xref: /openbmc/linux/drivers/vhost/vhost.c (revision 77ab8d5d)
1 /* Copyright (C) 2009 Red Hat, Inc.
2  * Copyright (C) 2006 Rusty Russell IBM Corporation
3  *
4  * Author: Michael S. Tsirkin <mst@redhat.com>
5  *
6  * Inspiration, some code, and most witty comments come from
7  * Documentation/virtual/lguest/lguest.c, by Rusty Russell
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.
10  *
11  * Generic code for virtio server in host kernel.
12  */
13 
14 #include <linux/eventfd.h>
15 #include <linux/vhost.h>
16 #include <linux/uio.h>
17 #include <linux/mm.h>
18 #include <linux/mmu_context.h>
19 #include <linux/miscdevice.h>
20 #include <linux/mutex.h>
21 #include <linux/poll.h>
22 #include <linux/file.h>
23 #include <linux/highmem.h>
24 #include <linux/slab.h>
25 #include <linux/vmalloc.h>
26 #include <linux/kthread.h>
27 #include <linux/cgroup.h>
28 #include <linux/module.h>
29 #include <linux/sort.h>
30 #include <linux/sched/mm.h>
31 #include <linux/sched/signal.h>
32 #include <linux/interval_tree_generic.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 = file->f_op->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 static void vhost_vq_reset(struct vhost_dev *dev,
302 			   struct vhost_virtqueue *vq)
303 {
304 	vq->num = 1;
305 	vq->desc = NULL;
306 	vq->avail = NULL;
307 	vq->used = NULL;
308 	vq->last_avail_idx = 0;
309 	vq->avail_idx = 0;
310 	vq->last_used_idx = 0;
311 	vq->signalled_used = 0;
312 	vq->signalled_used_valid = false;
313 	vq->used_flags = 0;
314 	vq->log_used = false;
315 	vq->log_addr = -1ull;
316 	vq->private_data = NULL;
317 	vq->acked_features = 0;
318 	vq->log_base = NULL;
319 	vq->error_ctx = NULL;
320 	vq->kick = NULL;
321 	vq->call_ctx = NULL;
322 	vq->log_ctx = NULL;
323 	vhost_reset_is_le(vq);
324 	vhost_disable_cross_endian(vq);
325 	vq->busyloop_timeout = 0;
326 	vq->umem = NULL;
327 	vq->iotlb = NULL;
328 	__vhost_vq_meta_reset(vq);
329 }
330 
331 static int vhost_worker(void *data)
332 {
333 	struct vhost_dev *dev = data;
334 	struct vhost_work *work, *work_next;
335 	struct llist_node *node;
336 	mm_segment_t oldfs = get_fs();
337 
338 	set_fs(USER_DS);
339 	use_mm(dev->mm);
340 
341 	for (;;) {
342 		/* mb paired w/ kthread_stop */
343 		set_current_state(TASK_INTERRUPTIBLE);
344 
345 		if (kthread_should_stop()) {
346 			__set_current_state(TASK_RUNNING);
347 			break;
348 		}
349 
350 		node = llist_del_all(&dev->work_list);
351 		if (!node)
352 			schedule();
353 
354 		node = llist_reverse_order(node);
355 		/* make sure flag is seen after deletion */
356 		smp_wmb();
357 		llist_for_each_entry_safe(work, work_next, node, node) {
358 			clear_bit(VHOST_WORK_QUEUED, &work->flags);
359 			__set_current_state(TASK_RUNNING);
360 			work->fn(work);
361 			if (need_resched())
362 				schedule();
363 		}
364 	}
365 	unuse_mm(dev->mm);
366 	set_fs(oldfs);
367 	return 0;
368 }
369 
370 static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq)
371 {
372 	kfree(vq->indirect);
373 	vq->indirect = NULL;
374 	kfree(vq->log);
375 	vq->log = NULL;
376 	kfree(vq->heads);
377 	vq->heads = NULL;
378 }
379 
380 /* Helper to allocate iovec buffers for all vqs. */
381 static long vhost_dev_alloc_iovecs(struct vhost_dev *dev)
382 {
383 	struct vhost_virtqueue *vq;
384 	int i;
385 
386 	for (i = 0; i < dev->nvqs; ++i) {
387 		vq = dev->vqs[i];
388 		vq->indirect = kmalloc(sizeof *vq->indirect * UIO_MAXIOV,
389 				       GFP_KERNEL);
390 		vq->log = kmalloc(sizeof *vq->log * UIO_MAXIOV, GFP_KERNEL);
