xref: /linux/drivers/vdpa/vdpa_user/vduse_dev.c (revision dd093fb0) 
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VDUSE: vDPA Device in Userspace
4  *
5  * Copyright (C) 2020-2021 Bytedance Inc. and/or its affiliates. All rights reserved.
6  *
7  * Author: Xie Yongji <xieyongji@bytedance.com>
8  *
9  */
10 
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/cdev.h>
14 #include <linux/device.h>
15 #include <linux/eventfd.h>
16 #include <linux/slab.h>
17 #include <linux/wait.h>
18 #include <linux/dma-map-ops.h>
19 #include <linux/poll.h>
20 #include <linux/file.h>
21 #include <linux/uio.h>
22 #include <linux/vdpa.h>
23 #include <linux/nospec.h>
24 #include <linux/vmalloc.h>
25 #include <linux/sched/mm.h>
26 #include <uapi/linux/vduse.h>
27 #include <uapi/linux/vdpa.h>
28 #include <uapi/linux/virtio_config.h>
29 #include <uapi/linux/virtio_ids.h>
30 #include <uapi/linux/virtio_blk.h>
31 #include <linux/mod_devicetable.h>
32 
33 #include "iova_domain.h"
34 
35 #define DRV_AUTHOR   "Yongji Xie <xieyongji@bytedance.com>"
36 #define DRV_DESC     "vDPA Device in Userspace"
37 #define DRV_LICENSE  "GPL v2"
38 
39 #define VDUSE_DEV_MAX (1U << MINORBITS)
40 #define VDUSE_BOUNCE_SIZE (64 * 1024 * 1024)
41 #define VDUSE_IOVA_SIZE (128 * 1024 * 1024)
42 #define VDUSE_MSG_DEFAULT_TIMEOUT 30
43 
44 struct vduse_virtqueue {
45 	u16 index;
46 	u16 num_max;
47 	u32 num;
48 	u64 desc_addr;
49 	u64 driver_addr;
50 	u64 device_addr;
51 	struct vdpa_vq_state state;
52 	bool ready;
53 	bool kicked;
54 	spinlock_t kick_lock;
55 	spinlock_t irq_lock;
56 	struct eventfd_ctx *kickfd;
57 	struct vdpa_callback cb;
58 	struct work_struct inject;
59 	struct work_struct kick;
60 };
61 
62 struct vduse_dev;
63 
64 struct vduse_vdpa {
65 	struct vdpa_device vdpa;
66 	struct vduse_dev *dev;
67 };
68 
69 struct vduse_umem {
70 	unsigned long iova;
71 	unsigned long npages;
72 	struct page **pages;
73 	struct mm_struct *mm;
74 };
75 
76 struct vduse_dev {
77 	struct vduse_vdpa *vdev;
78 	struct device *dev;
79 	struct vduse_virtqueue *vqs;
80 	struct vduse_iova_domain *domain;
81 	char *name;
82 	struct mutex lock;
83 	spinlock_t msg_lock;
84 	u64 msg_unique;
85 	u32 msg_timeout;
86 	wait_queue_head_t waitq;
87 	struct list_head send_list;
88 	struct list_head recv_list;
89 	struct vdpa_callback config_cb;
90 	struct work_struct inject;
91 	spinlock_t irq_lock;
92 	struct rw_semaphore rwsem;
93 	int minor;
94 	bool broken;
95 	bool connected;
96 	u64 api_version;
97 	u64 device_features;
98 	u64 driver_features;
99 	u32 device_id;
100 	u32 vendor_id;
101 	u32 generation;
102 	u32 config_size;
103 	void *config;
104 	u8 status;
105 	u32 vq_num;
106 	u32 vq_align;
107 	struct vduse_umem *umem;
108 	struct mutex mem_lock;
109 };
110 
111 struct vduse_dev_msg {
112 	struct vduse_dev_request req;
113 	struct vduse_dev_response resp;
114 	struct list_head list;
115 	wait_queue_head_t waitq;
116 	bool completed;
117 };
118 
119 struct vduse_control {
120 	u64 api_version;
121 };
122 
123 static DEFINE_MUTEX(vduse_lock);
124 static DEFINE_IDR(vduse_idr);
125 
126 static dev_t vduse_major;
127 static struct class *vduse_class;
128 static struct cdev vduse_ctrl_cdev;
129 static struct cdev vduse_cdev;
130 static struct workqueue_struct *vduse_irq_wq;
131 
132 static u32 allowed_device_id[] = {
133 	VIRTIO_ID_BLOCK,
134 };
135 
136 static inline struct vduse_dev *vdpa_to_vduse(struct vdpa_device *vdpa)
137 {
138 	struct vduse_vdpa *vdev = container_of(vdpa, struct vduse_vdpa, vdpa);
139 
140 	return vdev->dev;
141 }
142 
143 static inline struct vduse_dev *dev_to_vduse(struct device *dev)
144 {
145 	struct vdpa_device *vdpa = dev_to_vdpa(dev);
146 
147 	return vdpa_to_vduse(vdpa);
148 }
149 
150 static struct vduse_dev_msg *vduse_find_msg(struct list_head *head,
151 					    uint32_t request_id)
152 {
153 	struct vduse_dev_msg *msg;
154 
155 	list_for_each_entry(msg, head, list) {
156 		if (msg->req.request_id == request_id) {
157 			list_del(&msg->list);
158 			return msg;
159 		}
160 	}
161 
162 	return NULL;
163 }
164 
165 static struct vduse_dev_msg *vduse_dequeue_msg(struct list_head *head)
166 {
167 	struct vduse_dev_msg *msg = NULL;
168 
169 	if (!list_empty(head)) {
170 		msg = list_first_entry(head, struct vduse_dev_msg, list);
171 		list_del(&msg->list);
172 	}
173 
174 	return msg;
175 }
176 
177 static void vduse_enqueue_msg(struct list_head *head,
178 			      struct vduse_dev_msg *msg)
179 {
180 	list_add_tail(&msg->list, head);
181 }
182 
183 static void vduse_dev_broken(struct vduse_dev *dev)
184 {
185 	struct vduse_dev_msg *msg, *tmp;
186 
187 	if (unlikely(dev->broken))
188 		return;
189 
190 	list_splice_init(&dev->recv_list, &dev->send_list);
191 	list_for_each_entry_safe(msg, tmp, &dev->send_list, list) {
192 		list_del(&msg->list);
193 		msg->completed = 1;
194 		msg->resp.result = VDUSE_REQ_RESULT_FAILED;
195 		wake_up(&msg->waitq);
196 	}
197 	dev->broken = true;
198 	wake_up(&dev->waitq);
199 }
200 
201 static int vduse_dev_msg_sync(struct vduse_dev *dev,
202 			      struct vduse_dev_msg *msg)
203 {
204 	int ret;
205 
206 	if (unlikely(dev->broken))
207 		return -EIO;
208 
209 	init_waitqueue_head(&msg->waitq);
210 	spin_lock(&dev->msg_lock);
211 	if (unlikely(dev->broken)) {
212 		spin_unlock(&dev->msg_lock);
213 		return -EIO;
214 	}
215 	msg->req.request_id = dev->msg_unique++;
216 	vduse_enqueue_msg(&dev->send_list, msg);
217 	wake_up(&dev->waitq);
218 	spin_unlock(&dev->msg_lock);
219 	if (dev->msg_timeout)
220 		ret = wait_event_killable_timeout(msg->waitq, msg->completed,
221 						  (long)dev->msg_timeout * HZ);
222 	else
223 		ret = wait_event_killable(msg->waitq, msg->completed);
224 
225 	spin_lock(&dev->msg_lock);
226 	if (!msg->completed) {
227 		list_del(&msg->list);
228 		msg->resp.result = VDUSE_REQ_RESULT_FAILED;
229 		/* Mark the device as malfunction when there is a timeout */
230 		if (!ret)
231 			vduse_dev_broken(dev);
232 	}
233 	ret = (msg->resp.result == VDUSE_REQ_RESULT_OK) ? 0 : -EIO;
234 	spin_unlock(&dev->msg_lock);
235 
236 	return ret;
237 }
238 
239 static int vduse_dev_get_vq_state_packed(struct vduse_dev *dev,
240 					 struct vduse_virtqueue *vq,
241 					 struct vdpa_vq_state_packed *packed)
242 {
243 	struct vduse_dev_msg msg = { 0 };
