xref: /openbmc/linux/drivers/vhost/vhost.c (revision 568b9de4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (C) 2009 Red Hat, Inc.
3  * Copyright (C) 2006 Rusty Russell IBM Corporation
4  *
5  * Author: Michael S. Tsirkin <mst@redhat.com>
6  *
7  * Inspiration, some code, and most witty comments come from
8  * Documentation/virtual/lguest/lguest.c, by Rusty Russell
9  *
10  * Generic code for virtio server in host kernel.
11  */
12 
13 #include <linux/eventfd.h>
14 #include <linux/vhost.h>
15 #include <linux/uio.h>
16 #include <linux/mm.h>
17 #include <linux/mmu_context.h>
18 #include <linux/miscdevice.h>
19 #include <linux/mutex.h>
20 #include <linux/poll.h>
21 #include <linux/file.h>
22 #include <linux/highmem.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/kthread.h>
26 #include <linux/cgroup.h>
27 #include <linux/module.h>
28 #include <linux/sort.h>
29 #include <linux/sched/mm.h>
30 #include <linux/sched/signal.h>
31 #include <linux/interval_tree_generic.h>
32 #include <linux/nospec.h>
33 
34 #include "vhost.h"
35 
36 static ushort max_mem_regions = 64;
37 module_param(max_mem_regions, ushort, 0444);
38 MODULE_PARM_DESC(max_mem_regions,
39 	"Maximum number of memory regions in memory map. (default: 64)");
40 static int max_iotlb_entries = 2048;
41 module_param(max_iotlb_entries, int, 0444);
42 MODULE_PARM_DESC(max_iotlb_entries,
43 	"Maximum number of iotlb entries. (default: 2048)");
44 
45 enum {
46 	VHOST_MEMORY_F_LOG = 0x1,
47 };
48 
49 #define vhost_used_event(vq) ((__virtio16 __user *)&vq->avail->ring[vq->num])
50 #define vhost_avail_event(vq) ((__virtio16 __user *)&vq->used->ring[vq->num])
51 
52 INTERVAL_TREE_DEFINE(struct vhost_umem_node,
53 		     rb, __u64, __subtree_last,
54 		     START, LAST, static inline, vhost_umem_interval_tree);
55 
56 #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
57 static void vhost_disable_cross_endian(struct vhost_virtqueue *vq)
58 {
59 	vq->user_be = !virtio_legacy_is_little_endian();
60 }
61 
62 static void vhost_enable_cross_endian_big(struct vhost_virtqueue *vq)
63 {
64 	vq->user_be = true;
65 }
66 
67 static void vhost_enable_cross_endian_little(struct vhost_virtqueue *vq)
68 {
69 	vq->user_be = false;
70 }
71 
72 static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp)
73 {
74 	struct vhost_vring_state s;
75 
76 	if (vq->private_data)
77 		return -EBUSY;
78 
79 	if (copy_from_user(&s, argp, sizeof(s)))
80 		return -EFAULT;
81 
82 	if (s.num != VHOST_VRING_LITTLE_ENDIAN &&
83 	    s.num != VHOST_VRING_BIG_ENDIAN)
84 		return -EINVAL;
85 
86 	if (s.num == VHOST_VRING_BIG_ENDIAN)
87 		vhost_enable_cross_endian_big(vq);
88 	else
89 		vhost_enable_cross_endian_little(vq);
90 
91 	return 0;
92 }
93 
94 static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx,
95 				   int __user *argp)
96 {
97 	struct vhost_vring_state s = {
98 		.index = idx,
99 		.num = vq->user_be
100 	};
101 
102 	if (copy_to_user(argp, &s, sizeof(s)))
103 		return -EFAULT;
104 
105 	return 0;
106 }
107 
108 static void vhost_init_is_le(struct vhost_virtqueue *vq)
109 {
110 	/* Note for legacy virtio: user_be is initialized at reset time
111 	 * according to the host endianness. If userspace does not set an
112 	 * explicit endianness, the default behavior is native endian, as
113 	 * expected by legacy virtio.
114 	 */
115 	vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) || !vq->user_be;
116 }
117 #else
118 static void vhost_disable_cross_endian(struct vhost_virtqueue *vq)
119 {
120 }
121 
122 static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp)
123 {
124 	return -ENOIOCTLCMD;
125 }
126 
127 static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx,
128 				   int __user *argp)
129 {
130 	return -ENOIOCTLCMD;
131 }
132 
133 static void vhost_init_is_le(struct vhost_virtqueue *vq)
134 {
135 	vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1)
136 		|| virtio_legacy_is_little_endian();
137 }
138 #endif /* CONFIG_VHOST_CROSS_ENDIAN_LEGACY */
139 
140 static void vhost_reset_is_le(struct vhost_virtqueue *vq)
141 {
142 	vhost_init_is_le(vq);
143 }
144 
145 struct vhost_flush_struct {
146 	struct vhost_work work;
147 	struct completion wait_event;
148 };
149 
150 static void vhost_flush_work(struct vhost_work *work)
151 {
152 	struct vhost_flush_struct *s;
153 
154 	s = container_of(work, struct vhost_flush_struct, work);
155 	complete(&s->wait_event);
156 }
157 
158 static void vhost_poll_func(struct file *file, wait_queue_head_t *wqh,
159 			    poll_table *pt)
160 {
161 	struct vhost_poll *poll;
162 
163 	poll = container_of(pt, struct vhost_poll, table);
164 	poll->wqh = wqh;
165 	add_wait_queue(wqh, &poll->wait);
166 }
167 
168 static int vhost_poll_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync,
169 			     void *key)
170 {
171 	struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait);
172 
173 	if (!(key_to_poll(key) & poll->mask))
174 		return 0;
175 
176 	vhost_poll_queue(poll);
177 	return 0;
178 }
179 
180 void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn)
181 {
182 	clear_bit(VHOST_WORK_QUEUED, &work->flags);
183 	work->fn = fn;
184 }
185 EXPORT_SYMBOL_GPL(vhost_work_init);
186 
187 /* Init poll structure */
188 void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
189 		     __poll_t mask, struct vhost_dev *dev)
190 {
191 	init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup);
192 	init_poll_funcptr(&poll->table, vhost_poll_func);
193 	poll->mask = mask;
194 	poll->dev = dev;
195 	poll->wqh = NULL;
196 
197 	vhost_work_init(&poll->work, fn);
198 }
199 EXPORT_SYMBOL_GPL(vhost_poll_init);
200 
201 /* Start polling a file. We add ourselves to file's wait queue. The caller must
202  * keep a reference to a file until after vhost_poll_stop is called. */
203 int vhost_poll_start(struct vhost_poll *poll, struct file *file)
204 {
205 	__poll_t mask;
206 	int ret = 0;
207 
208 	if (poll->wqh)
209 		return 0;
210 
211 	mask = vfs_poll(file, &poll->table);
212 	if (mask)
213 		vhost_poll_wakeup(&poll->wait, 0, 0, poll_to_key(mask));
214 	if (mask & EPOLLERR) {
215 		vhost_poll_stop(poll);
216 		ret = -EINVAL;
217 	}
218 
219 	return ret;
220 }
221 EXPORT_SYMBOL_GPL(vhost_poll_start);
222 
223 /* Stop polling a file. After this function returns, it becomes safe to drop the
224  * file reference. You must also flush afterwards. */
225 void vhost_poll_stop(struct vhost_poll *poll)
226 {
227 	if (poll->wqh) {
228 		remove_wait_queue(poll->wqh, &poll->wait);
229 		poll->wqh = NULL;
230 	}
231 }
232 EXPORT_SYMBOL_GPL(vhost_poll_stop);
233 
234 void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work)
235 {
236 	struct vhost_flush_struct flush;
237 
238 	if (dev->worker) {
239 		init_completion(&flush.wait_event);
240 		vhost_work_init(&flush.work, vhost_flush_work);
241 
242 		vhost_work_queue(dev, &flush.work);
243 		wait_for_completion(&flush.wait_event);
244 	}
245 }
246 EXPORT_SYMBOL_GPL(vhost_work_flush);
247 
248 /* Flush any work that has been scheduled. When calling this, don't hold any
249  * locks that are also used by the callback. */
250 void vhost_poll_flush(struct vhost_poll *poll)
251 {
252 	vhost_work_flush(poll->dev, &poll->work);
253 }
254 EXPORT_SYMBOL_GPL(vhost_poll_flush);
255 
256 void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work)
257 {
258 	if (!dev->worker)
259 		return;
260 
261 	if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) {
262 		/* We can only add the work to the list after we're
263 		 * sure it was not in the list.
264 		 * test_and_set_bit() implies a memory barrier.
265 		 */
266 		llist_add(&work->node, &dev->work_list);
267 		wake_up_process(dev->worker);
268 	}
269 }
270 EXPORT_SYMBOL_GPL(vhost_work_queue);
271 
272 /* A lockless hint for busy polling code to exit the loop */
273 bool vhost_has_work(struct vhost_dev *dev)
274 {
275 	return !llist_empty(&dev->work_list);
276 }
277 EXPORT_SYMBOL_GPL(vhost_has_work);
278 
279 void vhost_poll_queue(struct vhost_poll *poll)
280 {
281 	vhost_work_queue(poll->dev, &poll->work);
282 }
283 EXPORT_SYMBOL_GPL(vhost_poll_queue);
284 
285 static void __vhost_vq_meta_reset(struct vhost_virtqueue *vq)
286 {
287 	int j;
288 
289 	for (j = 0; j < VHOST_NUM_ADDRS; j++)
290 		vq->meta_iotlb[j] = NULL;
291 }
292 
293 static void vhost_vq_meta_reset(struct vhost_dev *d)
294 {
295 	int i;
296 
297 	for (i = 0; i < d->nvqs; ++i)
298 		__vhost_vq_meta_reset(d->vqs[i]);
299 }
300 
301 #if VHOST_ARCH_CAN_ACCEL_UACCESS
302 static void vhost_map_unprefetch(struct vhost_map *map)
303 {
304 	kfree(map->pages);
305 	map->pages = NULL;
306 	map->npages = 0;
307 	map->addr = NULL;
308 }
309 
310 static void vhost_uninit_vq_maps(struct vhost_virtqueue *vq)
311 {
312 	struct vhost_map *map[VHOST_NUM_ADDRS];
313 	int i;
314 
315 	spin_lock(&vq->mmu_lock);
316 	for (i = 0; i < VHOST_NUM_ADDRS; i++) {
317 		map[i] = rcu_dereference_protected(vq->maps[i],
318 				  lockdep_is_held(&vq->mmu_lock));
319 		if (map[i])
320 			rcu_assign_pointer(vq->maps[i], NULL);
321 	}
322 	spin_unlock(&vq->mmu_lock);
323 
324 	synchronize_rcu();
325 
326 	for (i = 0; i < VHOST_NUM_ADDRS; i++)
327 		if (map[i])
328 			vhost_map_unprefetch(map[i]);
329 
330 }
331 
332 static void vhost_reset_vq_maps(struct vhost_virtqueue *vq)
333 {
334 	int i;
335 
336 	vhost_uninit_vq_maps(vq);
337 	for (i = 0; i < VHOST_NUM_ADDRS; i++)
338 		vq->uaddrs[i].size = 0;
339 }
340 
341 static bool vhost_map_range_overlap(struct vhost_uaddr *uaddr,
342 				     unsigned long start,
343 				     unsigned long end)
344 {
345 	if (unlikely(!uaddr->size))
346 		return false;
347 
348 	return !(end < uaddr->uaddr || start > uaddr->uaddr - 1 + uaddr->size);
349 }
350 
351 static void vhost_invalidate_vq_start(struct vhost_virtqueue *vq,
352 				      int index,
353 				      unsigned long start,
354 				      unsigned long end)
355 {
356 	struct vhost_uaddr *uaddr = &vq->uaddrs[index];
357 	struct vhost_map *map;
358 	int i;
359 
360 	if (!vhost_map_range_overlap(uaddr, start, end))
361 		return;
362 
363 	spin_lock(&vq->mmu_lock);
364 	++vq->invalidate_count;
365 
366 	map = rcu_dereference_protected(vq->maps[index],
367 					lockdep_is_held(&vq->mmu_lock));
368 	if (map) {
369 		if (uaddr->write) {
370 			for (i = 0; i < map->npages; i++)
371 				set_page_dirty(map->pages[i]);
372 		}
373 		rcu_assign_pointer(vq->maps[index], NULL);
374 	}
375 	spin_unlock(&vq->mmu_lock);
376 
377 	if (map) {