391 		vq->heads = kmalloc(sizeof *vq->heads * UIO_MAXIOV, GFP_KERNEL);
392 		if (!vq->indirect || !vq->log || !vq->heads)
393 			goto err_nomem;
394 	}
395 	return 0;
396 
397 err_nomem:
398 	for (; i >= 0; --i)
399 		vhost_vq_free_iovecs(dev->vqs[i]);
400 	return -ENOMEM;
401 }
402 
403 static void vhost_dev_free_iovecs(struct vhost_dev *dev)
404 {
405 	int i;
406 
407 	for (i = 0; i < dev->nvqs; ++i)
408 		vhost_vq_free_iovecs(dev->vqs[i]);
409 }
410 
411 void vhost_dev_init(struct vhost_dev *dev,
412 		    struct vhost_virtqueue **vqs, int nvqs)
413 {
414 	struct vhost_virtqueue *vq;
415 	int i;
416 
417 	dev->vqs = vqs;
418 	dev->nvqs = nvqs;
419 	mutex_init(&dev->mutex);
420 	dev->log_ctx = NULL;
421 	dev->umem = NULL;
422 	dev->iotlb = NULL;
423 	dev->mm = NULL;
424 	dev->worker = NULL;
425 	init_llist_head(&dev->work_list);
426 	init_waitqueue_head(&dev->wait);
427 	INIT_LIST_HEAD(&dev->read_list);
428 	INIT_LIST_HEAD(&dev->pending_list);
429 	spin_lock_init(&dev->iotlb_lock);
430 
431 
432 	for (i = 0; i < dev->nvqs; ++i) {
433 		vq = dev->vqs[i];
434 		vq->log = NULL;
435 		vq->indirect = NULL;
436 		vq->heads = NULL;
437 		vq->dev = dev;
438 		mutex_init(&vq->mutex);
439 		vhost_vq_reset(dev, vq);
440 		if (vq->handle_kick)
441 			vhost_poll_init(&vq->poll, vq->handle_kick,
442 					EPOLLIN, dev);
443 	}
444 }
445 EXPORT_SYMBOL_GPL(vhost_dev_init);
446 
447 /* Caller should have device mutex */
448 long vhost_dev_check_owner(struct vhost_dev *dev)
449 {
450 	/* Are you the owner? If not, I don't think you mean to do that */
451 	return dev->mm == current->mm ? 0 : -EPERM;
452 }
453 EXPORT_SYMBOL_GPL(vhost_dev_check_owner);
454 
455 struct vhost_attach_cgroups_struct {
456 	struct vhost_work work;
457 	struct task_struct *owner;
458 	int ret;
459 };
460 
461 static void vhost_attach_cgroups_work(struct vhost_work *work)
462 {
463 	struct vhost_attach_cgroups_struct *s;
464 
465 	s = container_of(work, struct vhost_attach_cgroups_struct, work);
466 	s->ret = cgroup_attach_task_all(s->owner, current);
467 }
468 
469 static int vhost_attach_cgroups(struct vhost_dev *dev)
470 {
471 	struct vhost_attach_cgroups_struct attach;
472 
473 	attach.owner = current;
474 	vhost_work_init(&attach.work, vhost_attach_cgroups_work);
475 	vhost_work_queue(dev, &attach.work);
476 	vhost_work_flush(dev, &attach.work);
477 	return attach.ret;
478 }
479 
480 /* Caller should have device mutex */
481 bool vhost_dev_has_owner(struct vhost_dev *dev)
482 {
483 	return dev->mm;
484 }
485 EXPORT_SYMBOL_GPL(vhost_dev_has_owner);
486 
487 /* Caller should have device mutex */
488 long vhost_dev_set_owner(struct vhost_dev *dev)
489 {
490 	struct task_struct *worker;
491 	int err;
492 
493 	/* Is there an owner already? */
494 	if (vhost_dev_has_owner(dev)) {
495 		err = -EBUSY;
496 		goto err_mm;
497 	}
498 
499 	/* No owner, become one */
500 	dev->mm = get_task_mm(current);
501 	worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid);
502 	if (IS_ERR(worker)) {
503 		err = PTR_ERR(worker);
504 		goto err_worker;
505 	}
506 
507 	dev->worker = worker;
508 	wake_up_process(worker);	/* avoid contributing to loadavg */
509 
510 	err = vhost_attach_cgroups(dev);
511 	if (err)
512 		goto err_cgroup;
513 
514 	err = vhost_dev_alloc_iovecs(dev);
515 	if (err)
516 		goto err_cgroup;
517 
518 	return 0;
519 err_cgroup:
520 	kthread_stop(worker);
521 	dev->worker = NULL;
522 err_worker:
523 	if (dev->mm)
524 		mmput(dev->mm);
525 	dev->mm = NULL;
526 err_mm:
527 	return err;
528 }