244 	int ret;
245 
246 	msg.req.type = VDUSE_GET_VQ_STATE;
247 	msg.req.vq_state.index = vq->index;
248 
249 	ret = vduse_dev_msg_sync(dev, &msg);
250 	if (ret)
251 		return ret;
252 
253 	packed->last_avail_counter =
254 			msg.resp.vq_state.packed.last_avail_counter & 0x0001;
255 	packed->last_avail_idx =
256 			msg.resp.vq_state.packed.last_avail_idx & 0x7FFF;
257 	packed->last_used_counter =
258 			msg.resp.vq_state.packed.last_used_counter & 0x0001;
259 	packed->last_used_idx =
260 			msg.resp.vq_state.packed.last_used_idx & 0x7FFF;
261 
262 	return 0;
263 }
264 
265 static int vduse_dev_get_vq_state_split(struct vduse_dev *dev,
266 					struct vduse_virtqueue *vq,
267 					struct vdpa_vq_state_split *split)
268 {
269 	struct vduse_dev_msg msg = { 0 };
270 	int ret;
271 
272 	msg.req.type = VDUSE_GET_VQ_STATE;
273 	msg.req.vq_state.index = vq->index;
274 
275 	ret = vduse_dev_msg_sync(dev, &msg);
276 	if (ret)
277 		return ret;
278 
279 	split->avail_index = msg.resp.vq_state.split.avail_index;
280 
281 	return 0;
282 }
283 
284 static int vduse_dev_set_status(struct vduse_dev *dev, u8 status)
285 {
286 	struct vduse_dev_msg msg = { 0 };
287 
288 	msg.req.type = VDUSE_SET_STATUS;
289 	msg.req.s.status = status;
290 
291 	return vduse_dev_msg_sync(dev, &msg);
292 }
293 
294 static int vduse_dev_update_iotlb(struct vduse_dev *dev,
295 				  u64 start, u64 last)
296 {
297 	struct vduse_dev_msg msg = { 0 };
298 
299 	if (last < start)
300 		return -EINVAL;
301 
302 	msg.req.type = VDUSE_UPDATE_IOTLB;
303 	msg.req.iova.start = start;
304 	msg.req.iova.last = last;
305 
306 	return vduse_dev_msg_sync(dev, &msg);
307 }
308 
309 static ssize_t vduse_dev_read_iter(struct kiocb *iocb, struct iov_iter *to)
310 {
311 	struct file *file = iocb->ki_filp;
312 	struct vduse_dev *dev = file->private_data;
313 	struct vduse_dev_msg *msg;
314 	int size = sizeof(struct vduse_dev_request);
315 	ssize_t ret;
316 
317 	if (iov_iter_count(to) < size)
318 		return -EINVAL;
319 
320 	spin_lock(&dev->msg_lock);
321 	while (1) {
322 		msg = vduse_dequeue_msg(&dev->send_list);
323 		if (msg)
324 			break;
325 
326 		ret = -EAGAIN;
327 		if (file->f_flags & O_NONBLOCK)
328 			goto unlock;
329 
330 		spin_unlock(&dev->msg_lock);
331 		ret = wait_event_interruptible_exclusive(dev->waitq,
332 					!list_empty(&dev->send_list));
333 		if (ret)
334 			return ret;
335 
336 		spin_lock(&dev->msg_lock);
337 	}
338 	spin_unlock(&dev->msg_lock);
339 	ret = copy_to_iter(&msg->req, size, to);
340 	spin_lock(&dev->msg_lock);
341 	if (ret != size) {
342 		ret = -EFAULT;
343 		vduse_enqueue_msg(&dev->send_list, msg);
344 		goto unlock;
345 	}
346 	vduse_enqueue_msg(&dev->recv_list, msg);
347 unlock:
348 	spin_unlock(&dev->msg_lock);
349 
350 	return ret;
351 }
352 
353 static bool is_mem_zero(const char *ptr, int size)
354 {
355 	int i;
356 
357 	for (i = 0; i < size; i++) {
358 		if (ptr[i])
359 			return false;
360 	}
361 	return true;
362 }
363 
364 static ssize_t vduse_dev_write_iter(struct kiocb *iocb, struct iov_iter *from)
365 {
366 	struct file *file = iocb->ki_filp;
367 	struct vduse_dev *dev = file->private_data;
368 	struct vduse_dev_response resp;
369 	struct vduse_dev_msg *msg;
370 	size_t ret;
371 
372 	ret = copy_from_iter(&resp, sizeof(resp), from);
373 	if (ret != sizeof(resp))
374 		return -EINVAL;
375 
376 	if (!is_mem_zero((const char *)resp.reserved, sizeof(resp.reserved)))
377 		return -EINVAL;
378 
379 	spin_lock(&dev->msg_lock);
380 	msg = vduse_find_msg(&dev->recv_list, resp.request_id);
381 	if (!msg) {
382 		ret = -ENOENT;
383 		goto unlock;
384 	}
385 
386 	memcpy(&msg->resp, &resp, sizeof(resp));
387 	msg->completed = 1;
388 	wake_up(&msg->waitq);
389 unlock:
390 	spin_unlock(&dev->msg_lock);
391 
392 	return ret;
393 }
394 
395 static __poll_t vduse_dev_poll(struct file *file, poll_table *wait)
396 {
397 	struct vduse_dev *dev = file->private_data;
398 	__poll_t mask = 0;
399 
400 	poll_wait(file, &dev->waitq, wait);
401 
402 	spin_lock(&dev->msg_lock);
403 
404 	if (unlikely(dev->broken))
405 		mask |= EPOLLERR;
406 	if (!list_empty(&dev->send_list))
407 		mask |= EPOLLIN | EPOLLRDNORM;
408 	if (!list_empty(&dev->recv_list))
409 		mask |= EPOLLOUT | EPOLLWRNORM;
410 
411 	spin_unlock(&dev->msg_lock);
412 
413 	return mask;
414 }
415 
416 static void vduse_dev_reset(struct vduse_dev *dev)
417 {
418 	int i;
419 	struct vduse_iova_domain *domain = dev->domain;
420 
421 	/* The coherent mappings are handled in vduse_dev_free_coherent() */
422 	if (domain->bounce_map)
423 		vduse_domain_reset_bounce_map(domain);
424 
425 	down_write(&dev->rwsem);
426 
427 	dev->status = 0;
428 	dev->driver_features = 0;
429 	dev->generation++;
430 	spin_lock(&dev->irq_lock);
431 	dev->config_cb.callback = NULL;
432 	dev->config_cb.private = NULL;
433 	spin_unlock(&dev->irq_lock);
434 	flush_work(&dev->inject);
435 
436 	for (i = 0; i < dev->vq_num; i++) {
437 		struct vduse_virtqueue *vq = &dev->vqs[i];
438 
439 		vq->ready = false;
440 		vq->desc_addr = 0;
441 		vq->driver_addr = 0;
442 		vq->device_addr = 0;
443 		vq->num = 0;
444 		memset(&vq->state, 0, sizeof(vq->state));
445 
446 		spin_lock(&vq->kick_lock);
447 		vq->kicked = false;
448 		if (vq->kickfd)
449 			eventfd_ctx_put(vq->kickfd);
450 		vq->kickfd = NULL;
451 		spin_unlock(&vq->kick_lock);
452 
453 		spin_lock(&vq->irq_lock);
454 		vq->cb.callback = NULL;
455 		vq->cb.private = NULL;
456 		spin_unlock(&vq->irq_lock);
457 		flush_work(&vq->inject);
458 		flush_work(&vq->kick);
459 	}
460 
461 	up_write(&dev->rwsem);
462 }
463 
464 static int vduse_vdpa_set_vq_address(struct vdpa_device *vdpa, u16 idx,
465 				u64 desc_area, u64 driver_area,
466 				u64 device_area)
467 {
468 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
469 	struct vduse_virtqueue *vq = &dev->vqs[idx];
470 
471 	vq->desc_addr = desc_area;
472 	vq->driver_addr = driver_area;
473 	vq->device_addr = device_area;
474 
475 	return 0;
476 }
477 
478 static void vduse_vq_kick(struct vduse_virtqueue *vq)
479 {
480 	spin_lock(&vq->kick_lock);
481 	if (!vq->ready)
482 		goto unlock;
483 
484 	if (vq->kickfd)
485 		eventfd_signal(vq->kickfd, 1);
486 	else
487 		vq->kicked = true;
488 unlock:
489 	spin_unlock(&vq->kick_lock);
490 }
491 
492 static void vduse_vq_kick_work(struct work_struct *work)