378 		synchronize_rcu();
379 		vhost_map_unprefetch(map);
380 	}
381 }
382 
383 static void vhost_invalidate_vq_end(struct vhost_virtqueue *vq,
384 				    int index,
385 				    unsigned long start,
386 				    unsigned long end)
387 {
388 	if (!vhost_map_range_overlap(&vq->uaddrs[index], start, end))
389 		return;
390 
391 	spin_lock(&vq->mmu_lock);
392 	--vq->invalidate_count;
393 	spin_unlock(&vq->mmu_lock);
394 }
395 
396 static int vhost_invalidate_range_start(struct mmu_notifier *mn,
397 					const struct mmu_notifier_range *range)
398 {
399 	struct vhost_dev *dev = container_of(mn, struct vhost_dev,
400 					     mmu_notifier);
401 	int i, j;
402 
403 	if (!mmu_notifier_range_blockable(range))
404 		return -EAGAIN;
405 
406 	for (i = 0; i < dev->nvqs; i++) {
407 		struct vhost_virtqueue *vq = dev->vqs[i];
408 
409 		for (j = 0; j < VHOST_NUM_ADDRS; j++)
410 			vhost_invalidate_vq_start(vq, j,
411 						  range->start,
412 						  range->end);
413 	}
414 
415 	return 0;
416 }
417 
418 static void vhost_invalidate_range_end(struct mmu_notifier *mn,
419 				       const struct mmu_notifier_range *range)
420 {
421 	struct vhost_dev *dev = container_of(mn, struct vhost_dev,
422 					     mmu_notifier);
423 	int i, j;
424 
425 	for (i = 0; i < dev->nvqs; i++) {
426 		struct vhost_virtqueue *vq = dev->vqs[i];
427 
428 		for (j = 0; j < VHOST_NUM_ADDRS; j++)
429 			vhost_invalidate_vq_end(vq, j,
430 						range->start,
431 						range->end);
432 	}
433 }
434 
435 static const struct mmu_notifier_ops vhost_mmu_notifier_ops = {
436 	.invalidate_range_start = vhost_invalidate_range_start,
437 	.invalidate_range_end = vhost_invalidate_range_end,
438 };
439 
440 static void vhost_init_maps(struct vhost_dev *dev)
441 {
442 	struct vhost_virtqueue *vq;
443 	int i, j;
444 
445 	dev->mmu_notifier.ops = &vhost_mmu_notifier_ops;
446 
447 	for (i = 0; i < dev->nvqs; ++i) {
448 		vq = dev->vqs[i];
449 		for (j = 0; j < VHOST_NUM_ADDRS; j++)
450 			RCU_INIT_POINTER(vq->maps[j], NULL);
451 	}
452 }
453 #endif
454 
455 static void vhost_vq_reset(struct vhost_dev *dev,
456 			   struct vhost_virtqueue *vq)
457 {
458 	vq->num = 1;
459 	vq->desc = NULL;
460 	vq->avail = NULL;
461 	vq->used = NULL;
462 	vq->last_avail_idx = 0;
463 	vq->avail_idx = 0;
464 	vq->last_used_idx = 0;
465 	vq->signalled_used = 0;
466 	vq->signalled_used_valid = false;
467 	vq->used_flags = 0;
468 	vq->log_used = false;
469 	vq->log_addr = -1ull;
470 	vq->private_data = NULL;
471 	vq->acked_features = 0;
472 	vq->acked_backend_features = 0;
473 	vq->log_base = NULL;
474 	vq->error_ctx = NULL;
475 	vq->kick = NULL;
476 	vq->call_ctx = NULL;
477 	vq->log_ctx = NULL;
478 	vhost_reset_is_le(vq);
479 	vhost_disable_cross_endian(vq);
480 	vq->busyloop_timeout = 0;
481 	vq->umem = NULL;
482 	vq->iotlb = NULL;
483 	vq->invalidate_count = 0;
484 	__vhost_vq_meta_reset(vq);
485 #if VHOST_ARCH_CAN_ACCEL_UACCESS
486 	vhost_reset_vq_maps(vq);
487 #endif
488 }
489 
490 static int vhost_worker(void *data)
491 {
492 	struct vhost_dev *dev = data;
493 	struct vhost_work *work, *work_next;
494 	struct llist_node *node;
495 	mm_segment_t oldfs = get_fs();
496 
497 	set_fs(USER_DS);
498 	use_mm(dev->mm);
499 
500 	for (;;) {
501 		/* mb paired w/ kthread_stop */
502 		set_current_state(TASK_INTERRUPTIBLE);
503 
504 		if (kthread_should_stop()) {
505 			__set_current_state(TASK_RUNNING);
506 			break;
507 		}
508 
509 		node = llist_del_all(&dev->work_list);
510 		if (!node)
511 			schedule();
512 
513 		node = llist_reverse_order(node);
514 		/* make sure flag is seen after deletion */
515 		smp_wmb();
516 		llist_for_each_entry_safe(work, work_next, node, node) {
517 			clear_bit(VHOST_WORK_QUEUED, &work->flags);
518 			__set_current_state(TASK_RUNNING);
519 			work->fn(work);
520 			if (need_resched())
521 				schedule();
522 		}
523 	}
524 	unuse_mm(dev->mm);
525 	set_fs(oldfs);
526 	return 0;
527 }
528 
529 static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq)
530 {
531 	kfree(vq->indirect);
532 	vq->indirect = NULL;
533 	kfree(vq->log);
534 	vq->log = NULL;
535 	kfree(vq->heads);
536 	vq->heads = NULL;
537 }
538 
539 /* Helper to allocate iovec buffers for all vqs. */
540 static long vhost_dev_alloc_iovecs(struct vhost_dev *dev)
541 {
542 	struct vhost_virtqueue *vq;
543 	int i;
544 
545 	for (i = 0; i < dev->nvqs; ++i) {
546 		vq = dev->vqs[i];
547 		vq->indirect = kmalloc_array(UIO_MAXIOV,
548 					     sizeof(*vq->indirect),
549 					     GFP_KERNEL);
550 		vq->log = kmalloc_array(dev->iov_limit, sizeof(*vq->log),
551 					GFP_KERNEL);
552 		vq->heads = kmalloc_array(dev->iov_limit, sizeof(*vq->heads),
553 					  GFP_KERNEL);
554 		if (!vq->indirect || !vq->log || !vq->heads)
555 			goto err_nomem;
556 	}
557 	return 0;
558 
559 err_nomem:
560 	for (; i >= 0; --i)
561 		vhost_vq_free_iovecs(dev->vqs[i]);
562 	return -ENOMEM;
563 }
564 
565 static void vhost_dev_free_iovecs(struct vhost_dev *dev)
566 {
567 	int i;
568 
569 	for (i = 0; i < dev->nvqs; ++i)
570 		vhost_vq_free_iovecs(dev->vqs[i]);
571 }
572 
573 bool vhost_exceeds_weight(struct vhost_virtqueue *vq,
574 			  int pkts, int total_len)
575 {
576 	struct vhost_dev *dev = vq->dev;
577 
578 	if ((dev->byte_weight && total_len >= dev->byte_weight) ||
579 	    pkts >= dev->weight) {
580 		vhost_poll_queue(&vq->poll);
581 		return true;
582 	}
583 
584 	return false;
585 }
586 EXPORT_SYMBOL_GPL(vhost_exceeds_weight);
587 
588 static size_t vhost_get_avail_size(struct vhost_virtqueue *vq,
589 				   unsigned int num)
590 {
591 	size_t event __maybe_unused =
592 	       vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
593 
594 	return sizeof(*vq->avail) +
595 	       sizeof(*vq->avail->ring) * num + event;
596 }
597 
598 static size_t vhost_get_used_size(struct vhost_virtqueue *vq,
599 				  unsigned int num)
600 {
601 	size_t event __maybe_unused =
602 	       vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
603 
604 	return sizeof(*vq->used) +
605 	       sizeof(*vq->used->ring) * num + event;
606 }
607 
608 static size_t vhost_get_desc_size(struct vhost_virtqueue *vq,
609 				  unsigned int num)
610 {
611 	return sizeof(*vq->desc) * num;
612 }
613 
614 void vhost_dev_init(struct vhost_dev *dev,
615 		    struct vhost_virtqueue **vqs, int nvqs,
616 		    int iov_limit, int weight, int byte_weight)
617 {
618 	struct vhost_virtqueue *vq;
619 	int i;
620 
621 	dev->vqs = vqs;
622 	dev->nvqs = nvqs;
623 	mutex_init(&dev->mutex);
624 	dev->log_ctx = NULL;
625 	dev->umem = NULL;
626 	dev->iotlb = NULL;
627 	dev->mm = NULL;
628 	dev->worker = NULL;
629 	dev->iov_limit = iov_limit;
630 	dev->weight = weight;
631 	dev->byte_weight = byte_weight;
632 	init_llist_head(&dev->work_list);
633 	init_waitqueue_head(&dev->wait);
634 	INIT_LIST_HEAD(&dev->read_list);
635 	INIT_LIST_HEAD(&dev->pending_list);
636 	spin_lock_init(&dev->iotlb_lock);
637 #if VHOST_ARCH_CAN_ACCEL_UACCESS
638 	vhost_init_maps(dev);
639 #endif
640 
641 	for (i = 0; i < dev->nvqs; ++i) {
642 		vq = dev->vqs[i];
643 		vq->log = NULL;
644 		vq->indirect = NULL;
645 		vq->heads = NULL;
646 		vq->dev = dev;
647 		mutex_init(&vq->mutex);
648 		spin_lock_init(&vq->mmu_lock);
649 		vhost_vq_reset(dev, vq);
650 		if (vq->handle_kick)
651 			vhost_poll_init(&vq->poll, vq->handle_kick,
652 					EPOLLIN, dev);
653 	}
654 }
655 EXPORT_SYMBOL_GPL(vhost_dev_init);
656 
657 /* Caller should have device mutex */
658 long vhost_dev_check_owner(struct vhost_dev *dev)
659 {
660 	/* Are you the owner? If not, I don't think you mean to do that */
661 	return dev->mm == current->mm ? 0 : -EPERM;
662 }
663 EXPORT_SYMBOL_GPL(vhost_dev_check_owner);
664 
665 struct vhost_attach_cgroups_struct {
666 	struct vhost_work work;
667 	struct task_struct *owner;
668 	int ret;
669 };
670 
671 static void vhost_attach_cgroups_work(struct vhost_work *work)
672 {
673 	struct vhost_attach_cgroups_struct *s;
674 
675 	s = container_of(work, struct vhost_attach_cgroups_struct, work);
676 	s->ret = cgroup_attach_task_all(s->owner, current);
677 }
678 
679 static int vhost_attach_cgroups(struct vhost_dev *dev)
680 {
681 	struct vhost_attach_cgroups_struct attach;
682 
683 	attach.owner = current;
684 	vhost_work_init(&attach.work, vhost_attach_cgroups_work);
685 	vhost_work_queue(dev, &attach.work);
686 	vhost_work_flush(dev, &attach.work);
687 	return attach.ret;
688 }
689 
690 /* Caller should have device mutex */
691 bool vhost_dev_has_owner(struct vhost_dev *dev)
692 {
693 	return dev->mm;
694 }
695 EXPORT_SYMBOL_GPL(vhost_dev_has_owner);
696 
697 /* Caller should have device mutex */
698 long vhost_dev_set_owner(struct vhost_dev *dev)
699 {
700 	struct task_struct *worker;
701 	int err;
702 
703 	/* Is there an owner already? */
704 	if (vhost_dev_has_owner(dev)) {
705 		err = -EBUSY;
706 		goto err_mm;
707 	}
708 
709 	/* No owner, become one */
710 	dev->mm = get_task_mm(current);
711 	worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid);
712 	if (IS_ERR(worker)) {
713 		err = PTR_ERR(worker);
714 		goto err_worker;
715 	}
716 
717 	dev->worker = worker;
718 	wake_up_process(worker);	/* avoid contributing to loadavg */
719 
720 	err = vhost_attach_cgroups(dev);
721 	if (err)
722 		goto err_cgroup;
723 
724 	err = vhost_dev_alloc_iovecs(dev);
725 	if (err)
726 		goto err_cgroup;
727 
728 #if VHOST_ARCH_CAN_ACCEL_UACCESS
729 	err = mmu_notifier_register(&dev->mmu_notifier, dev->mm);
730 	if (err)
731 		goto err_mmu_notifier;
732 #endif
733 
734 	return 0;
735 
736 #if VHOST_ARCH_CAN_ACCEL_UACCESS
737 err_mmu_notifier:
738 	vhost_dev_free_iovecs(dev);
739 #endif
740 err_cgroup:
741 	kthread_stop(worker);
742 	dev->worker = NULL;
743 err_worker:
744 	if (dev->mm)
745 		mmput(dev->mm);
746 	dev->mm = NULL;
747 err_mm:
748 	return err;
749 }
750 EXPORT_SYMBOL_GPL(vhost_dev_set_owner);
751 
752 struct vhost_umem *vhost_dev_reset_owner_prepare(void)
753 {
754 	return kvzalloc(sizeof(struct vhost_umem), GFP_KERNEL);
755 }
756 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare);
757 
758 /* Caller should have device mutex */
759 void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_umem *umem)
760 {
761 	int i;
762 
763 	vhost_dev_cleanup(dev);
764 
765 	/* Restore memory to default empty mapping. */
766 	INIT_LIST_HEAD(&umem->umem_list);
767 	dev->umem = umem;
768 	/* We don't need VQ locks below since vhost_dev_cleanup makes sure
769 	 * VQs aren't running.
