a reference to the vfio_device | |
854 | */ | |
855 | void *vfio_device_data(struct vfio_device *device) | |
856 | { | |
857 | return device->device_data; | |
858 | } | |
859 | EXPORT_SYMBOL_GPL(vfio_device_data); | |
860 | ||
e014e944 AW |
861 | /* Given a referenced group, check if it contains the device */ |
862 | static bool vfio_dev_present(struct vfio_group *group, struct device *dev) | |
cba3345c | 863 | { |
cba3345c AW |
864 | struct vfio_device *device; |
865 | ||
cba3345c | 866 | device = vfio_group_get_device(group, dev); |
e014e944 | 867 | if (!device) |
cba3345c | 868 | return false; |
cba3345c AW |
869 | |
870 | vfio_device_put(device); | |
cba3345c AW |
871 | return true; |
872 | } | |
873 | ||
874 | /* | |
875 | * Decrement the device reference count and wait for the device to be | |
876 | * removed. Open file descriptors for the device... */ | |
877 | void *vfio_del_group_dev(struct device *dev) | |
878 | { | |
879 | struct vfio_device *device = dev_get_drvdata(dev); | |
880 | struct vfio_group *group = device->group; | |
cba3345c | 881 | void *device_data = device->device_data; |
60720a0f | 882 | struct vfio_unbound_dev *unbound; |
13060b64 | 883 | unsigned int i = 0; |
db7d4d7f AW |
884 | long ret; |
885 | bool interrupted = false; | |
cba3345c | 886 | |
e014e944 AW |
887 | /* |
888 | * The group exists so long as we have a device reference. Get | |
889 | * a group reference and use it to scan for the device going away. | |
890 | */ | |
891 | vfio_group_get(group); | |
892 | ||
60720a0f AW |
893 | /* |
894 | * When the device is removed from the group, the group suddenly | |
895 | * becomes non-viable; the device has a driver (until the unbind | |
896 | * completes), but it's not present in the group. This is bad news | |
897 | * for any external users that need to re-acquire a group reference | |
898 | * in order to match and release their existing reference. To | |
899 | * solve this, we track such devices on the unbound_list to bridge | |
900 | * the gap until they're fully unbound. | |
901 | */ | |
902 | unbound = kzalloc(sizeof(*unbound), GFP_KERNEL); | |
903 | if (unbound) { | |
904 | unbound->dev = dev; | |
905 | mutex_lock(&group->unbound_lock); | |
906 | list_add(&unbound->unbound_next, &group->unbound_list); | |
907 | mutex_unlock(&group->unbound_lock); | |
908 | } | |
909 | WARN_ON(!unbound); | |
910 | ||
cba3345c AW |
911 | vfio_device_put(device); |
912 | ||
13060b64 AW |
913 | /* |
914 | * If the device is still present in the group after the above | |
915 | * 'put', then it is in use and we need to request it from the | |
916 | * bus driver. The driver may in turn need to request the | |
917 | * device from the user. We send the request on an arbitrary | |
918 | * interval with counter to allow the driver to take escalating | |
919 | * measures to release the device if it has the ability to do so. | |
920 | */ | |
921 | do { | |
922 | device = vfio_group_get_device(group, dev); | |
923 | if (!device) | |
924 | break; | |
925 | ||
926 | if (device->ops->request) | |
927 | device->ops->request(device_data, i++); | |
928 | ||
929 | vfio_device_put(device); | |
930 | ||
db7d4d7f AW |
931 | if (interrupted) { |
932 | ret = wait_event_timeout(vfio.release_q, | |
933 | !vfio_dev_present(group, dev), HZ * 10); | |
934 | } else { | |
935 | ret = wait_event_interruptible_timeout(vfio.release_q, | |
936 | !vfio_dev_present(group, dev), HZ * 10); | |
937 | if (ret == -ERESTARTSYS) { | |
938 | interrupted = true; | |
939 | dev_warn(dev, | |
940 | "Device is currently in use, task" | |
941 | " \"%s\" (%d) " | |
942 | "blocked until device is released", | |
943 | current->comm, task_pid_nr(current)); | |
944 | } | |
945 | } | |
946 | } while (ret <= 0); | |
e014e944 AW |
947 | |
948 | vfio_group_put(group); | |
cba3345c | 949 | |
cba3345c AW |
950 | return device_data; |
951 | } | |
952 | EXPORT_SYMBOL_GPL(vfio_del_group_dev); | |
953 | ||
954 | /** | |
955 | * VFIO base fd, /dev/vfio/vfio | |
956 | */ | |
957 | static long vfio_ioctl_check_extension(struct vfio_container *container, | |
958 | unsigned long arg) | |
959 | { | |
0b43c082 | 960 | struct vfio_iommu_driver *driver; |
cba3345c AW |
961 | long ret = 0; |
962 | ||
0b43c082 AW |
963 | down_read(&container->group_lock); |
964 | ||
965 | driver = container->iommu_driver; | |
966 | ||
cba3345c AW |