493 {
494 	struct vduse_virtqueue *vq = container_of(work,
495 					struct vduse_virtqueue, kick);
496 
497 	vduse_vq_kick(vq);
498 }
499 
500 static void vduse_vdpa_kick_vq(struct vdpa_device *vdpa, u16 idx)
501 {
502 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
503 	struct vduse_virtqueue *vq = &dev->vqs[idx];
504 
505 	if (!eventfd_signal_allowed()) {
506 		schedule_work(&vq->kick);
507 		return;
508 	}
509 	vduse_vq_kick(vq);
510 }
511 
512 static void vduse_vdpa_set_vq_cb(struct vdpa_device *vdpa, u16 idx,
513 			      struct vdpa_callback *cb)
514 {
515 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
516 	struct vduse_virtqueue *vq = &dev->vqs[idx];
517 
518 	spin_lock(&vq->irq_lock);
519 	vq->cb.callback = cb->callback;
520 	vq->cb.private = cb->private;
521 	spin_unlock(&vq->irq_lock);
522 }
523 
524 static void vduse_vdpa_set_vq_num(struct vdpa_device *vdpa, u16 idx, u32 num)
525 {
526 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
527 	struct vduse_virtqueue *vq = &dev->vqs[idx];
528 
529 	vq->num = num;
530 }
531 
532 static void vduse_vdpa_set_vq_ready(struct vdpa_device *vdpa,
533 					u16 idx, bool ready)
534 {
535 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
536 	struct vduse_virtqueue *vq = &dev->vqs[idx];
537 
538 	vq->ready = ready;
539 }
540 
541 static bool vduse_vdpa_get_vq_ready(struct vdpa_device *vdpa, u16 idx)
542 {
543 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
544 	struct vduse_virtqueue *vq = &dev->vqs[idx];
545 
546 	return vq->ready;
547 }
548 
549 static int vduse_vdpa_set_vq_state(struct vdpa_device *vdpa, u16 idx,
550 				const struct vdpa_vq_state *state)
551 {
552 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
553 	struct vduse_virtqueue *vq = &dev->vqs[idx];
554 
555 	if (dev->driver_features & BIT_ULL(VIRTIO_F_RING_PACKED)) {
556 		vq->state.packed.last_avail_counter =
557 				state->packed.last_avail_counter;
558 		vq->state.packed.last_avail_idx = state->packed.last_avail_idx;
559 		vq->state.packed.last_used_counter =
560 				state->packed.last_used_counter;
561 		vq->state.packed.last_used_idx = state->packed.last_used_idx;
562 	} else
563 		vq->state.split.avail_index = state->split.avail_index;
564 
565 	return 0;
566 }
567 
568 static int vduse_vdpa_get_vq_state(struct vdpa_device *vdpa, u16 idx,
569 				struct vdpa_vq_state *state)
570 {
571 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
572 	struct vduse_virtqueue *vq = &dev->vqs[idx];
573 
574 	if (dev->driver_features & BIT_ULL(VIRTIO_F_RING_PACKED))
575 		return vduse_dev_get_vq_state_packed(dev, vq, &state->packed);
576 
577 	return vduse_dev_get_vq_state_split(dev, vq, &state->split);
578 }
579 
580 static u32 vduse_vdpa_get_vq_align(struct vdpa_device *vdpa)
581 {
582 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
583 
584 	return dev->vq_align;
585 }
586 
587 static u64 vduse_vdpa_get_device_features(struct vdpa_device *vdpa)
588 {
589 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
590 
591 	return dev->device_features;
592 }
593 
594 static int vduse_vdpa_set_driver_features(struct vdpa_device *vdpa, u64 features)
595 {
596 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
597 
598 	dev->driver_features = features;
599 	return 0;
600 }
601 
602 static u64 vduse_vdpa_get_driver_features(struct vdpa_device *vdpa)
603 {
604 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
605 
606 	return dev->driver_features;
607 }
608 
609 static void vduse_vdpa_set_config_cb(struct vdpa_device *vdpa,
610 				  struct vdpa_callback *cb)
611 {
612 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
613 
614 	spin_lock(&dev->irq_lock);
615 	dev->config_cb.callback = cb->callback;
616 	dev->config_cb.private = cb->private;
617 	spin_unlock(&dev->irq_lock);
618 }
619 
620 static u16 vduse_vdpa_get_vq_num_max(struct vdpa_device *vdpa)
621 {
622 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
623 	u16 num_max = 0;
624 	int i;
625 
626 	for (i = 0; i < dev->vq_num; i++)
627 		if (num_max < dev->vqs[i].num_max)
628 			num_max = dev->vqs[i].num_max;
629 
630 	return num_max;
631 }
632 
633 static u32 vduse_vdpa_get_device_id(struct vdpa_device *vdpa)
634 {
635 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
636 
637 	return dev->device_id;
638 }
639 
640 static u32 vduse_vdpa_get_vendor_id(struct vdpa_device *vdpa)
641 {
642 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
643 
644 	return dev->vendor_id;
645 }
646 
647 static u8 vduse_vdpa_get_status(struct vdpa_device *vdpa)
648 {
649 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
650 
651 	return dev->status;
652 }
653 
654 static void vduse_vdpa_set_status(struct vdpa_device *vdpa, u8 status)
655 {
656 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
657 
658 	if (vduse_dev_set_status(dev, status))
659 		return;
660 
661 	dev->status = status;
662 }
663 
664 static size_t vduse_vdpa_get_config_size(struct vdpa_device *vdpa)
665 {
666 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
667 
668 	return dev->config_size;
669 }
670 
671 static void vduse_vdpa_get_config(struct vdpa_device *vdpa, unsigned int offset,
672 				  void *buf, unsigned int len)
673 {
674 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
675 
676 	/* Initialize the buffer in case of partial copy. */
677 	memset(buf, 0, len);
678 
679 	if (offset > dev->config_size)
680 		return;
681 
682 	if (len > dev->config_size - offset)
683 		len = dev->config_size - offset;
684 
685 	memcpy(buf, dev->config + offset, len);
686 }
687 
688 static void vduse_vdpa_set_config(struct vdpa_device *vdpa, unsigned int offset,
689 			const void *buf, unsigned int len)
690 {
691 	/* Now we only support read-only configuration space */
692 }
693 
694 static int vduse_vdpa_reset(struct vdpa_device *vdpa)
695 {
696 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
697 	int ret = vduse_dev_set_status(dev, 0);
698 
699 	vduse_dev_reset(dev);
700 
701 	return ret;
702 }
703 
704 static u32 vduse_vdpa_get_generation(struct vdpa_device *vdpa)
705 {
706 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
707 
708 	return dev->generation;
709 }
710 
711 static int vduse_vdpa_set_map(struct vdpa_device *vdpa,
712 				unsigned int asid,