770 	 */
771 	for (i = 0; i < dev->nvqs; ++i)
772 		dev->vqs[i]->umem = umem;
773 }
774 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner);
775 
776 void vhost_dev_stop(struct vhost_dev *dev)
777 {
778 	int i;
779 
780 	for (i = 0; i < dev->nvqs; ++i) {
781 		if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) {
782 			vhost_poll_stop(&dev->vqs[i]->poll);
783 			vhost_poll_flush(&dev->vqs[i]->poll);
784 		}
785 	}
786 }
787 EXPORT_SYMBOL_GPL(vhost_dev_stop);
788 
789 static void vhost_umem_free(struct vhost_umem *umem,
790 			    struct vhost_umem_node *node)
791 {
792 	vhost_umem_interval_tree_remove(node, &umem->umem_tree);
793 	list_del(&node->link);
794 	kfree(node);
795 	umem->numem--;
796 }
797 
798 static void vhost_umem_clean(struct vhost_umem *umem)
799 {
800 	struct vhost_umem_node *node, *tmp;
801 
802 	if (!umem)
803 		return;
804 
805 	list_for_each_entry_safe(node, tmp, &umem->umem_list, link)
806 		vhost_umem_free(umem, node);
807 
808 	kvfree(umem);
809 }
810 
811 static void vhost_clear_msg(struct vhost_dev *dev)
812 {
813 	struct vhost_msg_node *node, *n;
814 
815 	spin_lock(&dev->iotlb_lock);
816 
817 	list_for_each_entry_safe(node, n, &dev->read_list, node) {
818 		list_del(&node->node);
819 		kfree(node);
820 	}
821 
822 	list_for_each_entry_safe(node, n, &dev->pending_list, node) {
823 		list_del(&node->node);
824 		kfree(node);
825 	}
826 
827 	spin_unlock(&dev->iotlb_lock);
828 }
829 
830 #if VHOST_ARCH_CAN_ACCEL_UACCESS
831 static void vhost_setup_uaddr(struct vhost_virtqueue *vq,
832 			      int index, unsigned long uaddr,
833 			      size_t size, bool write)
834 {
835 	struct vhost_uaddr *addr = &vq->uaddrs[index];
836 
837 	addr->uaddr = uaddr;
838 	addr->size = size;
839 	addr->write = write;
840 }
841 
842 static void vhost_setup_vq_uaddr(struct vhost_virtqueue *vq)
843 {
844 	vhost_setup_uaddr(vq, VHOST_ADDR_DESC,
845 			  (unsigned long)vq->desc,
846 			  vhost_get_desc_size(vq, vq->num),
847 			  false);
848 	vhost_setup_uaddr(vq, VHOST_ADDR_AVAIL,
849 			  (unsigned long)vq->avail,
850 			  vhost_get_avail_size(vq, vq->num),
851 			  false);
852 	vhost_setup_uaddr(vq, VHOST_ADDR_USED,
853 			  (unsigned long)vq->used,
854 			  vhost_get_used_size(vq, vq->num),
855 			  true);
856 }
857 
858 static int vhost_map_prefetch(struct vhost_virtqueue *vq,
859 			       int index)
860 {
861 	struct vhost_map *map;
862 	struct vhost_uaddr *uaddr = &vq->uaddrs[index];
863 	struct page **pages;
864 	int npages = DIV_ROUND_UP(uaddr->size, PAGE_SIZE);
865 	int npinned;
866 	void *vaddr, *v;
867 	int err;
868 	int i;
869 
870 	spin_lock(&vq->mmu_lock);
871 
872 	err = -EFAULT;
873 	if (vq->invalidate_count)
874 		goto err;
875 
876 	err = -ENOMEM;
877 	map = kmalloc(sizeof(*map), GFP_ATOMIC);
878 	if (!map)
879 		goto err;
880 
881 	pages = kmalloc_array(npages, sizeof(struct page *), GFP_ATOMIC);
882 	if (!pages)
883 		goto err_pages;
884 
885 	err = EFAULT;
886 	npinned = __get_user_pages_fast(uaddr->uaddr, npages,
887 					uaddr->write, pages);
888 	if (npinned > 0)
889 		release_pages(pages, npinned);
890 	if (npinned != npages)
891 		goto err_gup;
892 
893 	for (i = 0; i < npinned; i++)
894 		if (PageHighMem(pages[i]))
895 			goto err_gup;
896 
897 	vaddr = v = page_address(pages[0]);
898 
899 	/* For simplicity, fallback to userspace address if VA is not
900 	 * contigious.
901 	 */
902 	for (i = 1; i < npinned; i++) {
903 		v += PAGE_SIZE;
904 		if (v != page_address(pages[i]))
905 			goto err_gup;
906 	}
907 
908 	map->addr = vaddr + (uaddr->uaddr & (PAGE_SIZE - 1));
909 	map->npages = npages;
910 	map->pages = pages;
911 
912 	rcu_assign_pointer(vq->maps[index], map);
913 	/* No need for a synchronize_rcu(). This function should be
914 	 * called by dev->worker so we are serialized with all
915 	 * readers.
916 	 */
917 	spin_unlock(&vq->mmu_lock);
918 
919 	return 0;
920 
921 err_gup:
922 	kfree(pages);
923 err_pages:
924 	kfree(map);
925 err:
926 	spin_unlock(&vq->mmu_lock);
927 	return err;
928 }
929 #endif
930 
931 void vhost_dev_cleanup(struct vhost_dev *dev)
932 {
933 	int i;
934 
935 	for (i = 0; i < dev->nvqs; ++i) {
936 		if (dev->vqs[i]->error_ctx)
937 			eventfd_ctx_put(dev->vqs[i]->error_ctx);
938 		if (dev->vqs[i]->kick)
939 			fput(dev->vqs[i]->kick);
940 		if (dev->vqs[i]->call_ctx)
941 			eventfd_ctx_put(dev->vqs[i]->call_ctx);
942 		vhost_vq_reset(dev, dev->vqs[i]);
943 	}
944 	vhost_dev_free_iovecs(dev);
945 	if (dev->log_ctx)
946 		eventfd_ctx_put(dev->log_ctx);
947 	dev->log_ctx = NULL;
948 	/* No one will access memory at this point */
949 	vhost_umem_clean(dev->umem);
950 	dev->umem = NULL;
951 	vhost_umem_clean(dev->iotlb);
952 	dev->iotlb = NULL;
953 	vhost_clear_msg(dev);
954 	wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
955 	WARN_ON(!llist_empty(&dev->work_list));
956 	if (dev->worker) {
957 		kthread_stop(dev->worker);
958 		dev->worker = NULL;
959 	}
960 	if (dev->mm) {
961 #if VHOST_ARCH_CAN_ACCEL_UACCESS
962 		mmu_notifier_unregister(&dev->mmu_notifier, dev->mm);
963 #endif
964 		mmput(dev->mm);
965 	}
966 #if VHOST_ARCH_CAN_ACCEL_UACCESS
967 	for (i = 0; i < dev->nvqs; i++)
968 		vhost_uninit_vq_maps(dev->vqs[i]);
969 #endif
970 	dev->mm = NULL;
971 }
972 EXPORT_SYMBOL_GPL(vhost_dev_cleanup);
973 
974 static bool log_access_ok(void __user *log_base, u64 addr, unsigned long sz)
975 {
976 	u64 a = addr / VHOST_PAGE_SIZE / 8;
977 
978 	/* Make sure 64 bit math will not overflow. */
979 	if (a > ULONG_MAX - (unsigned long)log_base ||
980 	    a + (unsigned long)log_base > ULONG_MAX)
981 		return false;
982 
983 	return access_ok(log_base + a,
984 			 (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8);
985 }
986 
987 static bool vhost_overflow(u64 uaddr, u64 size)
988 {
989 	/* Make sure 64 bit math will not overflow. */
990 	return uaddr > ULONG_MAX || size > ULONG_MAX || uaddr > ULONG_MAX - size;
991 }
992 
993 /* Caller should have vq mutex and device mutex. */
994 static bool vq_memory_access_ok(void __user *log_base, struct vhost_umem *umem,
995 				int log_all)
996 {
997 	struct vhost_umem_node *node;
998 
999 	if (!umem)
1000 		return false;
1001 
1002 	list_for_each_entry(node, &umem->umem_list, link) {
1003 		unsigned long a = node->userspace_addr;
1004 
1005 		if (vhost_overflow(node->userspace_addr, node->size))
1006 			return false;
1007 
1008 
1009 		if (!access_ok((void __user *)a,
1010 				    node->size))
1011 			return false;
1012 		else if (log_all && !log_access_ok(log_base,
1013 						   node->start,
1014 						   node->size))
1015 			return false;
1016 	}
1017 	return true;
1018 }
1019 
1020 static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq,
1021 					       u64 addr, unsigned int size,
1022 					       int type)
1023 {
1024 	const struct vhost_umem_node *node = vq->meta_iotlb[type];
1025 
1026 	if (!node)
1027 		return NULL;
1028 
1029 	return (void *)(uintptr_t)(node->userspace_addr + addr - node->start);
1030 }
1031 
1032 /* Can we switch to this memory table? */
1033 /* Caller should have device mutex but not vq mutex */
1034 static bool memory_access_ok(struct vhost_dev *d, struct vhost_umem *umem,
1035 			     int log_all)
1036 {
1037 	int i;
1038 
1039 	for (i = 0; i < d->nvqs; ++i) {
1040 		bool ok;
1041 		bool log;
1042 
1043 		mutex_lock(&d->vqs[i]->mutex);
1044 		log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL);
1045 		/* If ring is inactive, will check when it's enabled. */
1046 		if (d->vqs[i]->private_data)
1047 			ok = vq_memory_access_ok(d->vqs[i]->log_base,
1048 						 umem, log);
1049 		else
1050 			ok = true;
1051 		mutex_unlock(&d->vqs[i]->mutex);
1052 		if (!ok)
1053 			return false;
1054 	}
1055 	return true;
1056 }
1057 
1058 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
1059 			  struct iovec iov[], int iov_size, int access);
1060 
1061 static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to,
1062 			      const void *from, unsigned size)
1063 {
1064 	int ret;
1065 
1066 	if (!vq->iotlb)
1067 		return __copy_to_user(to, from, size);
1068 	else {
1069 		/* This function should be called after iotlb
1070 		 * prefetch, which means we're sure that all vq
1071 		 * could be access through iotlb. So -EAGAIN should
1072 		 * not happen in this case.
1073 		 */
1074 		struct iov_iter t;
1075 		void __user *uaddr = vhost_vq_meta_fetch(vq,
1076 				     (u64)(uintptr_t)to, size,
1077 				     VHOST_ADDR_USED);
1078 
1079 		if (uaddr)
1080 			return __copy_to_user(uaddr, from, size);
1081 
1082 		ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov,
1083 				     ARRAY_SIZE(vq->iotlb_iov),
1084 				     VHOST_ACCESS_WO);
1085 		if (ret < 0)
1086 			goto out;
1087 		iov_iter_init(&t, WRITE, vq->iotlb_iov, ret, size);
1088 		ret = copy_to_iter(from, size, &t);
1089 		if (ret == size)
1090 			ret = 0;
1091 	}
1092 out:
1093 	return ret;
1094 }
1095 
1096 static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to,
1097 				void __user *from, unsigned size)
1098 {
1099 	int ret;
1100 
1101 	if (!vq->iotlb)
1102 		return __copy_from_user(to, from, size);
1103 	else {
1104 		/* This function should be called after iotlb
1105 		 * prefetch, which means we're sure that vq
1106 		 * could be access through iotlb. So -EAGAIN should
1107 		 * not happen in this case.
1108 		 */
1109 		void __user *uaddr = vhost_vq_meta_fetch(vq,
1110 				     (u64)(uintptr_t)from, size,
1111 				     VHOST_ADDR_DESC);
1112 		struct iov_iter f;
1113 
1114 		if (uaddr)
1115 			return __copy_from_user(to, uaddr, size);
1116 
1117 		ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov,
1118 				     ARRAY_SIZE(vq->iotlb_iov),
1119 				     VHOST_ACCESS_RO);
1120 		if (ret < 0) {
1121 			vq_err(vq, "IOTLB translation failure: uaddr "
1122 			       "%p size 0x%llx\n", from,
1123 			       (unsigned long long) size);
1124 			goto out;
1125 		}
1126 		iov_iter_init(&f, READ, vq->iotlb_iov, ret, size);
1127 		ret = copy_from_iter(to, size, &f);
1128 		if (ret == size)
1129 			ret = 0;
1130 	}
1131 
1132 out:
1133 	return ret;
1134 }
1135 
1136 static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq,
1137 					  void __user *addr, unsigned int size,
1138 					  int type)
1139 {
1140 	int ret;
1141 
1142 	ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov,
1143 			     ARRAY_SIZE(vq->iotlb_iov),
1144 			     VHOST_ACCESS_RO);
1145 	if (ret < 0) {
1146 		vq_err(vq, "IOTLB translation failure: uaddr "
1147 			"%p size 0x%llx\n", addr,
1148 			(unsigned long long) size);
1149 		return NULL;
1150 	}
1151 
1152 	if (ret != 1 || vq->iotlb_iov[0].iov_len != size) {
1153 		vq_err(vq, "Non atomic userspace memory access: uaddr "
1154 			"%p size 0x%llx\n", addr,
1155 			(unsigned long long) size);
1156 		return NULL;
1157 	}
1158 
1159 	return vq->iotlb_iov[0].iov_base;
1160 }
1161 
1162 /* This function should be called after iotlb
1163  * prefetch, which means we're sure that vq
1164  * could be access through iotlb. So -EAGAIN should
1165  * not happen in this case.