967 | switch (arg) { |
968 | /* No base extensions yet */ | |
969 | default: | |
970 | /* | |
971 | * If no driver is set, poll all registered drivers for | |
972 | * extensions and return the first positive result. If | |
973 | * a driver is already set, further queries will be passed | |
974 | * only to that driver. | |
975 | */ | |
976 | if (!driver) { | |
977 | mutex_lock(&vfio.iommu_drivers_lock); | |
ae5515d6 AW |
978 | list_for_each_entry(driver, &vfio.iommu_drivers_list, |
979 | vfio_next) { | |
03a76b60 AW |
980 | |
981 | #ifdef CONFIG_VFIO_NOIOMMU | |
982 | if (!list_empty(&container->group_list) && | |
983 | (container->noiommu != | |
984 | (driver->ops == &vfio_noiommu_ops))) | |
985 | continue; | |
986 | #endif | |
987 | ||
cba3345c AW |
988 | if (!try_module_get(driver->ops->owner)) |
989 | continue; | |
990 | ||
991 | ret = driver->ops->ioctl(NULL, | |
992 | VFIO_CHECK_EXTENSION, | |
993 | arg); | |
994 | module_put(driver->ops->owner); | |
995 | if (ret > 0) | |
996 | break; | |
997 | } | |
998 | mutex_unlock(&vfio.iommu_drivers_lock); | |
999 | } else | |
1000 | ret = driver->ops->ioctl(container->iommu_data, | |
1001 | VFIO_CHECK_EXTENSION, arg); | |
1002 | } | |
1003 | ||
0b43c082 AW |
1004 | up_read(&container->group_lock); |
1005 | ||
cba3345c AW |
1006 | return ret; |
1007 | } | |
1008 | ||
9587f44a | 1009 | /* hold write lock on container->group_lock */ |
cba3345c AW |
1010 | static int __vfio_container_attach_groups(struct vfio_container *container, |
1011 | struct vfio_iommu_driver *driver, | |
1012 | void *data) | |
1013 | { | |
1014 | struct vfio_group *group; | |
1015 | int ret = -ENODEV; | |
1016 | ||
1017 | list_for_each_entry(group, &container->group_list, container_next) { | |
1018 | ret = driver->ops->attach_group(data, group->iommu_group); | |
1019 | if (ret) | |
1020 | goto unwind; | |
1021 | } | |
1022 | ||
1023 | return ret; | |
1024 | ||
1025 | unwind: | |
1026 | list_for_each_entry_continue_reverse(group, &container->group_list, | |
1027 | container_next) { | |
1028 | driver->ops->detach_group(data, group->iommu_group); | |
1029 | } | |
1030 | ||
1031 | return ret; | |
1032 | } | |
1033 | ||
1034 | static long vfio_ioctl_set_iommu(struct vfio_container *container, | |
1035 | unsigned long arg) | |
1036 | { | |
1037 | struct vfio_iommu_driver *driver; | |
1038 | long ret = -ENODEV; | |
1039 | ||
9587f44a | 1040 | down_write(&container->group_lock); |
cba3345c AW |
1041 | |
1042 | /* | |
1043 | * The container is designed to be an unprivileged interface while | |
1044 | * the group can be assigned to specific users. Therefore, only by | |
1045 | * adding a group to a container does the user get the privilege of | |
1046 | * enabling the iommu, which may allocate finite resources. There | |
1047 | * is no unset_iommu, but by removing all the groups from a container, | |
1048 | * the container is deprivileged and returns to an unset state. | |
1049 | */ | |
1050 | if (list_empty(&container->group_list) || container->iommu_driver) { | |
9587f44a | 1051 | up_write(&container->group_lock); |
cba3345c AW |
1052 | return -EINVAL; |
1053 | } | |
1054 | ||
1055 | mutex_lock(&vfio.iommu_drivers_lock); | |
ae5515d6 | 1056 | list_for_each_entry(driver, &vfio.iommu_drivers_list, vfio_next) { |
cba3345c AW |
1057 | void *data; |
1058 | ||
03a76b60 AW |
1059 | #ifdef CONFIG_VFIO_NOIOMMU |
1060 | /* | |
1061 | * Only noiommu containers can use vfio-noiommu and noiommu | |
1062 | * containers can only use vfio-noiommu. | |
1063 | */ | |
1064 | if (container->noiommu != (driver->ops == &vfio_noiommu_ops)) | |
1065 | continue; | |
1066 | #endif | |
1067 | ||
cba3345c AW |
1068 | if (!try_module_get(driver->ops->owner)) |
1069 | continue; | |
1070 | ||
1071 | /* | |
1072 | * The arg magic for SET_IOMMU is the same as CHECK_EXTENSION, | |
1073 | * so test which iommu driver reported support for this | |
1074 | * extension and call open on them. We also pass them the | |
1075 | * magic, allowing a single driver to support multiple | |
1076 | * interfaces if they'd like. | |
1077 | */ | |
1078 | if (driver->ops->ioctl(NULL, VFIO_CHECK_EXTENSION, arg) <= 0) { | |
1079 | module_put(driver->ops->owner); | |
1080 | continue; | |
1081 | } | |
1082 | ||
1083 | /* module reference holds the driver we're working on */ | |
1084 | mutex_unlock(&vfio.iommu_drivers_lock); | |
1085 | ||