529 EXPORT_SYMBOL_GPL(vhost_dev_set_owner);
530 
531 struct vhost_umem *vhost_dev_reset_owner_prepare(void)
532 {
533 	return kvzalloc(sizeof(struct vhost_umem), GFP_KERNEL);
534 }
535 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare);
536 
537 /* Caller should have device mutex */
538 void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_umem *umem)
539 {
540 	int i;
541 
542 	vhost_dev_cleanup(dev);
543 
544 	/* Restore memory to default empty mapping. */
545 	INIT_LIST_HEAD(&umem->umem_list);
546 	dev->umem = umem;
547 	/* We don't need VQ locks below since vhost_dev_cleanup makes sure
548 	 * VQs aren't running.
549 	 */
550 	for (i = 0; i < dev->nvqs; ++i)
551 		dev->vqs[i]->umem = umem;
552 }
553 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner);
554 
555 void vhost_dev_stop(struct vhost_dev *dev)
556 {
557 	int i;
558 
559 	for (i = 0; i < dev->nvqs; ++i) {
560 		if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) {
561 			vhost_poll_stop(&dev->vqs[i]->poll);
562 			vhost_poll_flush(&dev->vqs[i]->poll);
563 		}
564 	}
565 }
566 EXPORT_SYMBOL_GPL(vhost_dev_stop);
567 
568 static void vhost_umem_free(struct vhost_umem *umem,
569 			    struct vhost_umem_node *node)
570 {
571 	vhost_umem_interval_tree_remove(node, &umem->umem_tree);
572 	list_del(&node->link);
573 	kfree(node);
574 	umem->numem--;
575 }
576 
577 static void vhost_umem_clean(struct vhost_umem *umem)
578 {
579 	struct vhost_umem_node *node, *tmp;
580 
581 	if (!umem)
582 		return;
583 
584 	list_for_each_entry_safe(node, tmp, &umem->umem_list, link)
585 		vhost_umem_free(umem, node);
586 
587 	kvfree(umem);
588 }
589 
590 static void vhost_clear_msg(struct vhost_dev *dev)
591 {
592 	struct vhost_msg_node *node, *n;
593 
594 	spin_lock(&dev->iotlb_lock);
595 
596 	list_for_each_entry_safe(node, n, &dev->read_list, node) {
597 		list_del(&node->node);
598 		kfree(node);
599 	}
600 
601 	list_for_each_entry_safe(node, n, &dev->pending_list, node) {
602 		list_del(&node->node);
603 		kfree(node);
604 	}
605 
606 	spin_unlock(&dev->iotlb_lock);
607 }
608 
609 void vhost_dev_cleanup(struct vhost_dev *dev)
610 {
611 	int i;
612 
613 	for (i = 0; i < dev->nvqs; ++i) {
614 		if (dev->vqs[i]->error_ctx)
615 			eventfd_ctx_put(dev->vqs[i]->error_ctx);
616 		if (dev->vqs[i]->kick)
617 			fput(dev->vqs[i]->kick);
618 		if (dev->vqs[i]->call_ctx)
619 			eventfd_ctx_put(dev->vqs[i]->call_ctx);
620 		vhost_vq_reset(dev, dev->vqs[i]);
621 	}
622 	vhost_dev_free_iovecs(dev);
623 	if (dev->log_ctx)
624 		eventfd_ctx_put(dev->log_ctx);
625 	dev->log_ctx = NULL;
626 	/* No one will access memory at this point */
627 	vhost_umem_clean(dev->umem);
628 	dev->umem = NULL;
629 	vhost_umem_clean(dev->iotlb);
630 	dev->iotlb = NULL;
631 	vhost_clear_msg(dev);
632 	wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
633 	WARN_ON(!llist_empty(&dev->work_list));
634 	if (dev->worker) {
635 		kthread_stop(dev->worker);
636 		dev->worker = NULL;
637 	}
638 	if (dev->mm)
639 		mmput(dev->mm);
640 	dev->mm = NULL;
641 }
642 EXPORT_SYMBOL_GPL(vhost_dev_cleanup);
643 
644 static bool log_access_ok(void __user *log_base, u64 addr, unsigned long sz)
645 {
646 	u64 a = addr / VHOST_PAGE_SIZE / 8;
647 
648 	/* Make sure 64 bit math will not overflow. */
649 	if (a > ULONG_MAX - (unsigned long)log_base ||
650 	    a + (unsigned long)log_base > ULONG_MAX)
651 		return false;
652 
653 	return access_ok(VERIFY_WRITE, log_base + a,
654 			 (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8);