1166  */
1167 static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq,
1168 					    void *addr, unsigned int size,
1169 					    int type)
1170 {
1171 	void __user *uaddr = vhost_vq_meta_fetch(vq,
1172 			     (u64)(uintptr_t)addr, size, type);
1173 	if (uaddr)
1174 		return uaddr;
1175 
1176 	return __vhost_get_user_slow(vq, addr, size, type);
1177 }
1178 
1179 #define vhost_put_user(vq, x, ptr)		\
1180 ({ \
1181 	int ret = -EFAULT; \
1182 	if (!vq->iotlb) { \
1183 		ret = __put_user(x, ptr); \
1184 	} else { \
1185 		__typeof__(ptr) to = \
1186 			(__typeof__(ptr)) __vhost_get_user(vq, ptr,	\
1187 					  sizeof(*ptr), VHOST_ADDR_USED); \
1188 		if (to != NULL) \
1189 			ret = __put_user(x, to); \
1190 		else \
1191 			ret = -EFAULT;	\
1192 	} \
1193 	ret; \
1194 })
1195 
1196 static inline int vhost_put_avail_event(struct vhost_virtqueue *vq)
1197 {
1198 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1199 	struct vhost_map *map;
1200 	struct vring_used *used;
1201 
1202 	if (!vq->iotlb) {
1203 		rcu_read_lock();
1204 
1205 		map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
1206 		if (likely(map)) {
1207 			used = map->addr;
1208 			*((__virtio16 *)&used->ring[vq->num]) =
1209 				cpu_to_vhost16(vq, vq->avail_idx);
1210 			rcu_read_unlock();
1211 			return 0;
1212 		}
1213 
1214 		rcu_read_unlock();
1215 	}
1216 #endif
1217 
1218 	return vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx),
1219 			      vhost_avail_event(vq));
1220 }
1221 
1222 static inline int vhost_put_used(struct vhost_virtqueue *vq,
1223 				 struct vring_used_elem *head, int idx,
1224 				 int count)
1225 {
1226 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1227 	struct vhost_map *map;
1228 	struct vring_used *used;
1229 	size_t size;
1230 
1231 	if (!vq->iotlb) {
1232 		rcu_read_lock();
1233 
1234 		map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
1235 		if (likely(map)) {
1236 			used = map->addr;
1237 			size = count * sizeof(*head);
1238 			memcpy(used->ring + idx, head, size);
1239 			rcu_read_unlock();
1240 			return 0;
1241 		}
1242 
1243 		rcu_read_unlock();
1244 	}
1245 #endif
1246 
1247 	return vhost_copy_to_user(vq, vq->used->ring + idx, head,
1248 				  count * sizeof(*head));
1249 }
1250 
1251 static inline int vhost_put_used_flags(struct vhost_virtqueue *vq)
1252 
1253 {
1254 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1255 	struct vhost_map *map;
1256 	struct vring_used *used;
1257 
1258 	if (!vq->iotlb) {
1259 		rcu_read_lock();
1260 
1261 		map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
1262 		if (likely(map)) {
1263 			used = map->addr;
1264 			used->flags = cpu_to_vhost16(vq, vq->used_flags);
1265 			rcu_read_unlock();
1266 			return 0;
1267 		}
1268 
1269 		rcu_read_unlock();
1270 	}
1271 #endif
1272 
1273 	return vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags),
1274 			      &vq->used->flags);
1275 }
1276 
1277 static inline int vhost_put_used_idx(struct vhost_virtqueue *vq)
1278 
1279 {
1280 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1281 	struct vhost_map *map;
1282 	struct vring_used *used;
1283 
1284 	if (!vq->iotlb) {
1285 		rcu_read_lock();
1286 
1287 		map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
1288 		if (likely(map)) {
1289 			used = map->addr;
1290 			used->idx = cpu_to_vhost16(vq, vq->last_used_idx);
1291 			rcu_read_unlock();
1292 			return 0;
1293 		}
1294 
1295 		rcu_read_unlock();
1296 	}
1297 #endif
1298 
1299 	return vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx),
1300 			      &vq->used->idx);
1301 }
1302 
1303 #define vhost_get_user(vq, x, ptr, type)		\
1304 ({ \
1305 	int ret; \
1306 	if (!vq->iotlb) { \
1307 		ret = __get_user(x, ptr); \
1308 	} else { \
1309 		__typeof__(ptr) from = \
1310 			(__typeof__(ptr)) __vhost_get_user(vq, ptr, \
1311 							   sizeof(*ptr), \
1312 							   type); \
1313 		if (from != NULL) \
1314 			ret = __get_user(x, from); \
1315 		else \
1316 			ret = -EFAULT; \
1317 	} \
1318 	ret; \
1319 })
1320 
1321 #define vhost_get_avail(vq, x, ptr) \
1322 	vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL)
1323 
1324 #define vhost_get_used(vq, x, ptr) \
1325 	vhost_get_user(vq, x, ptr, VHOST_ADDR_USED)
1326 
1327 static void vhost_dev_lock_vqs(struct vhost_dev *d)
1328 {
1329 	int i = 0;
1330 	for (i = 0; i < d->nvqs; ++i)
1331 		mutex_lock_nested(&d->vqs[i]->mutex, i);
1332 }
1333 
1334 static void vhost_dev_unlock_vqs(struct vhost_dev *d)
1335 {
1336 	int i = 0;
1337 	for (i = 0; i < d->nvqs; ++i)
1338 		mutex_unlock(&d->vqs[i]->mutex);
1339 }
1340 
1341 static inline int vhost_get_avail_idx(struct vhost_virtqueue *vq,
1342 				      __virtio16 *idx)
1343 {
1344 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1345 	struct vhost_map *map;
1346 	struct vring_avail *avail;
1347 
1348 	if (!vq->iotlb) {
1349 		rcu_read_lock();
1350 
1351 		map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
1352 		if (likely(map)) {
1353 			avail = map->addr;
1354 			*idx = avail->idx;
1355 			rcu_read_unlock();
1356 			return 0;
1357 		}
1358 
1359 		rcu_read_unlock();
1360 	}
1361 #endif
1362 
1363 	return vhost_get_avail(vq, *idx, &vq->avail->idx);
1364 }
1365 
1366 static inline int vhost_get_avail_head(struct vhost_virtqueue *vq,
1367 				       __virtio16 *head, int idx)
1368 {
1369 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1370 	struct vhost_map *map;
1371 	struct vring_avail *avail;
1372 
1373 	if (!vq->iotlb) {
1374 		rcu_read_lock();
1375 
1376 		map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
1377 		if (likely(map)) {
1378 			avail = map->addr;
1379 			*head = avail->ring[idx & (vq->num - 1)];
1380 			rcu_read_unlock();
1381 			return 0;
1382 		}
1383 
1384 		rcu_read_unlock();
1385 	}
1386 #endif
1387 
1388 	return vhost_get_avail(vq, *head,
1389 			       &vq->avail->ring[idx & (vq->num - 1)]);
1390 }
1391 
1392 static inline int vhost_get_avail_flags(struct vhost_virtqueue *vq,
1393 					__virtio16 *flags)
1394 {
1395 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1396 	struct vhost_map *map;
1397 	struct vring_avail *avail;
1398 
1399 	if (!vq->iotlb) {
1400 		rcu_read_lock();
1401 
1402 		map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
1403 		if (likely(map)) {
1404 			avail = map->addr;
1405 			*flags = avail->flags;
1406 			rcu_read_unlock();
1407 			return 0;
1408 		}
1409 
1410 		rcu_read_unlock();
1411 	}
1412 #endif
1413 
1414 	return vhost_get_avail(vq, *flags, &vq->avail->flags);
1415 }
1416 
1417 static inline int vhost_get_used_event(struct vhost_virtqueue *vq,
1418 				       __virtio16 *event)
1419 {
1420 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1421 	struct vhost_map *map;
1422 	struct vring_avail *avail;
1423 
1424 	if (!vq->iotlb) {
1425 		rcu_read_lock();
1426 		map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
1427 		if (likely(map)) {
1428 			avail = map->addr;
1429 			*event = (__virtio16)avail->ring[vq->num];
1430 			rcu_read_unlock();
1431 			return 0;
1432 		}
1433 		rcu_read_unlock();
1434 	}
1435 #endif
1436 
1437 	return vhost_get_avail(vq, *event, vhost_used_event(vq));
1438 }
1439 
1440 static inline int vhost_get_used_idx(struct vhost_virtqueue *vq,
1441 				     __virtio16 *idx)
1442 {
1443 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1444 	struct vhost_map *map;
1445 	struct vring_used *used;
1446 
1447 	if (!vq->iotlb) {
1448 		rcu_read_lock();
1449 
1450 		map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
1451 		if (likely(map)) {
1452 			used = map->addr;
1453 			*idx = used->idx;
1454 			rcu_read_unlock();
1455 			return 0;
1456 		}
1457 
1458 		rcu_read_unlock();
1459 	}
1460 #endif
1461 
1462 	return vhost_get_used(vq, *idx, &vq->used->idx);
1463 }
1464 
1465 static inline int vhost_get_desc(struct vhost_virtqueue *vq,
1466 				 struct vring_desc *desc, int idx)
1467 {
1468 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1469 	struct vhost_map *map;
1470 	struct vring_desc *d;
1471 
1472 	if (!vq->iotlb) {
1473 		rcu_read_lock();
1474 
1475 		map = rcu_dereference(vq->maps[VHOST_ADDR_DESC]);
1476 		if (likely(map)) {
1477 			d = map->addr;
1478 			*desc = *(d + idx);
1479 			rcu_read_unlock();
1480 			return 0;
1481 		}
1482 
1483 		rcu_read_unlock();
1484 	}
1485 #endif
1486 
1487 	return vhost_copy_from_user(vq, desc, vq->desc + idx, sizeof(*desc));
1488 }
1489 
1490 static int vhost_new_umem_range(struct vhost_umem *umem,
1491 				u64 start, u64 size, u64 end,
1492 				u64 userspace_addr, int perm)
1493 {
1494 	struct vhost_umem_node *tmp, *node;
1495 
1496 	if (!size)
1497 		return -EFAULT;
1498 
1499 	node = kmalloc(sizeof(*node), GFP_ATOMIC);
1500 	if (!node)
1501 		return -ENOMEM;
1502 
1503 	if (umem->numem == max_iotlb_entries) {
1504 		tmp = list_first_entry(&umem->umem_list, typeof(*tmp), link);
1505 		vhost_umem_free(umem, tmp);
1506 	}
1507 
1508 	node->start = start;
1509 	node->size = size;
1510 	node->last = end;
1511 	node->userspace_addr = userspace_addr;
1512 	node->perm = perm;
1513 	INIT_LIST_HEAD(&node->link);
1514 	list_add_tail(&node->link, &umem->umem_list);
1515 	vhost_umem_interval_tree_insert(node, &umem->umem_tree);
1516 	umem->numem++;
1517 
1518 	return 0;
1519 }
1520 
1521 static void vhost_del_umem_range(struct vhost_umem *umem,
1522 				 u64 start, u64 end)
1523 {
1524 	struct vhost_umem_node *node;
1525 
1526 	while ((node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1527 							   start, end)))
1528 		vhost_umem_free(umem, node);
1529 }
1530 
1531 static void vhost_iotlb_notify_vq(struct vhost_dev *d,
1532 				  struct vhost_iotlb_msg *msg)
1533 {
1534 	struct vhost_msg_node *node, *n;
1535 
1536 	spin_lock(&d->iotlb_lock);
1537 
1538 	list_for_each_entry_safe(node, n, &d->pending_list, node) {
1539 		struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb;
1540 		if (msg->iova <= vq_msg->iova &&
1541 		    msg->iova + msg->size - 1 >= vq_msg->iova &&
1542 		    vq_msg->type == VHOST_IOTLB_MISS) {
1543 			vhost_poll_queue(&node->vq->poll);
1544 			list_del(&node->node);
1545 			kfree(node);
1546 		}
1547 	}
1548 
1549 	spin_unlock(&d->iotlb_lock);
1550 }
1551 
1552 static bool umem_access_ok(u64 uaddr, u64 size, int access)
1553 {
1554 	unsigned long a = uaddr;
1555 
1556 	/* Make sure 64 bit math will not overflow. */
1557 	if (vhost_overflow(uaddr, size))
1558 		return false;
1559 
1560 	if ((access & VHOST_ACCESS_RO) &&
1561 	    !access_ok((void __user *)a, size))
1562 		return false;
1563 	if ((access & VHOST_ACCESS_WO) &&
1564 	    !access_ok((void __user *)a, size))
1565 		return false;
1566 	return true;
1567 }
1568 
1569 static int vhost_process_iotlb_msg(struct vhost_dev *dev,
1570 				   struct vhost_iotlb_msg *msg)
1571 {
1572 	int ret = 0;
1573 
1574 	mutex_lock(&dev->mutex);
1575 	vhost_dev_lock_vqs(dev);
1576 	switch (msg->type) {
1577 	case VHOST_IOTLB_UPDATE:
1578 		if (!dev->iotlb) {
1579 			ret = -EFAULT;
1580 			break;
1581 		}
1582 		if (!umem_access_ok(msg->uaddr, msg->size, msg->perm)) {
1583 			ret = -EFAULT;
1584 			break;
1585 		}
1586 		vhost_vq_meta_reset(dev);
1587 		if (vhost_new_umem_range(dev->iotlb, msg->iova, msg->size,
1588 					 msg->iova + msg->size - 1,
1589 					 msg->uaddr, msg->perm)) {
1590 			ret = -ENOMEM;
1591 			break;
1592 		}
1593 		vhost_iotlb_notify_vq(dev, msg);
1594 		break;
1595 	case VHOST_IOTLB_INVALIDATE:
1596 		if (!dev->iotlb) {
1597 			ret = -EFAULT;
1598 			break;
1599 		}
1600 		vhost_vq_meta_reset(dev);
1601 		vhost_del_umem_range(dev->iotlb, msg->iova,
1602 				     msg->iova + msg->size - 1);
1603 		break;
1604 	default:
1605 		ret = -EINVAL;
1606 		break;
1607 	}
1608 
1609 	vhost_dev_unlock_vqs(dev);
1610 	mutex_unlock(&dev->mutex);
1611 
1612 	return ret;
1613 }
1614 ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
1615 			     struct iov_iter *from)
1616 {
1617 	struct vhost_iotlb_msg msg;
1618 	size_t offset;
1619 	int type, ret;
1620 
1621 	ret = copy_from_iter(&type, sizeof(type), from);
1622 	if (ret != sizeof(type)) {
1623 		ret = -EINVAL;
1624 		goto done;
1625 	}
1626 
1627 	switch (type) {
1628 	case VHOST_IOTLB_MSG:
1629 		/* There maybe a hole after type for V1 message type,
1630 		 * so skip it here.