1086 | data = driver->ops->open(arg); | |
1087 | if (IS_ERR(data)) { | |
1088 | ret = PTR_ERR(data); | |
1089 | module_put(driver->ops->owner); | |
1090 | goto skip_drivers_unlock; | |
1091 | } | |
1092 | ||
1093 | ret = __vfio_container_attach_groups(container, driver, data); | |
1094 | if (!ret) { | |
1095 | container->iommu_driver = driver; | |
1096 | container->iommu_data = data; | |
1097 | } else { | |
1098 | driver->ops->release(data); | |
1099 | module_put(driver->ops->owner); | |
1100 | } | |
1101 | ||
1102 | goto skip_drivers_unlock; | |
1103 | } | |
1104 | ||
1105 | mutex_unlock(&vfio.iommu_drivers_lock); | |
1106 | skip_drivers_unlock: | |
9587f44a | 1107 | up_write(&container->group_lock); |
cba3345c AW |
1108 | |
1109 | return ret; | |
1110 | } | |
1111 | ||
1112 | static long vfio_fops_unl_ioctl(struct file *filep, | |
1113 | unsigned int cmd, unsigned long arg) | |
1114 | { | |
1115 | struct vfio_container *container = filep->private_data; | |
1116 | struct vfio_iommu_driver *driver; | |
1117 | void *data; | |
1118 | long ret = -EINVAL; | |
1119 | ||
1120 | if (!container) | |
1121 | return ret; | |
1122 | ||
cba3345c AW |
1123 | switch (cmd) { |
1124 | case VFIO_GET_API_VERSION: | |
1125 | ret = VFIO_API_VERSION; | |
1126 | break; | |
1127 | case VFIO_CHECK_EXTENSION: | |
1128 | ret = vfio_ioctl_check_extension(container, arg); | |
1129 | break; | |
1130 | case VFIO_SET_IOMMU: | |
1131 | ret = vfio_ioctl_set_iommu(container, arg); | |
1132 | break; | |
1133 | default: | |
0b43c082 AW |
1134 | down_read(&container->group_lock); |
1135 | ||
1136 | driver = container->iommu_driver; | |
1137 | data = container->iommu_data; | |
1138 | ||
cba3345c AW |
1139 | if (driver) /* passthrough all unrecognized ioctls */ |
1140 | ret = driver->ops->ioctl(data, cmd, arg); | |
0b43c082 AW |
1141 | |
1142 | up_read(&container->group_lock); | |
cba3345c AW |
1143 | } |
1144 | ||
1145 | return ret; | |
1146 | } | |
1147 | ||
1148 | #ifdef CONFIG_COMPAT | |
1149 | static long vfio_fops_compat_ioctl(struct file *filep, | |
1150 | unsigned int cmd, unsigned long arg) | |
1151 | { | |
1152 | arg = (unsigned long)compat_ptr(arg); | |
1153 | return vfio_fops_unl_ioctl(filep, cmd, arg); | |
1154 | } | |
1155 | #endif /* CONFIG_COMPAT */ | |
1156 | ||
1157 | static int vfio_fops_open(struct inode *inode, struct file *filep) | |
1158 | { | |
1159 | struct vfio_container *container; | |
1160 | ||
1161 | container = kzalloc(sizeof(*container), GFP_KERNEL); | |
1162 | if (!container) | |
1163 | return -ENOMEM; | |
1164 | ||
1165 | INIT_LIST_HEAD(&container->group_list); | |
9587f44a | 1166 | init_rwsem(&container->group_lock); |
cba3345c AW |
1167 | kref_init(&container->kref); |
1168 | ||
1169 | filep->private_data = container; | |
1170 | ||
1171 | return 0; | |
1172 | } | |
1173 | ||
1174 | static int vfio_fops_release(struct inode *inode, struct file *filep) | |
1175 | { | |
1176 | struct vfio_container *container = filep->private_data; | |
1177 | ||
1178 | filep->private_data = NULL; | |
1179 | ||
1180 | vfio_container_put(container); | |
1181 | ||
1182 | return 0; | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * Once an iommu driver is set, we optionally pass read/write/mmap | |
1187 | * on to the driver, allowing management interfaces beyond ioctl. | |
1188 | */ | |
1189 | static ssize_t vfio_fops_read(struct file *filep, char __user *buf, | |
1190 | size_t count, loff_t *ppos) | |
1191 | { | |
1192 | struct vfio_container *container = filep->private_data; | |
0b43c082 AW |
1193 | struct vfio_iommu_driver *driver; |
1194 | ssize_t ret = -EINVAL; | |
cba3345c | 1195 | |
0b43c082 AW |
1196 | down_read(&container->group_lock); |
1197 | ||
1198 | driver = container->iommu_driver; | |
1199 | if (likely(driver && driver->ops->read)) | |
1200 | ret = driver->ops->read(container->iommu_data, | |
1201 | buf, count, ppos); | |
cba3345c | 1202 | |
0b43c082 AW |
1203 | up_read(&container->group_lock); |
1204 | ||
1205 | return ret; | |
cba3345c AW |
1206 | } |
1207 | ||
1208 | static ssize_t vfio_fops_write(struct file *filep, const char __user *buf, | |
1209 | size_t count, loff_t *ppos) | |
1210 | { | |
1211 | struct vfio_container *container = filep->private_data; | |
0b43c082 AW |
1212 | struct vfio_iommu_driver *driver; |
1213 | ssize_t ret = -EINVAL; | |
cba3345c | 1214 | |
0b43c082 AW |