713 				struct vhost_iotlb *iotlb)
714 {
715 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
716 	int ret;
717 
718 	ret = vduse_domain_set_map(dev->domain, iotlb);
719 	if (ret)
720 		return ret;
721 
722 	ret = vduse_dev_update_iotlb(dev, 0ULL, ULLONG_MAX);
723 	if (ret) {
724 		vduse_domain_clear_map(dev->domain, iotlb);
725 		return ret;
726 	}
727 
728 	return 0;
729 }
730 
731 static void vduse_vdpa_free(struct vdpa_device *vdpa)
732 {
733 	struct vduse_dev *dev = vdpa_to_vduse(vdpa);
734 
735 	dev->vdev = NULL;
736 }
737 
738 static const struct vdpa_config_ops vduse_vdpa_config_ops = {
739 	.set_vq_address		= vduse_vdpa_set_vq_address,
740 	.kick_vq		= vduse_vdpa_kick_vq,
741 	.set_vq_cb		= vduse_vdpa_set_vq_cb,
742 	.set_vq_num             = vduse_vdpa_set_vq_num,
743 	.set_vq_ready		= vduse_vdpa_set_vq_ready,
744 	.get_vq_ready		= vduse_vdpa_get_vq_ready,
745 	.set_vq_state		= vduse_vdpa_set_vq_state,
746 	.get_vq_state		= vduse_vdpa_get_vq_state,
747 	.get_vq_align		= vduse_vdpa_get_vq_align,
748 	.get_device_features	= vduse_vdpa_get_device_features,
749 	.set_driver_features	= vduse_vdpa_set_driver_features,
750 	.get_driver_features	= vduse_vdpa_get_driver_features,
751 	.set_config_cb		= vduse_vdpa_set_config_cb,
752 	.get_vq_num_max		= vduse_vdpa_get_vq_num_max,
753 	.get_device_id		= vduse_vdpa_get_device_id,
754 	.get_vendor_id		= vduse_vdpa_get_vendor_id,
755 	.get_status		= vduse_vdpa_get_status,
756 	.set_status		= vduse_vdpa_set_status,
757 	.get_config_size	= vduse_vdpa_get_config_size,
758 	.get_config		= vduse_vdpa_get_config,
759 	.set_config		= vduse_vdpa_set_config,
760 	.get_generation		= vduse_vdpa_get_generation,
761 	.reset			= vduse_vdpa_reset,
762 	.set_map		= vduse_vdpa_set_map,
763 	.free			= vduse_vdpa_free,
764 };
765 
766 static dma_addr_t vduse_dev_map_page(struct device *dev, struct page *page,
767 				     unsigned long offset, size_t size,
768 				     enum dma_data_direction dir,
769 				     unsigned long attrs)
770 {
771 	struct vduse_dev *vdev = dev_to_vduse(dev);
772 	struct vduse_iova_domain *domain = vdev->domain;
773 
774 	return vduse_domain_map_page(domain, page, offset, size, dir, attrs);
775 }
776 
777 static void vduse_dev_unmap_page(struct device *dev, dma_addr_t dma_addr,
778 				size_t size, enum dma_data_direction dir,
779 				unsigned long attrs)
780 {
781 	struct vduse_dev *vdev = dev_to_vduse(dev);
782 	struct vduse_iova_domain *domain = vdev->domain;
783 
784 	return vduse_domain_unmap_page(domain, dma_addr, size, dir, attrs);
785 }
786 
787 static void *vduse_dev_alloc_coherent(struct device *dev, size_t size,
788 					dma_addr_t *dma_addr, gfp_t flag,
789 					unsigned long attrs)
790 {
791 	struct vduse_dev *vdev = dev_to_vduse(dev);
792 	struct vduse_iova_domain *domain = vdev->domain;
793 	unsigned long iova;
794 	void *addr;
795 
796 	*dma_addr = DMA_MAPPING_ERROR;
797 	addr = vduse_domain_alloc_coherent(domain, size,
798 				(dma_addr_t *)&iova, flag, attrs);
799 	if (!addr)
800 		return NULL;
801 
802 	*dma_addr = (dma_addr_t)iova;
803 
804 	return addr;
805 }
806 
807 static void vduse_dev_free_coherent(struct device *dev, size_t size,
808 					void *vaddr, dma_addr_t dma_addr,
809 					unsigned long attrs)
810 {
811 	struct vduse_dev *vdev = dev_to_vduse(dev);
812 	struct vduse_iova_domain *domain = vdev->domain;
813 
814 	vduse_domain_free_coherent(domain, size, vaddr, dma_addr, attrs);
815 }
816 
817 static size_t vduse_dev_max_mapping_size(struct device *dev)
818 {
819 	struct vduse_dev *vdev = dev_to_vduse(dev);
820 	struct vduse_iova_domain *domain = vdev->domain;
821 
822 	return domain->bounce_size;
823 }
824 
825 static const struct dma_map_ops vduse_dev_dma_ops = {
826 	.map_page = vduse_dev_map_page,
827 	.unmap_page = vduse_dev_unmap_page,
828 	.alloc = vduse_dev_alloc_coherent,
829 	.free = vduse_dev_free_coherent,
830 	.max_mapping_size = vduse_dev_max_mapping_size,
831 };
832 
833 static unsigned int perm_to_file_flags(u8 perm)
834 {
835 	unsigned int flags = 0;
836 
837 	switch (perm) {
838 	case VDUSE_ACCESS_WO:
839 		flags |= O_WRONLY;
840 		break;
841 	case VDUSE_ACCESS_RO:
842 		flags |= O_RDONLY;
843 		break;
844 	case VDUSE_ACCESS_RW:
845 		flags |= O_RDWR;
846 		break;
847 	default:
848 		WARN(1, "invalidate vhost IOTLB permission\n");
849 		break;
850 	}
851 
852 	return flags;
853 }
854 
855 static int vduse_kickfd_setup(struct vduse_dev *dev,
856 			struct vduse_vq_eventfd *eventfd)
857 {
858 	struct eventfd_ctx *ctx = NULL;
859 	struct vduse_virtqueue *vq;
860 	u32 index;
861 
862 	if (eventfd->index >= dev->vq_num)
863 		return -EINVAL;
864 
865 	index = array_index_nospec(eventfd->index, dev->vq_num);
866 	vq = &dev->vqs[index];
867 	if (eventfd->fd >= 0) {
868 		ctx = eventfd_ctx_fdget(eventfd->fd);
869 		if (IS_ERR(ctx))
870 			return PTR_ERR(ctx);
871 	} else if (eventfd->fd != VDUSE_EVENTFD_DEASSIGN)
872 		return 0;
873 
874 	spin_lock(&vq->kick_lock);
875 	if (vq->kickfd)
876 		eventfd_ctx_put(vq->kickfd);
877 	vq->kickfd = ctx;
878 	if (vq->ready && vq->kicked && vq->kickfd) {
879 		eventfd_signal(vq->kickfd, 1);
880 		vq->kicked = false;
881 	}
882 	spin_unlock(&vq->kick_lock);
883 
884 	return 0;
885 }
886 
887 static bool vduse_dev_is_ready(struct vduse_dev *dev)
888 {
889 	int i;
890 
891 	for (i = 0; i < dev->vq_num; i++)
892 		if (!dev->vqs[i].num_max)
893 			return false;
894 
895 	return true;
896 }
897 
898 static void vduse_dev_irq_inject(struct work_struct *work)
899 {
900 	struct vduse_dev *dev = container_of(work, struct vduse_dev, inject);
901 
902 	spin_lock_irq(&dev->irq_lock);
903 	if (dev->config_cb.callback)
904 		dev->config_cb.callback(dev->config_cb.private);
905 	spin_unlock_irq(&dev->irq_lock);
906 }
907 
908 static void vduse_vq_irq_inject(struct work_struct *work)
909 {
910 	struct vduse_virtqueue *vq = container_of(work,
911 					struct vduse_virtqueue, inject);
912 
913 	spin_lock_irq(&vq->irq_lock);
914 	if (vq->ready && vq->cb.callback)
915 		vq->cb.callback(vq->cb.private);
916 	spin_unlock_irq(&vq->irq_lock);
917 }
918 
919 static int vduse_dev_queue_irq_work(struct vduse_dev *dev,
920 				    struct work_struct *irq_work)
921 {
922 	int ret = -EINVAL;
923 
924 	down_read(&dev->rwsem);
925 	if (!(dev->status & VIRTIO_CONFIG_S_DRIVER_OK))