655 }
656 
657 static bool vhost_overflow(u64 uaddr, u64 size)
658 {
659 	/* Make sure 64 bit math will not overflow. */
660 	return uaddr > ULONG_MAX || size > ULONG_MAX || uaddr > ULONG_MAX - size;
661 }
662 
663 /* Caller should have vq mutex and device mutex. */
664 static bool vq_memory_access_ok(void __user *log_base, struct vhost_umem *umem,
665 				int log_all)
666 {
667 	struct vhost_umem_node *node;
668 
669 	if (!umem)
670 		return false;
671 
672 	list_for_each_entry(node, &umem->umem_list, link) {
673 		unsigned long a = node->userspace_addr;
674 
675 		if (vhost_overflow(node->userspace_addr, node->size))
676 			return false;
677 
678 
679 		if (!access_ok(VERIFY_WRITE, (void __user *)a,
680 				    node->size))
681 			return false;
682 		else if (log_all && !log_access_ok(log_base,
683 						   node->start,
684 						   node->size))
685 			return false;
686 	}
687 	return true;
688 }
689 
690 static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq,
691 					       u64 addr, unsigned int size,
692 					       int type)
693 {
694 	const struct vhost_umem_node *node = vq->meta_iotlb[type];
695 
696 	if (!node)
697 		return NULL;
698 
699 	return (void *)(uintptr_t)(node->userspace_addr + addr - node->start);
700 }
701 
702 /* Can we switch to this memory table? */
703 /* Caller should have device mutex but not vq mutex */
704 static bool memory_access_ok(struct vhost_dev *d, struct vhost_umem *umem,
705 			     int log_all)
706 {
707 	int i;
708 
709 	for (i = 0; i < d->nvqs; ++i) {
710 		bool ok;
711 		bool log;
712 
713 		mutex_lock(&d->vqs[i]->mutex);
714 		log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL);
715 		/* If ring is inactive, will check when it's enabled. */
716 		if (d->vqs[i]->private_data)
717 			ok = vq_memory_access_ok(d->vqs[i]->log_base,
718 						 umem, log);
719 		else
720 			ok = true;
721 		mutex_unlock(&d->vqs[i]->mutex);
722 		if (!ok)
723 			return false;
724 	}
725 	return true;
726 }
727 
728 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
729 			  struct iovec iov[], int iov_size, int access);
730 
731 static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to,
732 			      const void *from, unsigned size)
733 {
734 	int ret;
735 
736 	if (!vq->iotlb)
737 		return __copy_to_user(to, from, size);
738 	else {
739 		/* This function should be called after iotlb
740 		 * prefetch, which means we're sure that all vq
741 		 * could be access through iotlb. So -EAGAIN should
742 		 * not happen in this case.
743 		 */
744 		struct iov_iter t;
745 		void __user *uaddr = vhost_vq_meta_fetch(vq,
746 				     (u64)(uintptr_t)to, size,
747 				     VHOST_ADDR_USED);
748 
749 		if (uaddr)
750 			return __copy_to_user(uaddr, from, size);
751 
752 		ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov,
753 				     ARRAY_SIZE(vq->iotlb_iov),
754 				     VHOST_ACCESS_WO);
755 		if (ret < 0)
756 			goto out;
757 		iov_iter_init(&t, WRITE, vq->iotlb_iov, ret, size);
758 		ret = copy_to_iter(from, size, &t);
759 		if (ret == size)
760 			ret = 0;
761 	}
762 out:
763 	return ret;
764 }
765 
766 static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to,
767 				void __user *from, unsigned size)
768 {
769 	int ret;
770 
771 	if (!vq->iotlb)
772 		return __copy_from_user(to, from, size);
773 	else {
774 		/* This function should be called after iotlb
775 		 * prefetch, which means we're sure that vq
776 		 * could be access through iotlb. So -EAGAIN should
777 		 * not happen in this case.