1215 | down_read(&container->group_lock); |
1216 | ||
1217 | driver = container->iommu_driver; | |
1218 | if (likely(driver && driver->ops->write)) | |
1219 | ret = driver->ops->write(container->iommu_data, | |
1220 | buf, count, ppos); | |
1221 | ||
1222 | up_read(&container->group_lock); | |
cba3345c | 1223 | |
0b43c082 | 1224 | return ret; |
cba3345c AW |
1225 | } |
1226 | ||
1227 | static int vfio_fops_mmap(struct file *filep, struct vm_area_struct *vma) | |
1228 | { | |
1229 | struct vfio_container *container = filep->private_data; | |
0b43c082 AW |
1230 | struct vfio_iommu_driver *driver; |
1231 | int ret = -EINVAL; | |
cba3345c | 1232 | |
0b43c082 | 1233 | down_read(&container->group_lock); |
cba3345c | 1234 | |
0b43c082 AW |
1235 | driver = container->iommu_driver; |
1236 | if (likely(driver && driver->ops->mmap)) | |
1237 | ret = driver->ops->mmap(container->iommu_data, vma); | |
1238 | ||
1239 | up_read(&container->group_lock); | |
1240 | ||
1241 | return ret; | |
cba3345c AW |
1242 | } |
1243 | ||
1244 | static const struct file_operations vfio_fops = { | |
1245 | .owner = THIS_MODULE, | |
1246 | .open = vfio_fops_open, | |
1247 | .release = vfio_fops_release, | |
1248 | .read = vfio_fops_read, | |
1249 | .write = vfio_fops_write, | |
1250 | .unlocked_ioctl = vfio_fops_unl_ioctl, | |
1251 | #ifdef CONFIG_COMPAT | |
1252 | .compat_ioctl = vfio_fops_compat_ioctl, | |
1253 | #endif | |
1254 | .mmap = vfio_fops_mmap, | |
1255 | }; | |
1256 | ||
1257 | /** | |
1258 | * VFIO Group fd, /dev/vfio/$GROUP | |
1259 | */ | |
1260 | static void __vfio_group_unset_container(struct vfio_group *group) | |
1261 | { | |
1262 | struct vfio_container *container = group->container; | |
1263 | struct vfio_iommu_driver *driver; | |
1264 | ||
9587f44a | 1265 | down_write(&container->group_lock); |
cba3345c AW |
1266 | |
1267 | driver = container->iommu_driver; | |
1268 | if (driver) | |
1269 | driver->ops->detach_group(container->iommu_data, | |
1270 | group->iommu_group); | |
1271 | ||
1272 | group->container = NULL; | |
1273 | list_del(&group->container_next); | |
1274 | ||
1275 | /* Detaching the last group deprivileges a container, remove iommu */ | |
1276 | if (driver && list_empty(&container->group_list)) { | |
1277 | driver->ops->release(container->iommu_data); | |
1278 | module_put(driver->ops->owner); | |
1279 | container->iommu_driver = NULL; | |
1280 | container->iommu_data = NULL; | |
1281 | } | |
1282 | ||
9587f44a | 1283 | up_write(&container->group_lock); |
cba3345c AW |
1284 | |
1285 | vfio_container_put(container); | |
1286 | } | |
1287 | ||
1288 | /* | |
1289 | * VFIO_GROUP_UNSET_CONTAINER should fail if there are other users or | |
1290 | * if there was no container to unset. Since the ioctl is called on | |
1291 | * the group, we know that still exists, therefore the only valid | |
1292 | * transition here is 1->0. | |
1293 | */ | |
1294 | static int vfio_group_unset_container(struct vfio_group *group) | |
1295 | { | |
1296 | int users = atomic_cmpxchg(&group->container_users, 1, 0); | |
1297 | ||
1298 | if (!users) | |
1299 | return -EINVAL; | |
1300 | if (users != 1) | |
1301 | return -EBUSY; | |
1302 | ||
1303 | __vfio_group_unset_container(group); | |
1304 | ||
1305 | return 0; | |
1306 | } | |
1307 | ||
1308 | /* | |
1309 | * When removing container users, anything that removes the last user | |
1310 | * implicitly removes the group from the container. That is, if the | |
1311 | * group file descriptor is closed, as well as any device file descriptors, | |
1312 | * the group is free. | |
1313 | */ | |
1314 | static void vfio_group_try_dissolve_container(struct vfio_group *group) | |
1315 | { | |
1316 | if (0 == atomic_dec_if_positive(&group->container_users)) | |
1317 | __vfio_group_unset_container(group); | |
1318 | } | |
1319 | ||
1320 | static int vfio_group_set_container(struct vfio_group *group, int container_fd) | |
1321 | { | |
2903ff01 | 1322 | struct fd f; |
cba3345c AW |
1323 | struct vfio_container *container; |
1324 | struct vfio_iommu_driver *driver; | |
2903ff01 | 1325 | int ret = 0; |
cba3345c AW |
1326 | |
1327 | if (atomic_read(&group->container_users)) | |
1328 | return -EINVAL; | |
1329 | ||
03a76b60 AW |
1330 | if (group->noiommu && !capable(CAP_SYS_RAWIO)) |
1331 | return -EPERM; | |
1332 | ||
2903ff01 AV |
1333 | f = fdget(container_fd); |