926 		goto unlock;
927 
928 	ret = 0;
929 	queue_work(vduse_irq_wq, irq_work);
930 unlock:
931 	up_read(&dev->rwsem);
932 
933 	return ret;
934 }
935 
936 static int vduse_dev_dereg_umem(struct vduse_dev *dev,
937 				u64 iova, u64 size)
938 {
939 	int ret;
940 
941 	mutex_lock(&dev->mem_lock);
942 	ret = -ENOENT;
943 	if (!dev->umem)
944 		goto unlock;
945 
946 	ret = -EINVAL;
947 	if (dev->umem->iova != iova || size != dev->domain->bounce_size)
948 		goto unlock;
949 
950 	vduse_domain_remove_user_bounce_pages(dev->domain);
951 	unpin_user_pages_dirty_lock(dev->umem->pages,
952 				    dev->umem->npages, true);
953 	atomic64_sub(dev->umem->npages, &dev->umem->mm->pinned_vm);
954 	mmdrop(dev->umem->mm);
955 	vfree(dev->umem->pages);
956 	kfree(dev->umem);
957 	dev->umem = NULL;
958 	ret = 0;
959 unlock:
960 	mutex_unlock(&dev->mem_lock);
961 	return ret;
962 }
963 
964 static int vduse_dev_reg_umem(struct vduse_dev *dev,
965 			      u64 iova, u64 uaddr, u64 size)
966 {
967 	struct page **page_list = NULL;
968 	struct vduse_umem *umem = NULL;
969 	long pinned = 0;
970 	unsigned long npages, lock_limit;
971 	int ret;
972 
973 	if (!dev->domain->bounce_map ||
974 	    size != dev->domain->bounce_size ||
975 	    iova != 0 || uaddr & ~PAGE_MASK)
976 		return -EINVAL;
977 
978 	mutex_lock(&dev->mem_lock);
979 	ret = -EEXIST;
980 	if (dev->umem)
981 		goto unlock;
982 
983 	ret = -ENOMEM;
984 	npages = size >> PAGE_SHIFT;
985 	page_list = __vmalloc(array_size(npages, sizeof(struct page *)),
986 			      GFP_KERNEL_ACCOUNT);
987 	umem = kzalloc(sizeof(*umem), GFP_KERNEL);
988 	if (!page_list || !umem)
989 		goto unlock;
990 
991 	mmap_read_lock(current->mm);
992 
993 	lock_limit = PFN_DOWN(rlimit(RLIMIT_MEMLOCK));
994 	if (npages + atomic64_read(&current->mm->pinned_vm) > lock_limit)
995 		goto out;
996 
997 	pinned = pin_user_pages(uaddr, npages, FOLL_LONGTERM | FOLL_WRITE,
998 				page_list, NULL);
999 	if (pinned != npages) {
1000 		ret = pinned < 0 ? pinned : -ENOMEM;
1001 		goto out;
1002 	}
1003 
1004 	ret = vduse_domain_add_user_bounce_pages(dev->domain,
1005 						 page_list, pinned);
1006 	if (ret)
1007 		goto out;
1008 
1009 	atomic64_add(npages, &current->mm->pinned_vm);
1010 
1011 	umem->pages = page_list;
1012 	umem->npages = pinned;
1013 	umem->iova = iova;
1014 	umem->mm = current->mm;
1015 	mmgrab(current->mm);
1016 
1017 	dev->umem = umem;
1018 out:
1019 	if (ret && pinned > 0)
1020 		unpin_user_pages(page_list, pinned);
1021 
1022 	mmap_read_unlock(current->mm);
1023 unlock:
1024 	if (ret) {
1025 		vfree(page_list);
1026 		kfree(umem);
1027 	}
1028 	mutex_unlock(&dev->mem_lock);
1029 	return ret;
1030 }
1031 
1032 static long vduse_dev_ioctl(struct file *file, unsigned int cmd,
1033 			    unsigned long arg)
1034 {
1035 	struct vduse_dev *dev = file->private_data;
1036 	void __user *argp = (void __user *)arg;
1037 	int ret;
1038 
1039 	if (unlikely(dev->broken))
1040 		return -EPERM;
1041 
1042 	switch (cmd) {
1043 	case VDUSE_IOTLB_GET_FD: {
1044 		struct vduse_iotlb_entry entry;
1045 		struct vhost_iotlb_map *map;
1046 		struct vdpa_map_file *map_file;
1047 		struct vduse_iova_domain *domain = dev->domain;
1048 		struct file *f = NULL;
1049 
1050 		ret = -EFAULT;
1051 		if (copy_from_user(&entry, argp, sizeof(entry)))
1052 			break;
1053 
1054 		ret = -EINVAL;
1055 		if (entry.start > entry.last)
1056 			break;
1057 
1058 		spin_lock(&domain->iotlb_lock);
1059 		map = vhost_iotlb_itree_first(domain->iotlb,
1060 					      entry.start, entry.last);
1061 		if (map) {
1062 			map_file = (struct vdpa_map_file *)map->opaque;
1063 			f = get_file(map_file->file);
1064 			entry.offset = map_file->offset;
1065 			entry.start = map->start;
1066 			entry.last = map->last;
1067 			entry.perm = map->perm;
1068 		}
1069 		spin_unlock(&domain->iotlb_lock);
1070 		ret = -EINVAL;
1071 		if (!f)
1072 			break;
1073 
1074 		ret = -EFAULT;
1075 		if (copy_to_user(argp, &entry, sizeof(entry))) {
1076 			fput(f);
1077 			break;
1078 		}
1079 		ret = receive_fd(f, perm_to_file_flags(entry.perm));
1080 		fput(f);
1081 		break;
1082 	}
1083 	case VDUSE_DEV_GET_FEATURES:
1084 		/*
1085 		 * Just mirror what driver wrote here.
1086 		 * The driver is expected to check FEATURE_OK later.
1087 		 */
1088 		ret = put_user(dev->driver_features, (u64 __user *)argp);
1089 		break;
1090 	case VDUSE_DEV_SET_CONFIG: {
1091 		struct vduse_config_data config;
1092 		unsigned long size = offsetof(struct vduse_config_data,
1093 					      buffer);
1094 
1095 		ret = -EFAULT;
1096 		if (copy_from_user(&config, argp, size))
1097 			break;
1098 
1099 		ret = -EINVAL;
1100 		if (config.offset > dev->config_size ||
1101 		    config.length == 0 ||
1102 		    config.length > dev->config_size - config.offset)
1103 			break;
1104 
1105 		ret = -EFAULT;
1106 		if (copy_from_user(dev->config + config.offset, argp + size,
1107 				   config.length))
1108 			break;
1109 
1110 		ret = 0;
1111 		break;
1112 	}
1113 	case VDUSE_DEV_INJECT_CONFIG_IRQ:
1114 		ret = vduse_dev_queue_irq_work(dev, &dev->inject);
1115 		break;
1116 	case VDUSE_VQ_SETUP: {
1117 		struct vduse_vq_config config;
1118 		u32 index;
1119 
1120 		ret = -EFAULT;
1121 		if (copy_from_user(&config, argp, sizeof(config)))
1122 			break;
1123 
1124 		ret = -EINVAL;
1125 		if (config.index >= dev->vq_num)
1126 			break;
1127 
1128 		if (!is_mem_zero((const char *)config.reserved,
1129 				 sizeof(config.reserved)))
1130 			break;
1131 
1132 		index = array_index_nospec(config.index, dev->vq_num);
1133 		dev->vqs[index].num_max = config.max_size;
1134 		ret = 0;
1135 		break;
1136 	}
1137 	case VDUSE_VQ_GET_INFO: {
1138 		struct vduse_vq_info vq_info;
1139 		struct vduse_virtqueue *vq;
1140 		u32 index;
1141 
1142 		ret = -EFAULT;
1143 		if (copy_from_user(&vq_info, argp, sizeof(vq_info)))
1144 			break;
1145 
1146 		ret = -EINVAL;
1147 		if (vq_info.index >= dev->vq_num)
1148 			break;
1149 
1150 		index = array_index_nospec(vq_info.index, dev->vq_num);
1151 		vq = &dev->vqs[index];
1152 		vq_info.desc_addr = vq->desc_addr;
1153 		vq_info.driver_addr = vq->driver_addr;
1154 		vq_info.device_addr = vq->device_addr;
1155 		vq_info.num = vq->num;
1156 
1157 		if (dev->driver_features & BIT_ULL(VIRTIO_F_RING_PACKED)) {