778 		 */
779 		void __user *uaddr = vhost_vq_meta_fetch(vq,
780 				     (u64)(uintptr_t)from, size,
781 				     VHOST_ADDR_DESC);
782 		struct iov_iter f;
783 
784 		if (uaddr)
785 			return __copy_from_user(to, uaddr, size);
786 
787 		ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov,
788 				     ARRAY_SIZE(vq->iotlb_iov),
789 				     VHOST_ACCESS_RO);
790 		if (ret < 0) {
791 			vq_err(vq, "IOTLB translation failure: uaddr "
792 			       "%p size 0x%llx\n", from,
793 			       (unsigned long long) size);
794 			goto out;
795 		}
796 		iov_iter_init(&f, READ, vq->iotlb_iov, ret, size);
797 		ret = copy_from_iter(to, size, &f);
798 		if (ret == size)
799 			ret = 0;
800 	}
801 
802 out:
803 	return ret;
804 }
805 
806 static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq,
807 					  void __user *addr, unsigned int size,
808 					  int type)
809 {
810 	int ret;
811 
812 	ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov,
813 			     ARRAY_SIZE(vq->iotlb_iov),
814 			     VHOST_ACCESS_RO);
815 	if (ret < 0) {
816 		vq_err(vq, "IOTLB translation failure: uaddr "
817 			"%p size 0x%llx\n", addr,
818 			(unsigned long long) size);
819 		return NULL;
820 	}
821 
822 	if (ret != 1 || vq->iotlb_iov[0].iov_len != size) {
823 		vq_err(vq, "Non atomic userspace memory access: uaddr "
824 			"%p size 0x%llx\n", addr,
825 			(unsigned long long) size);
826 		return NULL;
827 	}
828 
829 	return vq->iotlb_iov[0].iov_base;
830 }
831 
832 /* This function should be called after iotlb
833  * prefetch, which means we're sure that vq
834  * could be access through iotlb. So -EAGAIN should
835  * not happen in this case.
836  */
837 static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq,
838 					    void *addr, unsigned int size,
839 					    int type)
840 {
841 	void __user *uaddr = vhost_vq_meta_fetch(vq,
842 			     (u64)(uintptr_t)addr, size, type);
843 	if (uaddr)
844 		return uaddr;
845 
846 	return __vhost_get_user_slow(vq, addr, size, type);
847 }
848 
849 #define vhost_put_user(vq, x, ptr)		\
850 ({ \
851 	int ret = -EFAULT; \
852 	if (!vq->iotlb) { \
853 		ret = __put_user(x, ptr); \
854 	} else { \
855 		__typeof__(ptr) to = \
856 			(__typeof__(ptr)) __vhost_get_user(vq, ptr,	\
857 					  sizeof(*ptr), VHOST_ADDR_USED); \
858 		if (to != NULL) \
859 			ret = __put_user(x, to); \
860 		else \
861 			ret = -EFAULT;	\
862 	} \
863 	ret; \
864 })
865 
866 #define vhost_get_user(vq, x, ptr, type)		\
867 ({ \
868 	int ret; \
869 	if (!vq->iotlb) { \
870 		ret = __get_user(x, ptr); \
871 	} else { \
872 		__typeof__(ptr) from = \
873 			(__typeof__(ptr)) __vhost_get_user(vq, ptr, \
874 							   sizeof(*ptr), \
875 							   type); \
876 		if (from != NULL) \
877 			ret = __get_user(x, from); \
878 		else \
879 			ret = -EFAULT; \
880 	} \
881 	ret; \
882 })
883 
884 #define vhost_get_avail(vq, x, ptr) \
885 	vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL)
886 
887 #define vhost_get_used(vq, x, ptr) \
888 	vhost_get_user(vq, x, ptr, VHOST_ADDR_USED)
889 
890 static void vhost_dev_lock_vqs(struct vhost_dev *d)
891 {
892 	int i = 0;
893 	for (i = 0; i < d->nvqs; ++i)
894 		mutex_lock_nested(&d->vqs[i]->mutex, i);