1334 | if (!f.file) | |
cba3345c AW |
1335 | return -EBADF; |
1336 | ||
1337 | /* Sanity check, is this really our fd? */ | |
2903ff01 AV |
1338 | if (f.file->f_op != &vfio_fops) { |
1339 | fdput(f); | |
cba3345c AW |
1340 | return -EINVAL; |
1341 | } | |
1342 | ||
2903ff01 | 1343 | container = f.file->private_data; |
cba3345c AW |
1344 | WARN_ON(!container); /* fget ensures we don't race vfio_release */ |
1345 | ||
9587f44a | 1346 | down_write(&container->group_lock); |
cba3345c | 1347 | |
03a76b60 AW |
1348 | /* Real groups and fake groups cannot mix */ |
1349 | if (!list_empty(&container->group_list) && | |
1350 | container->noiommu != group->noiommu) { | |
1351 | ret = -EPERM; | |
1352 | goto unlock_out; | |
1353 | } | |
1354 | ||
cba3345c AW |
1355 | driver = container->iommu_driver; |
1356 | if (driver) { | |
1357 | ret = driver->ops->attach_group(container->iommu_data, | |
1358 | group->iommu_group); | |
1359 | if (ret) | |
1360 | goto unlock_out; | |
1361 | } | |
1362 | ||
1363 | group->container = container; | |
03a76b60 | 1364 | container->noiommu = group->noiommu; |
cba3345c AW |
1365 | list_add(&group->container_next, &container->group_list); |
1366 | ||
1367 | /* Get a reference on the container and mark a user within the group */ | |
1368 | vfio_container_get(container); | |
1369 | atomic_inc(&group->container_users); | |
1370 | ||
1371 | unlock_out: | |
9587f44a | 1372 | up_write(&container->group_lock); |
2903ff01 | 1373 | fdput(f); |
cba3345c AW |
1374 | return ret; |
1375 | } | |
1376 | ||
1377 | static bool vfio_group_viable(struct vfio_group *group) | |
1378 | { | |
1379 | return (iommu_group_for_each_dev(group->iommu_group, | |
1380 | group, vfio_dev_viable) == 0); | |
1381 | } | |
1382 | ||
1383 | static const struct file_operations vfio_device_fops; | |
1384 | ||
1385 | static int vfio_group_get_device_fd(struct vfio_group *group, char *buf) | |
1386 | { | |
1387 | struct vfio_device *device; | |
1388 | struct file *filep; | |
4bc94d5d | 1389 | int ret; |
cba3345c AW |
1390 | |
1391 | if (0 == atomic_read(&group->container_users) || | |
1392 | !group->container->iommu_driver || !vfio_group_viable(group)) | |
1393 | return -EINVAL; | |
1394 | ||
03a76b60 AW |
1395 | if (group->noiommu && !capable(CAP_SYS_RAWIO)) |
1396 | return -EPERM; | |
1397 | ||
4bc94d5d AW |
1398 | device = vfio_device_get_from_name(group, buf); |
1399 | if (!device) | |
1400 | return -ENODEV; | |
cba3345c | 1401 | |
4bc94d5d AW |
1402 | ret = device->ops->open(device->device_data); |
1403 | if (ret) { | |
1404 | vfio_device_put(device); | |
1405 | return ret; | |
1406 | } | |
cba3345c | 1407 | |
4bc94d5d AW |
1408 | /* |
1409 | * We can't use anon_inode_getfd() because we need to modify | |
1410 | * the f_mode flags directly to allow more than just ioctls | |
1411 | */ | |
1412 | ret = get_unused_fd_flags(O_CLOEXEC); | |
1413 | if (ret < 0) { | |
1414 | device->ops->release(device->device_data); | |
1415 | vfio_device_put(device); | |
1416 | return ret; | |
1417 | } | |
cba3345c | 1418 | |
4bc94d5d AW |
1419 | filep = anon_inode_getfile("[vfio-device]", &vfio_device_fops, |
1420 | device, O_RDWR); | |
1421 | if (IS_ERR(filep)) { | |
1422 | put_unused_fd(ret); | |
1423 | ret = PTR_ERR(filep); | |
1424 | device->ops->release(device->device_data); | |
1425 | vfio_device_put(device); | |
1426 | return ret; | |
1427 | } | |
1428 | ||
1429 | /* | |
1430 | * TODO: add an anon_inode interface to do this. | |
1431 | * Appears to be missing by lack of need rather than | |
1432 | * explicitly prevented. Now there's need. | |
1433 | */ | |
1434 | filep->f_mode |= (FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE); | |
cba3345c | 1435 | |
4bc94d5d | 1436 | atomic_inc(&group->container_users); |
31605deb | 1437 | |
4bc94d5d | 1438 | fd_install(ret, filep); |
cba3345c | 1439 | |
03a76b60 AW |
1440 | if (group->noiommu) |
1441 | dev_warn(device->dev, "vfio-noiommu device opened by user " | |
1442 | "(%s:%d)\n", current->comm, task_pid_nr(current)); | |
1443 | ||
cba3345c AW |
1444 | return ret; |
1445 | } | |
1446 | ||
1447 | static long vfio_group_fops_unl_ioctl(struct file *filep, | |
1448 | unsigned int cmd, unsigned long arg) | |
1449 | { | |
1450 | struct vfio_group *group = filep->private_data; | |
1451 | long ret = -ENOTTY; | |
1452 | ||
1453 | switch (cmd) { | |
1454 | case VFIO_GROUP_GET_STATUS: | |