1631 		 */
1632 		offset = offsetof(struct vhost_msg, iotlb) - sizeof(int);
1633 		break;
1634 	case VHOST_IOTLB_MSG_V2:
1635 		offset = sizeof(__u32);
1636 		break;
1637 	default:
1638 		ret = -EINVAL;
1639 		goto done;
1640 	}
1641 
1642 	iov_iter_advance(from, offset);
1643 	ret = copy_from_iter(&msg, sizeof(msg), from);
1644 	if (ret != sizeof(msg)) {
1645 		ret = -EINVAL;
1646 		goto done;
1647 	}
1648 	if (vhost_process_iotlb_msg(dev, &msg)) {
1649 		ret = -EFAULT;
1650 		goto done;
1651 	}
1652 
1653 	ret = (type == VHOST_IOTLB_MSG) ? sizeof(struct vhost_msg) :
1654 	      sizeof(struct vhost_msg_v2);
1655 done:
1656 	return ret;
1657 }
1658 EXPORT_SYMBOL(vhost_chr_write_iter);
1659 
1660 __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
1661 			    poll_table *wait)
1662 {
1663 	__poll_t mask = 0;
1664 
1665 	poll_wait(file, &dev->wait, wait);
1666 
1667 	if (!list_empty(&dev->read_list))
1668 		mask |= EPOLLIN | EPOLLRDNORM;
1669 
1670 	return mask;
1671 }
1672 EXPORT_SYMBOL(vhost_chr_poll);
1673 
1674 ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
1675 			    int noblock)
1676 {
1677 	DEFINE_WAIT(wait);
1678 	struct vhost_msg_node *node;
1679 	ssize_t ret = 0;
1680 	unsigned size = sizeof(struct vhost_msg);
1681 
1682 	if (iov_iter_count(to) < size)
1683 		return 0;
1684 
1685 	while (1) {
1686 		if (!noblock)
1687 			prepare_to_wait(&dev->wait, &wait,
1688 					TASK_INTERRUPTIBLE);
1689 
1690 		node = vhost_dequeue_msg(dev, &dev->read_list);
1691 		if (node)
1692 			break;
1693 		if (noblock) {
1694 			ret = -EAGAIN;
1695 			break;
1696 		}
1697 		if (signal_pending(current)) {
1698 			ret = -ERESTARTSYS;
1699 			break;
1700 		}
1701 		if (!dev->iotlb) {
1702 			ret = -EBADFD;
1703 			break;
1704 		}
1705 
1706 		schedule();
1707 	}
1708 
1709 	if (!noblock)
1710 		finish_wait(&dev->wait, &wait);
1711 
1712 	if (node) {
1713 		struct vhost_iotlb_msg *msg;
1714 		void *start = &node->msg;
1715 
1716 		switch (node->msg.type) {
1717 		case VHOST_IOTLB_MSG:
1718 			size = sizeof(node->msg);
1719 			msg = &node->msg.iotlb;
1720 			break;
1721 		case VHOST_IOTLB_MSG_V2:
1722 			size = sizeof(node->msg_v2);
1723 			msg = &node->msg_v2.iotlb;
1724 			break;
1725 		default:
1726 			BUG();
1727 			break;
1728 		}
1729 
1730 		ret = copy_to_iter(start, size, to);
1731 		if (ret != size || msg->type != VHOST_IOTLB_MISS) {
1732 			kfree(node);
1733 			return ret;
1734 		}
1735 		vhost_enqueue_msg(dev, &dev->pending_list, node);
1736 	}
1737 
1738 	return ret;
1739 }
1740 EXPORT_SYMBOL_GPL(vhost_chr_read_iter);
1741 
1742 static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access)
1743 {
1744 	struct vhost_dev *dev = vq->dev;
1745 	struct vhost_msg_node *node;
1746 	struct vhost_iotlb_msg *msg;
1747 	bool v2 = vhost_backend_has_feature(vq, VHOST_BACKEND_F_IOTLB_MSG_V2);
1748 
1749 	node = vhost_new_msg(vq, v2 ? VHOST_IOTLB_MSG_V2 : VHOST_IOTLB_MSG);
1750 	if (!node)
1751 		return -ENOMEM;
1752 
1753 	if (v2) {
1754 		node->msg_v2.type = VHOST_IOTLB_MSG_V2;
1755 		msg = &node->msg_v2.iotlb;
1756 	} else {
1757 		msg = &node->msg.iotlb;
1758 	}
1759 
1760 	msg->type = VHOST_IOTLB_MISS;
1761 	msg->iova = iova;
1762 	msg->perm = access;
1763 
1764 	vhost_enqueue_msg(dev, &dev->read_list, node);
1765 
1766 	return 0;
1767 }
1768 
1769 static bool vq_access_ok(struct vhost_virtqueue *vq, unsigned int num,
1770 			 struct vring_desc __user *desc,
1771 			 struct vring_avail __user *avail,
1772 			 struct vring_used __user *used)
1773 
1774 {
1775 	return access_ok(desc, vhost_get_desc_size(vq, num)) &&
1776 	       access_ok(avail, vhost_get_avail_size(vq, num)) &&
1777 	       access_ok(used, vhost_get_used_size(vq, num));
1778 }
1779 
1780 static void vhost_vq_meta_update(struct vhost_virtqueue *vq,
1781 				 const struct vhost_umem_node *node,
1782 				 int type)
1783 {
1784 	int access = (type == VHOST_ADDR_USED) ?
1785 		     VHOST_ACCESS_WO : VHOST_ACCESS_RO;
1786 
1787 	if (likely(node->perm & access))
1788 		vq->meta_iotlb[type] = node;
1789 }
1790 
1791 static bool iotlb_access_ok(struct vhost_virtqueue *vq,
1792 			    int access, u64 addr, u64 len, int type)
1793 {
1794 	const struct vhost_umem_node *node;
1795 	struct vhost_umem *umem = vq->iotlb;
1796 	u64 s = 0, size, orig_addr = addr, last = addr + len - 1;
1797 
1798 	if (vhost_vq_meta_fetch(vq, addr, len, type))
1799 		return true;
1800 
1801 	while (len > s) {
1802 		node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1803 							   addr,
1804 							   last);
1805 		if (node == NULL || node->start > addr) {
1806 			vhost_iotlb_miss(vq, addr, access);
1807 			return false;
1808 		} else if (!(node->perm & access)) {
1809 			/* Report the possible access violation by
1810 			 * request another translation from userspace.
1811 			 */
1812 			return false;
1813 		}
1814 
1815 		size = node->size - addr + node->start;
1816 
1817 		if (orig_addr == addr && size >= len)
1818 			vhost_vq_meta_update(vq, node, type);
1819 
1820 		s += size;
1821 		addr += size;
1822 	}
1823 
1824 	return true;
1825 }
1826 
1827 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1828 static void vhost_vq_map_prefetch(struct vhost_virtqueue *vq)
1829 {
1830 	struct vhost_map __rcu *map;
1831 	int i;
1832 
1833 	for (i = 0; i < VHOST_NUM_ADDRS; i++) {
1834 		rcu_read_lock();
1835 		map = rcu_dereference(vq->maps[i]);
1836 		rcu_read_unlock();
1837 		if (unlikely(!map))
1838 			vhost_map_prefetch(vq, i);
1839 	}
1840 }
1841 #endif
1842 
1843 int vq_meta_prefetch(struct vhost_virtqueue *vq)
1844 {
1845 	unsigned int num = vq->num;
1846 
1847 	if (!vq->iotlb) {
1848 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1849 		vhost_vq_map_prefetch(vq);
1850 #endif
1851 		return 1;
1852 	}
1853 
1854 	return iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->desc,
1855 			       vhost_get_desc_size(vq, num), VHOST_ADDR_DESC) &&
1856 	       iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->avail,
1857 			       vhost_get_avail_size(vq, num),
1858 			       VHOST_ADDR_AVAIL) &&
1859 	       iotlb_access_ok(vq, VHOST_ACCESS_WO, (u64)(uintptr_t)vq->used,
1860 			       vhost_get_used_size(vq, num), VHOST_ADDR_USED);
1861 }
1862 EXPORT_SYMBOL_GPL(vq_meta_prefetch);
1863 
1864 /* Can we log writes? */
1865 /* Caller should have device mutex but not vq mutex */
1866 bool vhost_log_access_ok(struct vhost_dev *dev)
1867 {
1868 	return memory_access_ok(dev, dev->umem, 1);
1869 }
1870 EXPORT_SYMBOL_GPL(vhost_log_access_ok);
1871 
1872 /* Verify access for write logging. */
1873 /* Caller should have vq mutex and device mutex */
1874 static bool vq_log_access_ok(struct vhost_virtqueue *vq,
1875 			     void __user *log_base)
1876 {
1877 	return vq_memory_access_ok(log_base, vq->umem,
1878 				   vhost_has_feature(vq, VHOST_F_LOG_ALL)) &&
1879 		(!vq->log_used || log_access_ok(log_base, vq->log_addr,
1880 				  vhost_get_used_size(vq, vq->num)));
1881 }
1882 
1883 /* Can we start vq? */
1884 /* Caller should have vq mutex and device mutex */
1885 bool vhost_vq_access_ok(struct vhost_virtqueue *vq)
1886 {
1887 	if (!vq_log_access_ok(vq, vq->log_base))
1888 		return false;
1889 
1890 	/* Access validation occurs at prefetch time with IOTLB */
1891 	if (vq->iotlb)
1892 		return true;
1893 
1894 	return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used);
1895 }
1896 EXPORT_SYMBOL_GPL(vhost_vq_access_ok);
1897 
1898 static struct vhost_umem *vhost_umem_alloc(void)
1899 {
1900 	struct vhost_umem *umem = kvzalloc(sizeof(*umem), GFP_KERNEL);
1901 
1902 	if (!umem)
1903 		return NULL;
1904 
1905 	umem->umem_tree = RB_ROOT_CACHED;
1906 	umem->numem = 0;
1907 	INIT_LIST_HEAD(&umem->umem_list);
1908 
1909 	return umem;
1910 }
1911 
1912 static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)
1913 {
1914 	struct vhost_memory mem, *newmem;
1915 	struct vhost_memory_region *region;
1916 	struct vhost_umem *newumem, *oldumem;
1917 	unsigned long size = offsetof(struct vhost_memory, regions);
1918 	int i;
1919 
1920 	if (copy_from_user(&mem, m, size))
1921 		return -EFAULT;
1922 	if (mem.padding)
1923 		return -EOPNOTSUPP;
1924 	if (mem.nregions > max_mem_regions)
1925 		return -E2BIG;
1926 	newmem = kvzalloc(struct_size(newmem, regions, mem.nregions),
1927 			GFP_KERNEL);
1928 	if (!newmem)
1929 		return -ENOMEM;
1930 
1931 	memcpy(newmem, &mem, size);
1932 	if (copy_from_user(newmem->regions, m->regions,
1933 			   mem.nregions * sizeof *m->regions)) {
1934 		kvfree(newmem);
1935 		return -EFAULT;
1936 	}
1937 
1938 	newumem = vhost_umem_alloc();
1939 	if (!newumem) {
1940 		kvfree(newmem);
1941 		return -ENOMEM;
1942 	}
1943 
1944 	for (region = newmem->regions;
1945 	     region < newmem->regions + mem.nregions;
1946 	     region++) {
1947 		if (vhost_new_umem_range(newumem,
1948 					 region->guest_phys_addr,
1949 					 region->memory_size,
1950 					 region->guest_phys_addr +
1951 					 region->memory_size - 1,
1952 					 region->userspace_addr,
1953 					 VHOST_ACCESS_RW))
1954 			goto err;
1955 	}
1956 
1957 	if (!memory_access_ok(d, newumem, 0))
1958 		goto err;
1959 
1960 	oldumem = d->umem;
1961 	d->umem = newumem;
1962 
1963 	/* All memory accesses are done under some VQ mutex. */
1964 	for (i = 0; i < d->nvqs; ++i) {
1965 		mutex_lock(&d->vqs[i]->mutex);
1966 		d->vqs[i]->umem = newumem;
1967 		mutex_unlock(&d->vqs[i]->mutex);
1968 	}
1969 
1970 	kvfree(newmem);
1971 	vhost_umem_clean(oldumem);
1972 	return 0;
1973 
1974 err:
1975 	vhost_umem_clean(newumem);
1976 	kvfree(newmem);
1977 	return -EFAULT;
1978 }
1979 
1980 static long vhost_vring_set_num(struct vhost_dev *d,
1981 				struct vhost_virtqueue *vq,
1982 				void __user *argp)
1983 {
1984 	struct vhost_vring_state s;
1985 
1986 	/* Resizing ring with an active backend?
1987 	 * You don't want to do that. */
1988 	if (vq->private_data)
1989 		return -EBUSY;
1990 
1991 	if (copy_from_user(&s, argp, sizeof s))
1992 		return -EFAULT;
1993 
1994 	if (!s.num || s.num > 0xffff || (s.num & (s.num - 1)))
1995 		return -EINVAL;
1996 	vq->num = s.num;
1997 
1998 	return 0;
1999 }
2000 
2001 static long vhost_vring_set_addr(struct vhost_dev *d,
2002 				 struct vhost_virtqueue *vq,
2003 				 void __user *argp)
2004 {
2005 	struct vhost_vring_addr a;
2006 
2007 	if (copy_from_user(&a, argp, sizeof a))
2008 		return -EFAULT;
2009 	if (a.flags & ~(0x1 << VHOST_VRING_F_LOG))
2010 		return -EOPNOTSUPP;
2011 
2012 	/* For 32bit, verify that the top 32bits of the user
2013 	   data are set to zero. */
2014 	if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr ||
2015 	    (u64)(unsigned long)a.used_user_addr != a.used_user_addr ||
2016 	    (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr)
2017 		return -EFAULT;
2018 
2019 	/* Make sure it's safe to cast pointers to vring types. */
2020 	BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE);
2021 	BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE);
2022 	if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) ||
2023 	    (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) ||
2024 	    (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1)))
2025 		return -EINVAL;
2026 
2027 	/* We only verify access here if backend is configured.