1158 			vq_info.packed.last_avail_counter =
1159 				vq->state.packed.last_avail_counter;
1160 			vq_info.packed.last_avail_idx =
1161 				vq->state.packed.last_avail_idx;
1162 			vq_info.packed.last_used_counter =
1163 				vq->state.packed.last_used_counter;
1164 			vq_info.packed.last_used_idx =
1165 				vq->state.packed.last_used_idx;
1166 		} else
1167 			vq_info.split.avail_index =
1168 				vq->state.split.avail_index;
1169 
1170 		vq_info.ready = vq->ready;
1171 
1172 		ret = -EFAULT;
1173 		if (copy_to_user(argp, &vq_info, sizeof(vq_info)))
1174 			break;
1175 
1176 		ret = 0;
1177 		break;
1178 	}
1179 	case VDUSE_VQ_SETUP_KICKFD: {
1180 		struct vduse_vq_eventfd eventfd;
1181 
1182 		ret = -EFAULT;
1183 		if (copy_from_user(&eventfd, argp, sizeof(eventfd)))
1184 			break;
1185 
1186 		ret = vduse_kickfd_setup(dev, &eventfd);
1187 		break;
1188 	}
1189 	case VDUSE_VQ_INJECT_IRQ: {
1190 		u32 index;
1191 
1192 		ret = -EFAULT;
1193 		if (get_user(index, (u32 __user *)argp))
1194 			break;
1195 
1196 		ret = -EINVAL;
1197 		if (index >= dev->vq_num)
1198 			break;
1199 
1200 		index = array_index_nospec(index, dev->vq_num);
1201 		ret = vduse_dev_queue_irq_work(dev, &dev->vqs[index].inject);
1202 		break;
1203 	}
1204 	case VDUSE_IOTLB_REG_UMEM: {
1205 		struct vduse_iova_umem umem;
1206 
1207 		ret = -EFAULT;
1208 		if (copy_from_user(&umem, argp, sizeof(umem)))
1209 			break;
1210 
1211 		ret = -EINVAL;
1212 		if (!is_mem_zero((const char *)umem.reserved,
1213 				 sizeof(umem.reserved)))
1214 			break;
1215 
1216 		ret = vduse_dev_reg_umem(dev, umem.iova,
1217 					 umem.uaddr, umem.size);
1218 		break;
1219 	}
1220 	case VDUSE_IOTLB_DEREG_UMEM: {
1221 		struct vduse_iova_umem umem;
1222 
1223 		ret = -EFAULT;
1224 		if (copy_from_user(&umem, argp, sizeof(umem)))
1225 			break;
1226 
1227 		ret = -EINVAL;
1228 		if (!is_mem_zero((const char *)umem.reserved,
1229 				 sizeof(umem.reserved)))
1230 			break;
1231 
1232 		ret = vduse_dev_dereg_umem(dev, umem.iova,
1233 					   umem.size);
1234 		break;
1235 	}
1236 	case VDUSE_IOTLB_GET_INFO: {
1237 		struct vduse_iova_info info;
1238 		struct vhost_iotlb_map *map;
1239 		struct vduse_iova_domain *domain = dev->domain;
1240 
1241 		ret = -EFAULT;
1242 		if (copy_from_user(&info, argp, sizeof(info)))
1243 			break;
1244 
1245 		ret = -EINVAL;
1246 		if (info.start > info.last)
1247 			break;
1248 
1249 		if (!is_mem_zero((const char *)info.reserved,
1250 				 sizeof(info.reserved)))
1251 			break;
1252 
1253 		spin_lock(&domain->iotlb_lock);
1254 		map = vhost_iotlb_itree_first(domain->iotlb,
1255 					      info.start, info.last);
1256 		if (map) {
1257 			info.start = map->start;
1258 			info.last = map->last;
1259 			info.capability = 0;
1260 			if (domain->bounce_map && map->start == 0 &&
1261 			    map->last == domain->bounce_size - 1)
1262 				info.capability |= VDUSE_IOVA_CAP_UMEM;
1263 		}
1264 		spin_unlock(&domain->iotlb_lock);
1265 		if (!map)
1266 			break;
1267 
1268 		ret = -EFAULT;
1269 		if (copy_to_user(argp, &info, sizeof(info)))
1270 			break;
1271 
1272 		ret = 0;
1273 		break;
1274 	}
1275 	default:
1276 		ret = -ENOIOCTLCMD;
1277 		break;
1278 	}
1279 
1280 	return ret;
1281 }
1282 
1283 static int vduse_dev_release(struct inode *inode, struct file *file)
1284 {
1285 	struct vduse_dev *dev = file->private_data;
1286 
1287 	vduse_dev_dereg_umem(dev, 0, dev->domain->bounce_size);
1288 	spin_lock(&dev->msg_lock);
1289 	/* Make sure the inflight messages can processed after reconncection */
1290 	list_splice_init(&dev->recv_list, &dev->send_list);
1291 	spin_unlock(&dev->msg_lock);
1292 	dev->connected = false;
1293 
1294 	return 0;
1295 }
1296 
1297 static struct vduse_dev *vduse_dev_get_from_minor(int minor)
1298 {
1299 	struct vduse_dev *dev;
1300 
1301 	mutex_lock(&vduse_lock);
1302 	dev = idr_find(&vduse_idr, minor);
1303 	mutex_unlock(&vduse_lock);
1304 
1305 	return dev;
1306 }
1307 
1308 static int vduse_dev_open(struct inode *inode, struct file *file)
1309 {
1310 	int ret;
1311 	struct vduse_dev *dev = vduse_dev_get_from_minor(iminor(inode));
1312 
1313 	if (!dev)
1314 		return -ENODEV;
1315 
1316 	ret = -EBUSY;
1317 	mutex_lock(&dev->lock);
1318 	if (dev->connected)
1319 		goto unlock;
1320 
1321 	ret = 0;
1322 	dev->connected = true;
1323 	file->private_data = dev;
1324 unlock:
1325 	mutex_unlock(&dev->lock);
1326 
1327 	return ret;
1328 }
1329 
1330 static const struct file_operations vduse_dev_fops = {
1331 	.owner		= THIS_MODULE,
1332 	.open		= vduse_dev_open,
1333 	.release	= vduse_dev_release,
1334 	.read_iter	= vduse_dev_read_iter,
1335 	.write_iter	= vduse_dev_write_iter,
1336 	.poll		= vduse_dev_poll,
1337 	.unlocked_ioctl	= vduse_dev_ioctl,
1338 	.compat_ioctl	= compat_ptr_ioctl,
1339 	.llseek		= noop_llseek,
1340 };
1341 
1342 static struct vduse_dev *vduse_dev_create(void)
1343 {
1344 	struct vduse_dev *dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1345 
1346 	if (!dev)
1347 		return NULL;
1348 
1349 	mutex_init(&dev->lock);
1350 	mutex_init(&dev->mem_lock);
1351 	spin_lock_init(&dev->msg_lock);
1352 	INIT_LIST_HEAD(&dev->send_list);
1353 	INIT_LIST_HEAD(&dev->recv_list);
1354 	spin_lock_init(&dev->irq_lock);
1355 	init_rwsem(&dev->rwsem);
1356 
1357 	INIT_WORK(&dev->inject, vduse_dev_irq_inject);
1358 	init_waitqueue_head(&dev->waitq);
1359 
1360 	return dev;
1361 }
1362 
1363 static void vduse_dev_destroy(struct vduse_dev *dev)
1364 {
1365 	kfree(dev);
1366 }
1367 
1368 static struct vduse_dev *vduse_find_dev(const char *name)
1369 {
1370 	struct vduse_dev *dev;
1371 	int id;
1372 
1373 	idr_for_each_entry(&vduse_idr, dev, id)
1374 		if (!strcmp(dev->name, name))
1375 			return dev;
1376 
1377 	return NULL;
1378 }
1379 
1380 static int vduse_destroy_dev(char *name)
1381 {
1382 	struct vduse_dev *dev = vduse_find_dev(name);
1383 
1384 	if (!dev)
1385 		return -EINVAL;
1386 
1387 	mutex_lock(&dev->lock);
1388 	if (dev->vdev || dev->connected) {
1389 		mutex_unlock(&dev->lock);
1390 		return -EBUSY;
1391 	}
1392 	dev->connected = true;
1393 	mutex_unlock(&dev->lock);
1394 
1395 	vduse_dev_reset(dev);
1396 	device_destroy(vduse_class, MKDEV(MAJOR(vduse_major), dev->minor));
1397 	idr_remove(&vduse_idr, dev->minor);
1398 	kvfree(dev->config);
1399 	kfree(dev->vqs);
1400 	vduse_domain_destroy(dev->domain);