1455 | { | |
1456 | struct vfio_group_status status; | |
1457 | unsigned long minsz; | |
1458 | ||
1459 | minsz = offsetofend(struct vfio_group_status, flags); | |
1460 | ||
1461 | if (copy_from_user(&status, (void __user *)arg, minsz)) | |
1462 | return -EFAULT; | |
1463 | ||
1464 | if (status.argsz < minsz) | |
1465 | return -EINVAL; | |
1466 | ||
1467 | status.flags = 0; | |
1468 | ||
1469 | if (vfio_group_viable(group)) | |
1470 | status.flags |= VFIO_GROUP_FLAGS_VIABLE; | |
1471 | ||
1472 | if (group->container) | |
1473 | status.flags |= VFIO_GROUP_FLAGS_CONTAINER_SET; | |
1474 | ||
1475 | if (copy_to_user((void __user *)arg, &status, minsz)) | |
1476 | return -EFAULT; | |
1477 | ||
1478 | ret = 0; | |
1479 | break; | |
1480 | } | |
1481 | case VFIO_GROUP_SET_CONTAINER: | |
1482 | { | |
1483 | int fd; | |
1484 | ||
1485 | if (get_user(fd, (int __user *)arg)) | |
1486 | return -EFAULT; | |
1487 | ||
1488 | if (fd < 0) | |
1489 | return -EINVAL; | |
1490 | ||
1491 | ret = vfio_group_set_container(group, fd); | |
1492 | break; | |
1493 | } | |
1494 | case VFIO_GROUP_UNSET_CONTAINER: | |
1495 | ret = vfio_group_unset_container(group); | |
1496 | break; | |
1497 | case VFIO_GROUP_GET_DEVICE_FD: | |
1498 | { | |
1499 | char *buf; | |
1500 | ||
1501 | buf = strndup_user((const char __user *)arg, PAGE_SIZE); | |
1502 | if (IS_ERR(buf)) | |
1503 | return PTR_ERR(buf); | |
1504 | ||
1505 | ret = vfio_group_get_device_fd(group, buf); | |
1506 | kfree(buf); | |
1507 | break; | |
1508 | } | |
1509 | } | |
1510 | ||
1511 | return ret; | |
1512 | } | |
1513 | ||
1514 | #ifdef CONFIG_COMPAT | |
1515 | static long vfio_group_fops_compat_ioctl(struct file *filep, | |
1516 | unsigned int cmd, unsigned long arg) | |
1517 | { | |
1518 | arg = (unsigned long)compat_ptr(arg); | |
1519 | return vfio_group_fops_unl_ioctl(filep, cmd, arg); | |
1520 | } | |
1521 | #endif /* CONFIG_COMPAT */ | |
1522 | ||
1523 | static int vfio_group_fops_open(struct inode *inode, struct file *filep) | |
1524 | { | |
1525 | struct vfio_group *group; | |
6d6768c6 | 1526 | int opened; |
cba3345c AW |
1527 | |
1528 | group = vfio_group_get_from_minor(iminor(inode)); | |
1529 | if (!group) | |
1530 | return -ENODEV; | |
1531 | ||
03a76b60 AW |
1532 | if (group->noiommu && !capable(CAP_SYS_RAWIO)) { |
1533 | vfio_group_put(group); | |
1534 | return -EPERM; | |
1535 | } | |
1536 | ||
6d6768c6 AW |
1537 | /* Do we need multiple instances of the group open? Seems not. */ |
1538 | opened = atomic_cmpxchg(&group->opened, 0, 1); | |
1539 | if (opened) { | |
1540 | vfio_group_put(group); | |
1541 | return -EBUSY; | |
1542 | } | |
1543 | ||
1544 | /* Is something still in use from a previous open? */ | |
cba3345c | 1545 | if (group->container) { |
6d6768c6 | 1546 | atomic_dec(&group->opened); |
cba3345c AW |
1547 | vfio_group_put(group); |
1548 | return -EBUSY; | |
1549 | } | |
1550 | ||
1551 | filep->private_data = group; | |
1552 | ||
1553 | return 0; | |
1554 | } | |
1555 | ||
1556 | static int vfio_group_fops_release(struct inode *inode, struct file *filep) | |
1557 | { | |
1558 | struct vfio_group *group = filep->private_data; | |
1559 | ||
1560 | filep->private_data = NULL; | |
1561 | ||
1562 | vfio_group_try_dissolve_container(group); | |
1563 | ||
6d6768c6 AW |
1564 | atomic_dec(&group->opened); |
1565 | ||
cba3345c AW |
1566 | vfio_group_put(group); |
1567 | ||
1568 | return 0; | |
1569 | } | |
1570 | ||
1571 | static const struct file_operations vfio_group_fops = { | |
1572 | .owner = THIS_MODULE, | |
1573 | .unlocked_ioctl = vfio_group_fops_unl_ioctl, | |
1574 | #ifdef CONFIG_COMPAT | |
1575 | .compat_ioctl = vfio_group_fops_compat_ioctl, | |
1576 | #endif | |
1577 | .open = vfio_group_fops_open, | |
1578 | .release = vfio_group_fops_release, | |
1579 | }; | |
1580 | ||
1581 | /** | |
1582 | * VFIO Device fd | |
1583 | */ | |
1584 | static int vfio_device_fops_release(struct inode *inode, struct file *filep) | |
1585 | { | |
1586 | struct vfio_device *device = filep->private_data; | |
1587 | ||
1588 | device->ops->release(device->device_data); | |
1589 | ||
1590 | vfio_group_try_dissolve_container(device->group); | |
1591 | ||
1592 | vfio_device_put(device); | |
1593 | ||
1594 | return 0; | |
1595 | } | |
1596 | ||
1597 | static long vfio_device_fops_unl_ioctl(struct file *filep, | |
1598 | unsigned int cmd, unsigned long arg) | |