2028 	 * If it is not, we don't as size might not have been setup.
2029 	 * We will verify when backend is configured. */
2030 	if (vq->private_data) {
2031 		if (!vq_access_ok(vq, vq->num,
2032 			(void __user *)(unsigned long)a.desc_user_addr,
2033 			(void __user *)(unsigned long)a.avail_user_addr,
2034 			(void __user *)(unsigned long)a.used_user_addr))
2035 			return -EINVAL;
2036 
2037 		/* Also validate log access for used ring if enabled. */
2038 		if ((a.flags & (0x1 << VHOST_VRING_F_LOG)) &&
2039 			!log_access_ok(vq->log_base, a.log_guest_addr,
2040 				sizeof *vq->used +
2041 				vq->num * sizeof *vq->used->ring))
2042 			return -EINVAL;
2043 	}
2044 
2045 	vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG));
2046 	vq->desc = (void __user *)(unsigned long)a.desc_user_addr;
2047 	vq->avail = (void __user *)(unsigned long)a.avail_user_addr;
2048 	vq->log_addr = a.log_guest_addr;
2049 	vq->used = (void __user *)(unsigned long)a.used_user_addr;
2050 
2051 	return 0;
2052 }
2053 
2054 static long vhost_vring_set_num_addr(struct vhost_dev *d,
2055 				     struct vhost_virtqueue *vq,
2056 				     unsigned int ioctl,
2057 				     void __user *argp)
2058 {
2059 	long r;
2060 
2061 	mutex_lock(&vq->mutex);
2062 
2063 #if VHOST_ARCH_CAN_ACCEL_UACCESS
2064 	/* Unregister MMU notifer to allow invalidation callback
2065 	 * can access vq->uaddrs[] without holding a lock.
2066 	 */
2067 	if (d->mm)
2068 		mmu_notifier_unregister(&d->mmu_notifier, d->mm);
2069 
2070 	vhost_uninit_vq_maps(vq);
2071 #endif
2072 
2073 	switch (ioctl) {
2074 	case VHOST_SET_VRING_NUM:
2075 		r = vhost_vring_set_num(d, vq, argp);
2076 		break;
2077 	case VHOST_SET_VRING_ADDR:
2078 		r = vhost_vring_set_addr(d, vq, argp);
2079 		break;
2080 	default:
2081 		BUG();
2082 	}
2083 
2084 #if VHOST_ARCH_CAN_ACCEL_UACCESS
2085 	vhost_setup_vq_uaddr(vq);
2086 
2087 	if (d->mm)
2088 		mmu_notifier_register(&d->mmu_notifier, d->mm);
2089 #endif
2090 
2091 	mutex_unlock(&vq->mutex);
2092 
2093 	return r;
2094 }
2095 long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
2096 {
2097 	struct file *eventfp, *filep = NULL;
2098 	bool pollstart = false, pollstop = false;
2099 	struct eventfd_ctx *ctx = NULL;
2100 	u32 __user *idxp = argp;
2101 	struct vhost_virtqueue *vq;
2102 	struct vhost_vring_state s;
2103 	struct vhost_vring_file f;
2104 	u32 idx;
2105 	long r;
2106 
2107 	r = get_user(idx, idxp);
2108 	if (r < 0)
2109 		return r;
2110 	if (idx >= d->nvqs)
2111 		return -ENOBUFS;
2112 
2113 	idx = array_index_nospec(idx, d->nvqs);
2114 	vq = d->vqs[idx];
2115 
2116 	if (ioctl == VHOST_SET_VRING_NUM ||
2117 	    ioctl == VHOST_SET_VRING_ADDR) {
2118 		return vhost_vring_set_num_addr(d, vq, ioctl, argp);
2119 	}
2120 
2121 	mutex_lock(&vq->mutex);
2122 
2123 	switch (ioctl) {
2124 	case VHOST_SET_VRING_BASE:
2125 		/* Moving base with an active backend?
2126 		 * You don't want to do that. */
2127 		if (vq->private_data) {
2128 			r = -EBUSY;
2129 			break;
2130 		}
2131 		if (copy_from_user(&s, argp, sizeof s)) {
2132 			r = -EFAULT;
2133 			break;
2134 		}
2135 		if (s.num > 0xffff) {
2136 			r = -EINVAL;
2137 			break;
2138 		}
2139 		vq->last_avail_idx = s.num;
2140 		/* Forget the cached index value. */
2141 		vq->avail_idx = vq->last_avail_idx;
2142 		break;
2143 	case VHOST_GET_VRING_BASE:
2144 		s.index = idx;
2145 		s.num = vq->last_avail_idx;
2146 		if (copy_to_user(argp, &s, sizeof s))
2147 			r = -EFAULT;
2148 		break;
2149 	case VHOST_SET_VRING_KICK:
2150 		if (copy_from_user(&f, argp, sizeof f)) {
2151 			r = -EFAULT;
2152 			break;
2153 		}
2154 		eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd);
2155 		if (IS_ERR(eventfp)) {
2156 			r = PTR_ERR(eventfp);
2157 			break;
2158 		}
2159 		if (eventfp != vq->kick) {
2160 			pollstop = (filep = vq->kick) != NULL;
2161 			pollstart = (vq->kick = eventfp) != NULL;
2162 		} else
2163 			filep = eventfp;
2164 		break;
2165 	case VHOST_SET_VRING_CALL:
2166 		if (copy_from_user(&f, argp, sizeof f)) {
2167 			r = -EFAULT;
2168 			break;
2169 		}
2170 		ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd);
2171 		if (IS_ERR(ctx)) {
2172 			r = PTR_ERR(ctx);
2173 			break;
2174 		}
2175 		swap(ctx, vq->call_ctx);
2176 		break;
2177 	case VHOST_SET_VRING_ERR:
2178 		if (copy_from_user(&f, argp, sizeof f)) {
2179 			r = -EFAULT;
2180 			break;
2181 		}
2182 		ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd);
2183 		if (IS_ERR(ctx)) {
2184 			r = PTR_ERR(ctx);
2185 			break;
2186 		}
2187 		swap(ctx, vq->error_ctx);
2188 		break;
2189 	case VHOST_SET_VRING_ENDIAN:
2190 		r = vhost_set_vring_endian(vq, argp);
2191 		break;
2192 	case VHOST_GET_VRING_ENDIAN:
2193 		r = vhost_get_vring_endian(vq, idx, argp);
2194 		break;
2195 	case VHOST_SET_VRING_BUSYLOOP_TIMEOUT:
2196 		if (copy_from_user(&s, argp, sizeof(s))) {
2197 			r = -EFAULT;
2198 			break;
2199 		}
2200 		vq->busyloop_timeout = s.num;
2201 		break;
2202 	case VHOST_GET_VRING_BUSYLOOP_TIMEOUT:
2203 		s.index = idx;
2204 		s.num = vq->busyloop_timeout;
2205 		if (copy_to_user(argp, &s, sizeof(s)))
2206 			r = -EFAULT;
2207 		break;
2208 	default:
2209 		r = -ENOIOCTLCMD;
2210 	}
2211 
2212 	if (pollstop && vq->handle_kick)
2213 		vhost_poll_stop(&vq->poll);
2214 
2215 	if (!IS_ERR_OR_NULL(ctx))
2216 		eventfd_ctx_put(ctx);
2217 	if (filep)
2218 		fput(filep);
2219 
2220 	if (pollstart && vq->handle_kick)
2221 		r = vhost_poll_start(&vq->poll, vq->kick);
2222 
2223 	mutex_unlock(&vq->mutex);
2224 
2225 	if (pollstop && vq->handle_kick)
2226 		vhost_poll_flush(&vq->poll);
2227 	return r;
2228 }
2229 EXPORT_SYMBOL_GPL(vhost_vring_ioctl);
2230 
2231 int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled)
2232 {
2233 	struct vhost_umem *niotlb, *oiotlb;
2234 	int i;
2235 
2236 	niotlb = vhost_umem_alloc();
2237 	if (!niotlb)
2238 		return -ENOMEM;
2239 
2240 	oiotlb = d->iotlb;
2241 	d->iotlb = niotlb;
2242 
2243 	for (i = 0; i < d->nvqs; ++i) {
2244 		struct vhost_virtqueue *vq = d->vqs[i];
2245 
2246 		mutex_lock(&vq->mutex);
2247 		vq->iotlb = niotlb;
2248 		__vhost_vq_meta_reset(vq);
2249 		mutex_unlock(&vq->mutex);
2250 	}
2251 
2252 	vhost_umem_clean(oiotlb);
2253 
2254 	return 0;
2255 }
2256 EXPORT_SYMBOL_GPL(vhost_init_device_iotlb);
2257 
2258 /* Caller must have device mutex */
2259 long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
2260 {
2261 	struct eventfd_ctx *ctx;
2262 	u64 p;
2263 	long r;
2264 	int i, fd;
2265 
2266 	/* If you are not the owner, you can become one */
2267 	if (ioctl == VHOST_SET_OWNER) {
2268 		r = vhost_dev_set_owner(d);
2269 		goto done;
2270 	}
2271 
2272 	/* You must be the owner to do anything else */
2273 	r = vhost_dev_check_owner(d);
2274 	if (r)
2275 		goto done;
2276 
2277 	switch (ioctl) {
2278 	case VHOST_SET_MEM_TABLE:
2279 		r = vhost_set_memory(d, argp);
2280 		break;
2281 	case VHOST_SET_LOG_BASE:
2282 		if (copy_from_user(&p, argp, sizeof p)) {
2283 			r = -EFAULT;
2284 			break;
2285 		}
2286 		if ((u64)(unsigned long)p != p) {
2287 			r = -EFAULT;
2288 			break;
2289 		}
2290 		for (i = 0; i < d->nvqs; ++i) {
2291 			struct vhost_virtqueue *vq;
2292 			void __user *base = (void __user *)(unsigned long)p;
2293 			vq = d->vqs[i];
2294 			mutex_lock(&vq->mutex);
2295 			/* If ring is inactive, will check when it's enabled. */
2296 			if (vq->private_data && !vq_log_access_ok(vq, base))
2297 				r = -EFAULT;
2298 			else
2299 				vq->log_base = base;
2300 			mutex_unlock(&vq->mutex);
2301 		}
2302 		break;
2303 	case VHOST_SET_LOG_FD:
2304 		r = get_user(fd, (int __user *)argp);
2305 		if (r < 0)
2306 			break;
2307 		ctx = fd == -1 ? NULL : eventfd_ctx_fdget(fd);
2308 		if (IS_ERR(ctx)) {
2309 			r = PTR_ERR(ctx);
2310 			break;
2311 		}
2312 		swap(ctx, d->log_ctx);
2313 		for (i = 0; i < d->nvqs; ++i) {
2314 			mutex_lock(&d->vqs[i]->mutex);
2315 			d->vqs[i]->log_ctx = d->log_ctx;
2316 			mutex_unlock(&d->vqs[i]->mutex);
2317 		}
2318 		if (ctx)
2319 			eventfd_ctx_put(ctx);
2320 		break;
2321 	default:
2322 		r = -ENOIOCTLCMD;
2323 		break;
2324 	}
2325 done:
2326 	return r;
2327 }
2328 EXPORT_SYMBOL_GPL(vhost_dev_ioctl);
2329 
2330 /* TODO: This is really inefficient.  We need something like get_user()
2331  * (instruction directly accesses the data, with an exception table entry
2332  * returning -EFAULT). See Documentation/x86/exception-tables.rst.
2333  */
2334 static int set_bit_to_user(int nr, void __user *addr)
2335 {
2336 	unsigned long log = (unsigned long)addr;
2337 	struct page *page;
2338 	void *base;
2339 	int bit = nr + (log % PAGE_SIZE) * 8;
2340 	int r;
2341 
2342 	r = get_user_pages_fast(log, 1, FOLL_WRITE, &page);
2343 	if (r < 0)
2344 		return r;
2345 	BUG_ON(r != 1);
2346 	base = kmap_atomic(page);
2347 	set_bit(bit, base);
2348 	kunmap_atomic(base);
2349 	set_page_dirty_lock(page);
2350 	put_page(page);
2351 	return 0;
2352 }
2353 
2354 static int log_write(void __user *log_base,
2355 		     u64 write_address, u64 write_length)
2356 {
2357 	u64 write_page = write_address / VHOST_PAGE_SIZE;
2358 	int r;
2359 
2360 	if (!write_length)
2361 		return 0;
2362 	write_length += write_address % VHOST_PAGE_SIZE;
2363 	for (;;) {
2364 		u64 base = (u64)(unsigned long)log_base;
2365 		u64 log = base + write_page / 8;
2366 		int bit = write_page % 8;
2367 		if ((u64)(unsigned long)log != log)
2368 			return -EFAULT;
2369 		r = set_bit_to_user(bit, (void __user *)(unsigned long)log);
2370 		if (r < 0)
2371 			return r;
2372 		if (write_length <= VHOST_PAGE_SIZE)
2373 			break;
2374 		write_length -= VHOST_PAGE_SIZE;
2375 		write_page += 1;
2376 	}
2377 	return r;
2378 }
2379 
2380 static int log_write_hva(struct vhost_virtqueue *vq, u64 hva, u64 len)
2381 {
2382 	struct vhost_umem *umem = vq->umem;
2383 	struct vhost_umem_node *u;
2384 	u64 start, end, l, min;
2385 	int r;
2386 	bool hit = false;
2387 
2388 	while (len) {
2389 		min = len;
2390 		/* More than one GPAs can be mapped into a single HVA. So
2391 		 * iterate all possible umems here to be safe.