1401 	kfree(dev->name);
1402 	vduse_dev_destroy(dev);
1403 	module_put(THIS_MODULE);
1404 
1405 	return 0;
1406 }
1407 
1408 static bool device_is_allowed(u32 device_id)
1409 {
1410 	int i;
1411 
1412 	for (i = 0; i < ARRAY_SIZE(allowed_device_id); i++)
1413 		if (allowed_device_id[i] == device_id)
1414 			return true;
1415 
1416 	return false;
1417 }
1418 
1419 static bool features_is_valid(u64 features)
1420 {
1421 	if (!(features & (1ULL << VIRTIO_F_ACCESS_PLATFORM)))
1422 		return false;
1423 
1424 	/* Now we only support read-only configuration space */
1425 	if (features & (1ULL << VIRTIO_BLK_F_CONFIG_WCE))
1426 		return false;
1427 
1428 	return true;
1429 }
1430 
1431 static bool vduse_validate_config(struct vduse_dev_config *config)
1432 {
1433 	if (!is_mem_zero((const char *)config->reserved,
1434 			 sizeof(config->reserved)))
1435 		return false;
1436 
1437 	if (config->vq_align > PAGE_SIZE)
1438 		return false;
1439 
1440 	if (config->config_size > PAGE_SIZE)
1441 		return false;
1442 
1443 	if (config->vq_num > 0xffff)
1444 		return false;
1445 
1446 	if (!device_is_allowed(config->device_id))
1447 		return false;
1448 
1449 	if (!features_is_valid(config->features))
1450 		return false;
1451 
1452 	return true;
1453 }
1454 
1455 static ssize_t msg_timeout_show(struct device *device,
1456 				struct device_attribute *attr, char *buf)
1457 {
1458 	struct vduse_dev *dev = dev_get_drvdata(device);
1459 
1460 	return sysfs_emit(buf, "%u\n", dev->msg_timeout);
1461 }
1462 
1463 static ssize_t msg_timeout_store(struct device *device,
1464 				 struct device_attribute *attr,
1465 				 const char *buf, size_t count)
1466 {
1467 	struct vduse_dev *dev = dev_get_drvdata(device);
1468 	int ret;
1469 
1470 	ret = kstrtouint(buf, 10, &dev->msg_timeout);
1471 	if (ret < 0)
1472 		return ret;
1473 
1474 	return count;
1475 }
1476 
1477 static DEVICE_ATTR_RW(msg_timeout);
1478 
1479 static struct attribute *vduse_dev_attrs[] = {
1480 	&dev_attr_msg_timeout.attr,
1481 	NULL
1482 };
1483 
1484 ATTRIBUTE_GROUPS(vduse_dev);
1485 
1486 static int vduse_create_dev(struct vduse_dev_config *config,
1487 			    void *config_buf, u64 api_version)
1488 {
1489 	int i, ret;
1490 	struct vduse_dev *dev;
1491 
1492 	ret = -EEXIST;
1493 	if (vduse_find_dev(config->name))
1494 		goto err;
1495 
1496 	ret = -ENOMEM;
1497 	dev = vduse_dev_create();
1498 	if (!dev)
1499 		goto err;
1500 
1501 	dev->api_version = api_version;
1502 	dev->device_features = config->features;
1503 	dev->device_id = config->device_id;
1504 	dev->vendor_id = config->vendor_id;
1505 	dev->name = kstrdup(config->name, GFP_KERNEL);
1506 	if (!dev->name)
1507 		goto err_str;
1508 
1509 	dev->domain = vduse_domain_create(VDUSE_IOVA_SIZE - 1,
1510 					  VDUSE_BOUNCE_SIZE);
1511 	if (!dev->domain)
1512 		goto err_domain;
1513 
1514 	dev->config = config_buf;
1515 	dev->config_size = config->config_size;
1516 	dev->vq_align = config->vq_align;
1517 	dev->vq_num = config->vq_num;
1518 	dev->vqs = kcalloc(dev->vq_num, sizeof(*dev->vqs), GFP_KERNEL);
1519 	if (!dev->vqs)
1520 		goto err_vqs;
1521 
1522 	for (i = 0; i < dev->vq_num; i++) {
1523 		dev->vqs[i].index = i;
1524 		INIT_WORK(&dev->vqs[i].inject, vduse_vq_irq_inject);
1525 		INIT_WORK(&dev->vqs[i].kick, vduse_vq_kick_work);
1526 		spin_lock_init(&dev->vqs[i].kick_lock);
1527 		spin_lock_init(&dev->vqs[i].irq_lock);
1528 	}
1529 
1530 	ret = idr_alloc(&vduse_idr, dev, 1, VDUSE_DEV_MAX, GFP_KERNEL);
1531 	if (ret < 0)
1532 		goto err_idr;
1533 
1534 	dev->minor = ret;
1535 	dev->msg_timeout = VDUSE_MSG_DEFAULT_TIMEOUT;
1536 	dev->dev = device_create_with_groups(vduse_class, NULL,
1537 				MKDEV(MAJOR(vduse_major), dev->minor),
1538 				dev, vduse_dev_groups, "%s", config->name);
1539 	if (IS_ERR(dev->dev)) {
1540 		ret = PTR_ERR(dev->dev);
1541 		goto err_dev;
1542 	}
1543 	__module_get(THIS_MODULE);
1544 
1545 	return 0;
1546 err_dev:
1547 	idr_remove(&vduse_idr, dev->minor);
1548 err_idr:
1549 	kfree(dev->vqs);
1550 err_vqs:
1551 	vduse_domain_destroy(dev->domain);
1552 err_domain:
1553 	kfree(dev->name);
1554 err_str:
1555 	vduse_dev_destroy(dev);
1556 err:
1557 	return ret;
1558 }
1559 
1560 static long vduse_ioctl(struct file *file, unsigned int cmd,
1561 			unsigned long arg)
1562 {
1563 	int ret;
1564 	void __user *argp = (void __user *)arg;
1565 	struct vduse_control *control = file->private_data;
1566 
1567 	mutex_lock(&vduse_lock);
1568 	switch (cmd) {
1569 	case VDUSE_GET_API_VERSION:
1570 		ret = put_user(control->api_version, (u64 __user *)argp);
1571 		break;
1572 	case VDUSE_SET_API_VERSION: {
1573 		u64 api_version;
1574 
1575 		ret = -EFAULT;
1576 		if (get_user(api_version, (u64 __user *)argp))
1577 			break;
1578 
1579 		ret = -EINVAL;
1580 		if (api_version > VDUSE_API_VERSION)
1581 			break;
1582 
1583 		ret = 0;
1584 		control->api_version = api_version;
1585 		break;
1586 	}
1587 	case VDUSE_CREATE_DEV: {
1588 		struct vduse_dev_config config;
1589 		unsigned long size = offsetof(struct vduse_dev_config, config);
1590 		void *buf;
1591 
1592 		ret = -EFAULT;
1593 		if (copy_from_user(&config, argp, size))
1594 			break;
1595 
1596 		ret = -EINVAL;
1597 		if (vduse_validate_config(&config) == false)
1598 			break;
1599 
1600 		buf = vmemdup_user(argp + size, config.config_size);
1601 		if (IS_ERR(buf)) {
1602 			ret = PTR_ERR(buf);
1603 			break;
1604 		}
1605 		config.name[VDUSE_NAME_MAX - 1] = '\0';
1606 		ret = vduse_create_dev(&config, buf, control->api_version);
1607 		if (ret)
1608 			kvfree(buf);
1609 		break;
1610 	}
1611 	case VDUSE_DESTROY_DEV: {
1612 		char name[VDUSE_NAME_MAX];
1613 
1614 		ret = -EFAULT;
1615 		if (copy_from_user(name, argp, VDUSE_NAME_MAX))
1616 			break;
1617 
1618 		name[VDUSE_NAME_MAX - 1] = '\0';
1619 		ret = vduse_destroy_dev(name);
1620 		break;
1621 	}
1622 	default:
1623 		ret = -EINVAL;
1624 		break;
1625 	}
1626 	mutex_unlock(&vduse_lock);
1627 
1628 	return ret;
1629 }
1630 
1631 static int vduse_release(struct inode *inode, struct file *file)
1632 {
1633 	struct vduse_control *control = file->private_data;
1634 
1635 	kfree(control);
1636 	return 0;
1637 }
1638 
1639 static int vduse_open(struct inode *inode, struct file *file)
1640 {
1641 	struct vduse_control *control;
1642 