1599 | { | |
1600 | struct vfio_device *device = filep->private_data; | |
1601 | ||
1602 | if (unlikely(!device->ops->ioctl)) | |
1603 | return -EINVAL; | |
1604 | ||
1605 | return device->ops->ioctl(device->device_data, cmd, arg); | |
1606 | } | |
1607 | ||
1608 | static ssize_t vfio_device_fops_read(struct file *filep, char __user *buf, | |
1609 | size_t count, loff_t *ppos) | |
1610 | { | |
1611 | struct vfio_device *device = filep->private_data; | |
1612 | ||
1613 | if (unlikely(!device->ops->read)) | |
1614 | return -EINVAL; | |
1615 | ||
1616 | return device->ops->read(device->device_data, buf, count, ppos); | |
1617 | } | |
1618 | ||
1619 | static ssize_t vfio_device_fops_write(struct file *filep, | |
1620 | const char __user *buf, | |
1621 | size_t count, loff_t *ppos) | |
1622 | { | |
1623 | struct vfio_device *device = filep->private_data; | |
1624 | ||
1625 | if (unlikely(!device->ops->write)) | |
1626 | return -EINVAL; | |
1627 | ||
1628 | return device->ops->write(device->device_data, buf, count, ppos); | |
1629 | } | |
1630 | ||
1631 | static int vfio_device_fops_mmap(struct file *filep, struct vm_area_struct *vma) | |
1632 | { | |
1633 | struct vfio_device *device = filep->private_data; | |
1634 | ||
1635 | if (unlikely(!device->ops->mmap)) | |
1636 | return -EINVAL; | |
1637 | ||
1638 | return device->ops->mmap(device->device_data, vma); | |
1639 | } | |
1640 | ||
1641 | #ifdef CONFIG_COMPAT | |
1642 | static long vfio_device_fops_compat_ioctl(struct file *filep, | |
1643 | unsigned int cmd, unsigned long arg) | |
1644 | { | |
1645 | arg = (unsigned long)compat_ptr(arg); | |
1646 | return vfio_device_fops_unl_ioctl(filep, cmd, arg); | |
1647 | } | |
1648 | #endif /* CONFIG_COMPAT */ | |
1649 | ||
1650 | static const struct file_operations vfio_device_fops = { | |
1651 | .owner = THIS_MODULE, | |
1652 | .release = vfio_device_fops_release, | |
1653 | .read = vfio_device_fops_read, | |
1654 | .write = vfio_device_fops_write, | |
1655 | .unlocked_ioctl = vfio_device_fops_unl_ioctl, | |
1656 | #ifdef CONFIG_COMPAT | |
1657 | .compat_ioctl = vfio_device_fops_compat_ioctl, | |
1658 | #endif | |
1659 | .mmap = vfio_device_fops_mmap, | |
1660 | }; | |
1661 | ||
6cdd9782 AK |
1662 | /** |
1663 | * External user API, exported by symbols to be linked dynamically. | |
1664 | * | |
1665 | * The protocol includes: | |
1666 | * 1. do normal VFIO init operation: | |
1667 | * - opening a new container; | |
1668 | * - attaching group(s) to it; | |
1669 | * - setting an IOMMU driver for a container. | |
1670 | * When IOMMU is set for a container, all groups in it are | |
1671 | * considered ready to use by an external user. | |
1672 | * | |
1673 | * 2. User space passes a group fd to an external user. | |
1674 | * The external user calls vfio_group_get_external_user() | |
1675 | * to verify that: | |
1676 | * - the group is initialized; | |
1677 | * - IOMMU is set for it. | |
1678 | * If both checks passed, vfio_group_get_external_user() | |
1679 | * increments the container user counter to prevent | |
1680 | * the VFIO group from disposal before KVM exits. | |
1681 | * | |
1682 | * 3. The external user calls vfio_external_user_iommu_id() | |
1683 | * to know an IOMMU ID. | |
1684 | * | |
1685 | * 4. When the external KVM finishes, it calls | |
1686 | * vfio_group_put_external_user() to release the VFIO group. | |
1687 | * This call decrements the container user counter. | |
1688 | */ | |
1689 | struct vfio_group *vfio_group_get_external_user(struct file *filep) | |
1690 | { | |
1691 | struct vfio_group *group = filep->private_data; | |
1692 | ||
1693 | if (filep->f_op != &vfio_group_fops) | |
1694 | return ERR_PTR(-EINVAL); | |
1695 | ||
1696 | if (!atomic_inc_not_zero(&group->container_users)) | |
1697 | return ERR_PTR(-EINVAL); | |
1698 | ||
03a76b60 AW |
1699 | if (group->noiommu) { |
1700 | atomic_dec(&group->container_users); | |
1701 | return ERR_PTR(-EPERM); | |
1702 | } | |
1703 | ||
6cdd9782 AK |
1704 | if (!group->container->iommu_driver || |
1705 | !vfio_group_viable(group)) { | |
1706 | atomic_dec(&group->container_users); | |
1707 | return ERR_PTR(-EINVAL); | |
1708 | } | |
1709 | ||
1710 | vfio_group_get(group); | |
1711 | ||
1712 | return group; | |
1713 | } | |
1714 | EXPORT_SYMBOL_GPL(vfio_group_get_external_user); | |
1715 | ||