2392 		 */
2393 		list_for_each_entry(u, &umem->umem_list, link) {
2394 			if (u->userspace_addr > hva - 1 + len ||
2395 			    u->userspace_addr - 1 + u->size < hva)
2396 				continue;
2397 			start = max(u->userspace_addr, hva);
2398 			end = min(u->userspace_addr - 1 + u->size,
2399 				  hva - 1 + len);
2400 			l = end - start + 1;
2401 			r = log_write(vq->log_base,
2402 				      u->start + start - u->userspace_addr,
2403 				      l);
2404 			if (r < 0)
2405 				return r;
2406 			hit = true;
2407 			min = min(l, min);
2408 		}
2409 
2410 		if (!hit)
2411 			return -EFAULT;
2412 
2413 		len -= min;
2414 		hva += min;
2415 	}
2416 
2417 	return 0;
2418 }
2419 
2420 static int log_used(struct vhost_virtqueue *vq, u64 used_offset, u64 len)
2421 {
2422 	struct iovec iov[64];
2423 	int i, ret;
2424 
2425 	if (!vq->iotlb)
2426 		return log_write(vq->log_base, vq->log_addr + used_offset, len);
2427 
2428 	ret = translate_desc(vq, (uintptr_t)vq->used + used_offset,
2429 			     len, iov, 64, VHOST_ACCESS_WO);
2430 	if (ret < 0)
2431 		return ret;
2432 
2433 	for (i = 0; i < ret; i++) {
2434 		ret = log_write_hva(vq,	(uintptr_t)iov[i].iov_base,
2435 				    iov[i].iov_len);
2436 		if (ret)
2437 			return ret;
2438 	}
2439 
2440 	return 0;
2441 }
2442 
2443 int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
2444 		    unsigned int log_num, u64 len, struct iovec *iov, int count)
2445 {
2446 	int i, r;
2447 
2448 	/* Make sure data written is seen before log. */
2449 	smp_wmb();
2450 
2451 	if (vq->iotlb) {
2452 		for (i = 0; i < count; i++) {
2453 			r = log_write_hva(vq, (uintptr_t)iov[i].iov_base,
2454 					  iov[i].iov_len);
2455 			if (r < 0)
2456 				return r;
2457 		}
2458 		return 0;
2459 	}
2460 
2461 	for (i = 0; i < log_num; ++i) {
2462 		u64 l = min(log[i].len, len);
2463 		r = log_write(vq->log_base, log[i].addr, l);
2464 		if (r < 0)
2465 			return r;
2466 		len -= l;
2467 		if (!len) {
2468 			if (vq->log_ctx)
2469 				eventfd_signal(vq->log_ctx, 1);
2470 			return 0;
2471 		}
2472 	}
2473 	/* Length written exceeds what we have stored. This is a bug. */
2474 	BUG();
2475 	return 0;
2476 }
2477 EXPORT_SYMBOL_GPL(vhost_log_write);
2478 
2479 static int vhost_update_used_flags(struct vhost_virtqueue *vq)
2480 {
2481 	void __user *used;
2482 	if (vhost_put_used_flags(vq))
2483 		return -EFAULT;
2484 	if (unlikely(vq->log_used)) {
2485 		/* Make sure the flag is seen before log. */
2486 		smp_wmb();
2487 		/* Log used flag write. */
2488 		used = &vq->used->flags;
2489 		log_used(vq, (used - (void __user *)vq->used),
2490 			 sizeof vq->used->flags);
2491 		if (vq->log_ctx)
2492 			eventfd_signal(vq->log_ctx, 1);
2493 	}
2494 	return 0;
2495 }
2496 
2497 static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event)
2498 {
2499 	if (vhost_put_avail_event(vq))
2500 		return -EFAULT;
2501 	if (unlikely(vq->log_used)) {
2502 		void __user *used;
2503 		/* Make sure the event is seen before log. */
2504 		smp_wmb();
2505 		/* Log avail event write */
2506 		used = vhost_avail_event(vq);
2507 		log_used(vq, (used - (void __user *)vq->used),
2508 			 sizeof *vhost_avail_event(vq));
2509 		if (vq->log_ctx)
2510 			eventfd_signal(vq->log_ctx, 1);
2511 	}
2512 	return 0;
2513 }
2514 
2515 int vhost_vq_init_access(struct vhost_virtqueue *vq)
2516 {
2517 	__virtio16 last_used_idx;
2518 	int r;
2519 	bool is_le = vq->is_le;
2520 
2521 	if (!vq->private_data)
2522 		return 0;
2523 
2524 	vhost_init_is_le(vq);
2525 
2526 	r = vhost_update_used_flags(vq);
2527 	if (r)
2528 		goto err;
2529 	vq->signalled_used_valid = false;
2530 	if (!vq->iotlb &&
2531 	    !access_ok(&vq->used->idx, sizeof vq->used->idx)) {
2532 		r = -EFAULT;
2533 		goto err;
2534 	}
2535 	r = vhost_get_used_idx(vq, &last_used_idx);
2536 	if (r) {
2537 		vq_err(vq, "Can't access used idx at %p\n",
2538 		       &vq->used->idx);
2539 		goto err;
2540 	}
2541 	vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx);
2542 	return 0;
2543 
2544 err:
2545 	vq->is_le = is_le;
2546 	return r;
2547 }
2548 EXPORT_SYMBOL_GPL(vhost_vq_init_access);
2549 
2550 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
2551 			  struct iovec iov[], int iov_size, int access)
2552 {
2553 	const struct vhost_umem_node *node;
2554 	struct vhost_dev *dev = vq->dev;
2555 	struct vhost_umem *umem = dev->iotlb ? dev->iotlb : dev->umem;
2556 	struct iovec *_iov;
2557 	u64 s = 0;
2558 	int ret = 0;
2559 
2560 	while ((u64)len > s) {
2561 		u64 size;
2562 		if (unlikely(ret >= iov_size)) {
2563 			ret = -ENOBUFS;
2564 			break;
2565 		}
2566 
2567 		node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
2568 							addr, addr + len - 1);
2569 		if (node == NULL || node->start > addr) {
2570 			if (umem != dev->iotlb) {
2571 				ret = -EFAULT;
2572 				break;
2573 			}
2574 			ret = -EAGAIN;
2575 			break;
2576 		} else if (!(node->perm & access)) {
2577 			ret = -EPERM;
2578 			break;
2579 		}
2580 
2581 		_iov = iov + ret;
2582 		size = node->size - addr + node->start;
2583 		_iov->iov_len = min((u64)len - s, size);
2584 		_iov->iov_base = (void __user *)(unsigned long)
2585 			(node->userspace_addr + addr - node->start);
2586 		s += size;
2587 		addr += size;
2588 		++ret;
2589 	}
2590 
2591 	if (ret == -EAGAIN)
2592 		vhost_iotlb_miss(vq, addr, access);
2593 	return ret;
2594 }
2595 
2596 /* Each buffer in the virtqueues is actually a chain of descriptors.  This
2597  * function returns the next descriptor in the chain,
2598  * or -1U if we're at the end. */
2599 static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc)
2600 {
2601 	unsigned int next;
2602 
2603 	/* If this descriptor says it doesn't chain, we're done. */
2604 	if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT)))
2605 		return -1U;
2606 
2607 	/* Check they're not leading us off end of descriptors. */
2608 	next = vhost16_to_cpu(vq, READ_ONCE(desc->next));
2609 	return next;
2610 }
2611 
2612 static int get_indirect(struct vhost_virtqueue *vq,
2613 			struct iovec iov[], unsigned int iov_size,
2614 			unsigned int *out_num, unsigned int *in_num,
2615 			struct vhost_log *log, unsigned int *log_num,
2616 			struct vring_desc *indirect)
2617 {
2618 	struct vring_desc desc;
2619 	unsigned int i = 0, count, found = 0;
2620 	u32 len = vhost32_to_cpu(vq, indirect->len);
2621 	struct iov_iter from;
2622 	int ret, access;
2623 
2624 	/* Sanity check */
2625 	if (unlikely(len % sizeof desc)) {
2626 		vq_err(vq, "Invalid length in indirect descriptor: "
2627 		       "len 0x%llx not multiple of 0x%zx\n",
2628 		       (unsigned long long)len,
2629 		       sizeof desc);
2630 		return -EINVAL;
2631 	}
2632 
2633 	ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect,
2634 			     UIO_MAXIOV, VHOST_ACCESS_RO);
2635 	if (unlikely(ret < 0)) {
2636 		if (ret != -EAGAIN)
2637 			vq_err(vq, "Translation failure %d in indirect.\n", ret);
2638 		return ret;
2639 	}
2640 	iov_iter_init(&from, READ, vq->indirect, ret, len);
2641 
2642 	/* We will use the result as an address to read from, so most
2643 	 * architectures only need a compiler barrier here. */
2644 	read_barrier_depends();
2645 
2646 	count = len / sizeof desc;
2647 	/* Buffers are chained via a 16 bit next field, so
2648 	 * we can have at most 2^16 of these. */
2649 	if (unlikely(count > USHRT_MAX + 1)) {
2650 		vq_err(vq, "Indirect buffer length too big: %d\n",
2651 		       indirect->len);
2652 		return -E2BIG;
2653 	}
2654 
2655 	do {
2656 		unsigned iov_count = *in_num + *out_num;
2657 		if (unlikely(++found > count)) {
2658 			vq_err(vq, "Loop detected: last one at %u "
2659 			       "indirect size %u\n",
2660 			       i, count);
2661 			return -EINVAL;
2662 		}
2663 		if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) {
2664 			vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n",
2665 			       i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
2666 			return -EINVAL;
2667 		}
2668 		if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) {
2669 			vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n",
2670 			       i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
2671 			return -EINVAL;
2672 		}
2673 
2674 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2675 			access = VHOST_ACCESS_WO;
2676 		else
2677 			access = VHOST_ACCESS_RO;
2678 
2679 		ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2680 				     vhost32_to_cpu(vq, desc.len), iov + iov_count,
2681 				     iov_size - iov_count, access);
2682 		if (unlikely(ret < 0)) {
2683 			if (ret != -EAGAIN)
2684 				vq_err(vq, "Translation failure %d indirect idx %d\n",
2685 					ret, i);
2686 			return ret;
2687 		}
2688 		/* If this is an input descriptor, increment that count. */
2689 		if (access == VHOST_ACCESS_WO) {
2690 			*in_num += ret;
2691 			if (unlikely(log)) {
2692 				log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2693 				log[*log_num].len = vhost32_to_cpu(vq, desc.len);
2694 				++*log_num;
2695 			}
2696 		} else {
2697 			/* If it's an output descriptor, they're all supposed
2698 			 * to come before any input descriptors. */
2699 			if (unlikely(*in_num)) {
2700 				vq_err(vq, "Indirect descriptor "
2701 				       "has out after in: idx %d\n", i);
2702 				return -EINVAL;
2703 			}
2704 			*out_num += ret;
2705 		}
2706 	} while ((i = next_desc(vq, &desc)) != -1);
2707 	return 0;
2708 }
2709 
2710 /* This looks in the virtqueue and for the first available buffer, and converts
2711  * it to an iovec for convenient access.  Since descriptors consist of some
2712  * number of output then some number of input descriptors, it's actually two
2713  * iovecs, but we pack them into one and note how many of each there were.
2714  *
2715  * This function returns the descriptor number found, or vq->num (which is
2716  * never a valid descriptor number) if none was found.  A negative code is
2717  * returned on error. */
2718 int vhost_get_vq_desc(struct vhost_virtqueue *vq,
2719 		      struct iovec iov[], unsigned int iov_size,
2720 		      unsigned int *out_num, unsigned int *in_num,
2721 		      struct vhost_log *log, unsigned int *log_num)
2722 {
2723 	struct vring_desc desc;
2724 	unsigned int i, head, found = 0;
2725 	u16 last_avail_idx;
2726 	__virtio16 avail_idx;
2727 	__virtio16 ring_head;
2728 	int ret, access;
2729 
2730 	/* Check it isn't doing very strange things with descriptor numbers. */
2731 	last_avail_idx = vq->last_avail_idx;
2732 
2733 	if (vq->avail_idx == vq->last_avail_idx) {
2734 		if (unlikely(vhost_get_avail_idx(vq, &avail_idx))) {
2735 			vq_err(vq, "Failed to access avail idx at %p\n",
2736 				&vq->avail->idx);
2737 			return -EFAULT;
2738 		}
2739 		vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
2740 
2741 		if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) {
2742 			vq_err(vq, "Guest moved used index from %u to %u",
2743 				last_avail_idx, vq->avail_idx);
2744 			return -EFAULT;
2745 		}
2746 
2747 		/* If there's nothing new since last we looked, return
2748 		 * invalid.
2749 		 */
2750 		if (vq->avail_idx == last_avail_idx)
2751 			return vq->num;
2752 
2753 		/* Only get avail ring entries after they have been
2754 		 * exposed by guest.