1643 	control = kmalloc(sizeof(struct vduse_control), GFP_KERNEL);
1644 	if (!control)
1645 		return -ENOMEM;
1646 
1647 	control->api_version = VDUSE_API_VERSION;
1648 	file->private_data = control;
1649 
1650 	return 0;
1651 }
1652 
1653 static const struct file_operations vduse_ctrl_fops = {
1654 	.owner		= THIS_MODULE,
1655 	.open		= vduse_open,
1656 	.release	= vduse_release,
1657 	.unlocked_ioctl	= vduse_ioctl,
1658 	.compat_ioctl	= compat_ptr_ioctl,
1659 	.llseek		= noop_llseek,
1660 };
1661 
1662 static char *vduse_devnode(const struct device *dev, umode_t *mode)
1663 {
1664 	return kasprintf(GFP_KERNEL, "vduse/%s", dev_name(dev));
1665 }
1666 
1667 struct vduse_mgmt_dev {
1668 	struct vdpa_mgmt_dev mgmt_dev;
1669 	struct device dev;
1670 };
1671 
1672 static struct vduse_mgmt_dev *vduse_mgmt;
1673 
1674 static int vduse_dev_init_vdpa(struct vduse_dev *dev, const char *name)
1675 {
1676 	struct vduse_vdpa *vdev;
1677 	int ret;
1678 
1679 	if (dev->vdev)
1680 		return -EEXIST;
1681 
1682 	vdev = vdpa_alloc_device(struct vduse_vdpa, vdpa, dev->dev,
1683 				 &vduse_vdpa_config_ops, 1, 1, name, true);
1684 	if (IS_ERR(vdev))
1685 		return PTR_ERR(vdev);
1686 
1687 	dev->vdev = vdev;
1688 	vdev->dev = dev;
1689 	vdev->vdpa.dev.dma_mask = &vdev->vdpa.dev.coherent_dma_mask;
1690 	ret = dma_set_mask_and_coherent(&vdev->vdpa.dev, DMA_BIT_MASK(64));
1691 	if (ret) {
1692 		put_device(&vdev->vdpa.dev);
1693 		return ret;
1694 	}
1695 	set_dma_ops(&vdev->vdpa.dev, &vduse_dev_dma_ops);
1696 	vdev->vdpa.dma_dev = &vdev->vdpa.dev;
1697 	vdev->vdpa.mdev = &vduse_mgmt->mgmt_dev;
1698 
1699 	return 0;
1700 }
1701 
1702 static int vdpa_dev_add(struct vdpa_mgmt_dev *mdev, const char *name,
1703 			const struct vdpa_dev_set_config *config)
1704 {
1705 	struct vduse_dev *dev;
1706 	int ret;
1707 
1708 	mutex_lock(&vduse_lock);
1709 	dev = vduse_find_dev(name);
1710 	if (!dev || !vduse_dev_is_ready(dev)) {
1711 		mutex_unlock(&vduse_lock);
1712 		return -EINVAL;
1713 	}
1714 	ret = vduse_dev_init_vdpa(dev, name);
1715 	mutex_unlock(&vduse_lock);
1716 	if (ret)
1717 		return ret;
1718 
1719 	ret = _vdpa_register_device(&dev->vdev->vdpa, dev->vq_num);
1720 	if (ret) {
1721 		put_device(&dev->vdev->vdpa.dev);
1722 		return ret;
1723 	}
1724 
1725 	return 0;
1726 }
1727 
1728 static void vdpa_dev_del(struct vdpa_mgmt_dev *mdev, struct vdpa_device *dev)
1729 {
1730 	_vdpa_unregister_device(dev);
1731 }
1732 
1733 static const struct vdpa_mgmtdev_ops vdpa_dev_mgmtdev_ops = {
1734 	.dev_add = vdpa_dev_add,
1735 	.dev_del = vdpa_dev_del,
1736 };
1737 
1738 static struct virtio_device_id id_table[] = {
1739 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1740 	{ 0 },
1741 };
1742 
1743 static void vduse_mgmtdev_release(struct device *dev)
1744 {
1745 	struct vduse_mgmt_dev *mgmt_dev;
1746 
1747 	mgmt_dev = container_of(dev, struct vduse_mgmt_dev, dev);
1748 	kfree(mgmt_dev);
1749 }
1750 
1751 static int vduse_mgmtdev_init(void)
1752 {
1753 	int ret;
1754 
1755 	vduse_mgmt = kzalloc(sizeof(*vduse_mgmt), GFP_KERNEL);
1756 	if (!vduse_mgmt)
1757 		return -ENOMEM;
1758 
1759 	ret = dev_set_name(&vduse_mgmt->dev, "vduse");
1760 	if (ret) {
1761 		kfree(vduse_mgmt);
1762 		return ret;
1763 	}
1764 
1765 	vduse_mgmt->dev.release = vduse_mgmtdev_release;
1766 
1767 	ret = device_register(&vduse_mgmt->dev);
1768 	if (ret)
1769 		goto dev_reg_err;
1770 
1771 	vduse_mgmt->mgmt_dev.id_table = id_table;
1772 	vduse_mgmt->mgmt_dev.ops = &vdpa_dev_mgmtdev_ops;
1773 	vduse_mgmt->mgmt_dev.device = &vduse_mgmt->dev;
1774 	ret = vdpa_mgmtdev_register(&vduse_mgmt->mgmt_dev);
1775 	if (ret)
1776 		device_unregister(&vduse_mgmt->dev);
1777 
1778 	return ret;
1779 
1780 dev_reg_err:
1781 	put_device(&vduse_mgmt->dev);
1782 	return ret;
1783 }
1784 
1785 static void vduse_mgmtdev_exit(void)
1786 {
1787 	vdpa_mgmtdev_unregister(&vduse_mgmt->mgmt_dev);
1788 	device_unregister(&vduse_mgmt->dev);
1789 }
1790 
1791 static int vduse_init(void)
1792 {
1793 	int ret;
1794 	struct device *dev;
1795 
1796 	vduse_class = class_create(THIS_MODULE, "vduse");
1797 	if (IS_ERR(vduse_class))
1798 		return PTR_ERR(vduse_class);
1799 
1800 	vduse_class->devnode = vduse_devnode;
1801 
1802 	ret = alloc_chrdev_region(&vduse_major, 0, VDUSE_DEV_MAX, "vduse");
1803 	if (ret)
1804 		goto err_chardev_region;
1805 
1806 	/* /dev/vduse/control */
1807 	cdev_init(&vduse_ctrl_cdev, &vduse_ctrl_fops);
1808 	vduse_ctrl_cdev.owner = THIS_MODULE;
1809 	ret = cdev_add(&vduse_ctrl_cdev, vduse_major, 1);
1810 	if (ret)
1811 		goto err_ctrl_cdev;
1812 
1813 	dev = device_create(vduse_class, NULL, vduse_major, NULL, "control");
1814 	if (IS_ERR(dev)) {
1815 		ret = PTR_ERR(dev);
1816 		goto err_device;
1817 	}
1818 
1819 	/* /dev/vduse/$DEVICE */
1820 	cdev_init(&vduse_cdev, &vduse_dev_fops);
1821 	vduse_cdev.owner = THIS_MODULE;
1822 	ret = cdev_add(&vduse_cdev, MKDEV(MAJOR(vduse_major), 1),
1823 		       VDUSE_DEV_MAX - 1);
1824 	if (ret)
1825 		goto err_cdev;
1826 
1827 	vduse_irq_wq = alloc_workqueue("vduse-irq",
1828 				WQ_HIGHPRI | WQ_SYSFS | WQ_UNBOUND, 0);
1829 	if (!vduse_irq_wq) {
1830 		ret = -ENOMEM;
1831 		goto err_wq;
1832 	}
1833 
1834 	ret = vduse_domain_init();
1835 	if (ret)
1836 		goto err_domain;
1837 
1838 	ret = vduse_mgmtdev_init();
1839 	if (ret)
1840 		goto err_mgmtdev;
1841 
1842 	return 0;
1843 err_mgmtdev:
1844 	vduse_domain_exit();
1845 err_domain:
1846 	destroy_workqueue(vduse_irq_wq);
1847 err_wq:
1848 	cdev_del(&vduse_cdev);
1849 err_cdev:
1850 	device_destroy(vduse_class, vduse_major);
1851 err_device:
1852 	cdev_del(&vduse_ctrl_cdev);
1853 err_ctrl_cdev:
1854 	unregister_chrdev_region(vduse_major, VDUSE_DEV_MAX);
1855 err_chardev_region:
1856 	class_destroy(vduse_class);
1857 	return ret;
1858 }
1859 module_init(vduse_init);
1860 
1861 static void vduse_exit(void)
1862 {
1863 	vduse_mgmtdev_exit();
1864 	vduse_domain_exit();
1865 	destroy_workqueue(vduse_irq_wq);
1866 	cdev_del(&vduse_cdev);
1867 	device_destroy(vduse_class, vduse_major);
1868 	cdev_del(&vduse_ctrl_cdev);
1869 	unregister_chrdev_region(vduse_major, VDUSE_DEV_MAX);
1870 	class_destroy(vduse_class);
1871 }
1872 module_exit(vduse_exit);
1873 
1874 MODULE_LICENSE(DRV_LICENSE);
1875 MODULE_AUTHOR(DRV_AUTHOR);
1876 MODULE_DESCRIPTION(DRV_DESC);
1877