1716 | void vfio_group_put_external_user(struct vfio_group *group) | |
1717 | { | |
1718 | vfio_group_put(group); | |
1719 | vfio_group_try_dissolve_container(group); | |
1720 | } | |
1721 | EXPORT_SYMBOL_GPL(vfio_group_put_external_user); | |
1722 | ||
1723 | int vfio_external_user_iommu_id(struct vfio_group *group) | |
1724 | { | |
1725 | return iommu_group_id(group->iommu_group); | |
1726 | } | |
1727 | EXPORT_SYMBOL_GPL(vfio_external_user_iommu_id); | |
1728 | ||
88d7ab89 AW |
1729 | long vfio_external_check_extension(struct vfio_group *group, unsigned long arg) |
1730 | { | |
1731 | return vfio_ioctl_check_extension(group->container, arg); | |
1732 | } | |
1733 | EXPORT_SYMBOL_GPL(vfio_external_check_extension); | |
1734 | ||
cba3345c AW |
1735 | /** |
1736 | * Module/class support | |
1737 | */ | |
1738 | static char *vfio_devnode(struct device *dev, umode_t *mode) | |
1739 | { | |
1740 | return kasprintf(GFP_KERNEL, "vfio/%s", dev_name(dev)); | |
1741 | } | |
1742 | ||
d1099901 AW |
1743 | static struct miscdevice vfio_dev = { |
1744 | .minor = VFIO_MINOR, | |
1745 | .name = "vfio", | |
1746 | .fops = &vfio_fops, | |
1747 | .nodename = "vfio/vfio", | |
1748 | .mode = S_IRUGO | S_IWUGO, | |
1749 | }; | |
1750 | ||
cba3345c AW |
1751 | static int __init vfio_init(void) |
1752 | { | |
1753 | int ret; | |
1754 | ||
1755 | idr_init(&vfio.group_idr); | |
1756 | mutex_init(&vfio.group_lock); | |
1757 | mutex_init(&vfio.iommu_drivers_lock); | |
1758 | INIT_LIST_HEAD(&vfio.group_list); | |
1759 | INIT_LIST_HEAD(&vfio.iommu_drivers_list); | |
1760 | init_waitqueue_head(&vfio.release_q); | |
1761 | ||
d1099901 AW |
1762 | ret = misc_register(&vfio_dev); |
1763 | if (ret) { | |
1764 | pr_err("vfio: misc device register failed\n"); | |
1765 | return ret; | |
1766 | } | |
1767 | ||
1768 | /* /dev/vfio/$GROUP */ | |
cba3345c AW |
1769 | vfio.class = class_create(THIS_MODULE, "vfio"); |
1770 | if (IS_ERR(vfio.class)) { | |
1771 | ret = PTR_ERR(vfio.class); | |
1772 | goto err_class; | |
1773 | } | |
1774 | ||
1775 | vfio.class->devnode = vfio_devnode; | |
1776 | ||
d1099901 | 1777 | ret = alloc_chrdev_region(&vfio.group_devt, 0, MINORMASK, "vfio"); |
cba3345c | 1778 | if (ret) |
d1099901 | 1779 | goto err_alloc_chrdev; |
cba3345c | 1780 | |
cba3345c | 1781 | cdev_init(&vfio.group_cdev, &vfio_group_fops); |
d1099901 | 1782 | ret = cdev_add(&vfio.group_cdev, vfio.group_devt, MINORMASK); |
cba3345c | 1783 | if (ret) |
d1099901 | 1784 | goto err_cdev_add; |
cba3345c AW |
1785 | |
1786 | pr_info(DRIVER_DESC " version: " DRIVER_VERSION "\n"); | |
1787 | ||
73fa0d10 AW |
1788 | /* |
1789 | * Attempt to load known iommu-drivers. This gives us a working | |
1790 | * environment without the user needing to explicitly load iommu | |
1791 | * drivers. | |
1792 | */ | |
1793 | request_module_nowait("vfio_iommu_type1"); | |
5ffd229c | 1794 | request_module_nowait("vfio_iommu_spapr_tce"); |
73fa0d10 | 1795 | |
03a76b60 AW |
1796 | #ifdef CONFIG_VFIO_NOIOMMU |
1797 | vfio_register_iommu_driver(&vfio_noiommu_ops); | |
1798 | #endif | |
cba3345c AW |
1799 | return 0; |
1800 | ||
d1099901 AW |
1801 | err_cdev_add: |
1802 | unregister_chrdev_region(vfio.group_devt, MINORMASK); | |
1803 | err_alloc_chrdev: | |
cba3345c AW |
1804 | class_destroy(vfio.class); |
1805 | vfio.class = NULL; | |
1806 | err_class: | |
d1099901 | 1807 | misc_deregister(&vfio_dev); |
cba3345c AW |
1808 | return ret; |
1809 | } | |
1810 | ||
1811 | static void __exit vfio_cleanup(void) | |
1812 | { | |
1813 | WARN_ON(!list_empty(&vfio.group_list)); | |
1814 | ||
03a76b60 AW |
1815 | #ifdef CONFIG_VFIO_NOIOMMU |
1816 | vfio_unregister_iommu_driver(&vfio_noiommu_ops); | |
1817 | #endif | |
cba3345c AW |
1818 | idr_destroy(&vfio.group_idr); |
1819 | cdev_del(&vfio.group_cdev); | |
d1099901 | 1820 | unregister_chrdev_region(vfio.group_devt, MINORMASK); |
cba3345c AW |
1821 | class_destroy(vfio.class); |
1822 | vfio.class = NULL; | |
d1099901 | 1823 | misc_deregister(&vfio_dev); |
cba3345c AW |
1824 | } |
1825 | ||
1826 | module_init(vfio_init); | |
1827 | module_exit(vfio_cleanup); | |
1828 | ||
1829 | MODULE_VERSION(DRIVER_VERSION); | |
1830 | MODULE_LICENSE("GPL v2"); | |
1831 | MODULE_AUTHOR(DRIVER_AUTHOR); | |
1832 | MODULE_DESCRIPTION(DRIVER_DESC); | |
d1099901 AW |
1833 | MODULE_ALIAS_MISCDEV(VFIO_MINOR); |
1834 | MODULE_ALIAS("devname:vfio/vfio"); |