2755 		 */
2756 		smp_rmb();
2757 	}
2758 
2759 	/* Grab the next descriptor number they're advertising, and increment
2760 	 * the index we've seen. */
2761 	if (unlikely(vhost_get_avail_head(vq, &ring_head, last_avail_idx))) {
2762 		vq_err(vq, "Failed to read head: idx %d address %p\n",
2763 		       last_avail_idx,
2764 		       &vq->avail->ring[last_avail_idx % vq->num]);
2765 		return -EFAULT;
2766 	}
2767 
2768 	head = vhost16_to_cpu(vq, ring_head);
2769 
2770 	/* If their number is silly, that's an error. */
2771 	if (unlikely(head >= vq->num)) {
2772 		vq_err(vq, "Guest says index %u > %u is available",
2773 		       head, vq->num);
2774 		return -EINVAL;
2775 	}
2776 
2777 	/* When we start there are none of either input nor output. */
2778 	*out_num = *in_num = 0;
2779 	if (unlikely(log))
2780 		*log_num = 0;
2781 
2782 	i = head;
2783 	do {
2784 		unsigned iov_count = *in_num + *out_num;
2785 		if (unlikely(i >= vq->num)) {
2786 			vq_err(vq, "Desc index is %u > %u, head = %u",
2787 			       i, vq->num, head);
2788 			return -EINVAL;
2789 		}
2790 		if (unlikely(++found > vq->num)) {
2791 			vq_err(vq, "Loop detected: last one at %u "
2792 			       "vq size %u head %u\n",
2793 			       i, vq->num, head);
2794 			return -EINVAL;
2795 		}
2796 		ret = vhost_get_desc(vq, &desc, i);
2797 		if (unlikely(ret)) {
2798 			vq_err(vq, "Failed to get descriptor: idx %d addr %p\n",
2799 			       i, vq->desc + i);
2800 			return -EFAULT;
2801 		}
2802 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) {
2803 			ret = get_indirect(vq, iov, iov_size,
2804 					   out_num, in_num,
2805 					   log, log_num, &desc);
2806 			if (unlikely(ret < 0)) {
2807 				if (ret != -EAGAIN)
2808 					vq_err(vq, "Failure detected "
2809 						"in indirect descriptor at idx %d\n", i);
2810 				return ret;
2811 			}
2812 			continue;
2813 		}
2814 
2815 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2816 			access = VHOST_ACCESS_WO;
2817 		else
2818 			access = VHOST_ACCESS_RO;
2819 		ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2820 				     vhost32_to_cpu(vq, desc.len), iov + iov_count,
2821 				     iov_size - iov_count, access);
2822 		if (unlikely(ret < 0)) {
2823 			if (ret != -EAGAIN)
2824 				vq_err(vq, "Translation failure %d descriptor idx %d\n",
2825 					ret, i);
2826 			return ret;
2827 		}
2828 		if (access == VHOST_ACCESS_WO) {
2829 			/* If this is an input descriptor,
2830 			 * increment that count. */
2831 			*in_num += ret;
2832 			if (unlikely(log)) {
2833 				log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2834 				log[*log_num].len = vhost32_to_cpu(vq, desc.len);
2835 				++*log_num;
2836 			}
2837 		} else {
2838 			/* If it's an output descriptor, they're all supposed
2839 			 * to come before any input descriptors. */
2840 			if (unlikely(*in_num)) {
2841 				vq_err(vq, "Descriptor has out after in: "
2842 				       "idx %d\n", i);
2843 				return -EINVAL;
2844 			}
2845 			*out_num += ret;
2846 		}
2847 	} while ((i = next_desc(vq, &desc)) != -1);
2848 
2849 	/* On success, increment avail index. */
2850 	vq->last_avail_idx++;
2851 
2852 	/* Assume notifications from guest are disabled at this point,
2853 	 * if they aren't we would need to update avail_event index. */
2854 	BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY));
2855 	return head;
2856 }
2857 EXPORT_SYMBOL_GPL(vhost_get_vq_desc);
2858 
2859 /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */
2860 void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n)
2861 {
2862 	vq->last_avail_idx -= n;
2863 }
2864 EXPORT_SYMBOL_GPL(vhost_discard_vq_desc);
2865 
2866 /* After we've used one of their buffers, we tell them about it.  We'll then
2867  * want to notify the guest, using eventfd. */
2868 int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len)
2869 {
2870 	struct vring_used_elem heads = {
2871 		cpu_to_vhost32(vq, head),
2872 		cpu_to_vhost32(vq, len)
2873 	};
2874 
2875 	return vhost_add_used_n(vq, &heads, 1);
2876 }
2877 EXPORT_SYMBOL_GPL(vhost_add_used);
2878 
2879 static int __vhost_add_used_n(struct vhost_virtqueue *vq,
2880 			    struct vring_used_elem *heads,
2881 			    unsigned count)
2882 {
2883 	struct vring_used_elem __user *used;
2884 	u16 old, new;
2885 	int start;
2886 
2887 	start = vq->last_used_idx & (vq->num - 1);
2888 	used = vq->used->ring + start;
2889 	if (vhost_put_used(vq, heads, start, count)) {
2890 		vq_err(vq, "Failed to write used");
2891 		return -EFAULT;
2892 	}
2893 	if (unlikely(vq->log_used)) {
2894 		/* Make sure data is seen before log. */
2895 		smp_wmb();
2896 		/* Log used ring entry write. */
2897 		log_used(vq, ((void __user *)used - (void __user *)vq->used),
2898 			 count * sizeof *used);
2899 	}
2900 	old = vq->last_used_idx;
2901 	new = (vq->last_used_idx += count);
2902 	/* If the driver never bothers to signal in a very long while,
2903 	 * used index might wrap around. If that happens, invalidate
2904 	 * signalled_used index we stored. TODO: make sure driver
2905 	 * signals at least once in 2^16 and remove this. */
2906 	if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old)))
2907 		vq->signalled_used_valid = false;
2908 	return 0;
2909 }
2910 
2911 /* After we've used one of their buffers, we tell them about it.  We'll then
2912  * want to notify the guest, using eventfd. */
2913 int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads,
2914 		     unsigned count)
2915 {
2916 	int start, n, r;
2917 
2918 	start = vq->last_used_idx & (vq->num - 1);
2919 	n = vq->num - start;
2920 	if (n < count) {
2921 		r = __vhost_add_used_n(vq, heads, n);
2922 		if (r < 0)
2923 			return r;
2924 		heads += n;
2925 		count -= n;
2926 	}
2927 	r = __vhost_add_used_n(vq, heads, count);
2928 
2929 	/* Make sure buffer is written before we update index. */
2930 	smp_wmb();
2931 	if (vhost_put_used_idx(vq)) {
2932 		vq_err(vq, "Failed to increment used idx");
2933 		return -EFAULT;
2934 	}
2935 	if (unlikely(vq->log_used)) {
2936 		/* Make sure used idx is seen before log. */
2937 		smp_wmb();
2938 		/* Log used index update. */
2939 		log_used(vq, offsetof(struct vring_used, idx),
2940 			 sizeof vq->used->idx);
2941 		if (vq->log_ctx)
2942 			eventfd_signal(vq->log_ctx, 1);
2943 	}
2944 	return r;
2945 }
2946 EXPORT_SYMBOL_GPL(vhost_add_used_n);
2947 
2948 static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2949 {
2950 	__u16 old, new;
2951 	__virtio16 event;
2952 	bool v;
2953 	/* Flush out used index updates. This is paired
2954 	 * with the barrier that the Guest executes when enabling
2955 	 * interrupts. */
2956 	smp_mb();
2957 
2958 	if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2959 	    unlikely(vq->avail_idx == vq->last_avail_idx))
2960 		return true;
2961 
2962 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2963 		__virtio16 flags;
2964 		if (vhost_get_avail_flags(vq, &flags)) {
2965 			vq_err(vq, "Failed to get flags");
2966 			return true;
2967 		}
2968 		return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT));
2969 	}
2970 	old = vq->signalled_used;
2971 	v = vq->signalled_used_valid;
2972 	new = vq->signalled_used = vq->last_used_idx;
2973 	vq->signalled_used_valid = true;
2974 
2975 	if (unlikely(!v))
2976 		return true;
2977 
2978 	if (vhost_get_used_event(vq, &event)) {
2979 		vq_err(vq, "Failed to get used event idx");
2980 		return true;
2981 	}
2982 	return vring_need_event(vhost16_to_cpu(vq, event), new, old);
2983 }
2984 
2985 /* This actually signals the guest, using eventfd. */
2986 void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2987 {
2988 	/* Signal the Guest tell them we used something up. */
2989 	if (vq->call_ctx && vhost_notify(dev, vq))
2990 		eventfd_signal(vq->call_ctx, 1);
2991 }
2992 EXPORT_SYMBOL_GPL(vhost_signal);
2993 
2994 /* And here's the combo meal deal.  Supersize me! */
2995 void vhost_add_used_and_signal(struct vhost_dev *dev,
2996 			       struct vhost_virtqueue *vq,
2997 			       unsigned int head, int len)
2998 {
2999 	vhost_add_used(vq, head, len);
3000 	vhost_signal(dev, vq);
3001 }
3002 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal);
3003 
3004 /* multi-buffer version of vhost_add_used_and_signal */
3005 void vhost_add_used_and_signal_n(struct vhost_dev *dev,
3006 				 struct vhost_virtqueue *vq,
3007 				 struct vring_used_elem *heads, unsigned count)
3008 {
3009 	vhost_add_used_n(vq, heads, count);
3010 	vhost_signal(dev, vq);
3011 }
3012 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n);
3013 
3014 /* return true if we're sure that avaiable ring is empty */
3015 bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq)
3016 {
3017 	__virtio16 avail_idx;
3018 	int r;
3019 
3020 	if (vq->avail_idx != vq->last_avail_idx)
3021 		return false;
3022 
3023 	r = vhost_get_avail_idx(vq, &avail_idx);
3024 	if (unlikely(r))
3025 		return false;
3026 	vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
3027 
3028 	return vq->avail_idx == vq->last_avail_idx;
3029 }
3030 EXPORT_SYMBOL_GPL(vhost_vq_avail_empty);
3031 
3032 /* OK, now we need to know about added descriptors. */
3033 bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
3034 {
3035 	__virtio16 avail_idx;
3036 	int r;
3037 
3038 	if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY))
3039 		return false;
3040 	vq->used_flags &= ~VRING_USED_F_NO_NOTIFY;
3041 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
3042 		r = vhost_update_used_flags(vq);
3043 		if (r) {
3044 			vq_err(vq, "Failed to enable notification at %p: %d\n",
3045 			       &vq->used->flags, r);
3046 			return false;
3047 		}
3048 	} else {
3049 		r = vhost_update_avail_event(vq, vq->avail_idx);
3050 		if (r) {
3051 			vq_err(vq, "Failed to update avail event index at %p: %d\n",
3052 			       vhost_avail_event(vq), r);
3053 			return false;
3054 		}
3055 	}
3056 	/* They could have slipped one in as we were doing that: make
3057 	 * sure it's written, then check again. */
3058 	smp_mb();
3059 	r = vhost_get_avail_idx(vq, &avail_idx);
3060 	if (r) {
3061 		vq_err(vq, "Failed to check avail idx at %p: %d\n",
3062 		       &vq->avail->idx, r);
3063 		return false;
3064 	}
3065 
3066 	return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx;
3067 }
3068 EXPORT_SYMBOL_GPL(vhost_enable_notify);
3069 
3070 /* We don't need to be notified again. */
3071 void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
3072 {
3073 	int r;
3074 
3075 	if (vq->used_flags & VRING_USED_F_NO_NOTIFY)
3076 		return;
3077 	vq->used_flags |= VRING_USED_F_NO_NOTIFY;
3078 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
3079 		r = vhost_update_used_flags(vq);
3080 		if (r)
3081 			vq_err(vq, "Failed to enable notification at %p: %d\n",
3082 			       &vq->used->flags, r);
3083 	}
3084 }
3085 EXPORT_SYMBOL_GPL(vhost_disable_notify);
3086 
3087 /* Create a new message. */
3088 struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type)
3089 {
3090 	struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL);
3091 	if (!node)
3092 		return NULL;
3093 
3094 	/* Make sure all padding within the structure is initialized. */
3095 	memset(&node->msg, 0, sizeof node->msg);
3096 	node->vq = vq;
3097 	node->msg.type = type;
3098 	return node;
3099 }
3100 EXPORT_SYMBOL_GPL(vhost_new_msg);
3101 
3102 void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head,
3103 		       struct vhost_msg_node *node)
3104 {
3105 	spin_lock(&dev->iotlb_lock);
3106 	list_add_tail(&node->node, head);
3107 	spin_unlock(&dev->iotlb_lock);
3108 
3109 	wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
3110 }
3111 EXPORT_SYMBOL_GPL(vhost_enqueue_msg);
3112 
3113 struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
3114 					 struct list_head *head)
3115 {
3116 	struct vhost_msg_node *node = NULL;
3117 
3118 	spin_lock(&dev->iotlb_lock);
3119 	if (!list_empty(head)) {
3120 		node = list_first_entry(head, struct vhost_msg_node,
3121 					node);
3122 		list_del(&node->node);
3123 	}
3124 	spin_unlock(&dev->iotlb_lock);
3125 
3126 	return node;
3127 }
3128 EXPORT_SYMBOL_GPL(vhost_dequeue_msg);
3129 
3130 
3131 static int __init vhost_init(void)
3132 {
3133 	return 0;
3134 }
3135 
3136 static void __exit vhost_exit(void)
3137 {
3138 }
3139 
3140 module_init(vhost_init);
3141 module_exit(vhost_exit);
3142 
3143 MODULE_VERSION("0.0.1");
3144 MODULE_LICENSE("GPL v2");
3145 MODULE_AUTHOR("Michael S. Tsirkin");
3146 MODULE_DESCRIPTION("Host kernel accelerator for virtio");
3147