895 }
896 
897 static void vhost_dev_unlock_vqs(struct vhost_dev *d)
898 {
899 	int i = 0;
900 	for (i = 0; i < d->nvqs; ++i)
901 		mutex_unlock(&d->vqs[i]->mutex);
902 }
903 
904 static int vhost_new_umem_range(struct vhost_umem *umem,
905 				u64 start, u64 size, u64 end,
906 				u64 userspace_addr, int perm)
907 {
908 	struct vhost_umem_node *tmp, *node = kmalloc(sizeof(*node), GFP_ATOMIC);
909 
910 	if (!node)
911 		return -ENOMEM;
912 
913 	if (umem->numem == max_iotlb_entries) {
914 		tmp = list_first_entry(&umem->umem_list, typeof(*tmp), link);
915 		vhost_umem_free(umem, tmp);
916 	}
917 
918 	node->start = start;
919 	node->size = size;
920 	node->last = end;
921 	node->userspace_addr = userspace_addr;
922 	node->perm = perm;
923 	INIT_LIST_HEAD(&node->link);
924 	list_add_tail(&node->link, &umem->umem_list);
925 	vhost_umem_interval_tree_insert(node, &umem->umem_tree);
926 	umem->numem++;
927 
928 	return 0;
929 }
930 
931 static void vhost_del_umem_range(struct vhost_umem *umem,
932 				 u64 start, u64 end)
933 {
934 	struct vhost_umem_node *node;
935 
936 	while ((node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
937 							   start, end)))
938 		vhost_umem_free(umem, node);
939 }
940 
941 static void vhost_iotlb_notify_vq(struct vhost_dev *d,
942 				  struct vhost_iotlb_msg *msg)
943 {
944 	struct vhost_msg_node *node, *n;
945 
946 	spin_lock(&d->iotlb_lock);
947 
948 	list_for_each_entry_safe(node, n, &d->pending_list, node) {
949 		struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb;
950 		if (msg->iova <= vq_msg->iova &&
951 		    msg->iova + msg->size - 1 > vq_msg->iova &&
952 		    vq_msg->type == VHOST_IOTLB_MISS) {
953 			vhost_poll_queue(&node->vq->poll);
954 			list_del(&node->node);
955 			kfree(node);
956 		}
957 	}
958 
959 	spin_unlock(&d->iotlb_lock);
960 }
961 
962 static bool umem_access_ok(u64 uaddr, u64 size, int access)
963 {
964 	unsigned long a = uaddr;
965 
966 	/* Make sure 64 bit math will not overflow. */
967 	if (vhost_overflow(uaddr, size))
968 		return false;
969 
970 	if ((access & VHOST_ACCESS_RO) &&
971 	    !access_ok(VERIFY_READ, (void __user *)a, size))
972 		return false;
973 	if ((access & VHOST_ACCESS_WO) &&
974 	    !access_ok(VERIFY_WRITE, (void __user *)a, size))
975 		return false;
976 	return true;
977 }
978 
979 static int vhost_process_iotlb_msg(struct vhost_dev *dev,
980 				   struct vhost_iotlb_msg *msg)
981 {
982 	int ret = 0;
983 
984 	mutex_lock(&dev->mutex);
985 	vhost_dev_lock_vqs(dev);
986 	switch (msg->type) {
987 	case VHOST_IOTLB_UPDATE:
988 		if (!dev->iotlb) {
989 			ret = -EFAULT;
990 			break;
991 		}
992 		if (!umem_access_ok(msg->uaddr, msg->size, msg->perm)) {
993 			ret = -EFAULT;
994 			break;
995 		}
996 		vhost_vq_meta_reset(dev);
997 		if (vhost_new_umem_range(dev->iotlb, msg->iova, msg->size,
998 					 msg->iova + msg->size - 1,
999 					 msg->uaddr, msg->perm)) {
1000 			ret = -ENOMEM;
1001 			break;
1002 		}
1003 		vhost_iotlb_notify_vq(dev, msg);
1004 		break;
1005 	case VHOST_IOTLB_INVALIDATE:
1006 		if (!dev->iotlb) {
1007 			ret = -EFAULT;
1008 			break;
1009 		}
1010 		vhost_vq_meta_reset(dev);
1011 		vhost_del_umem_range(dev->iotlb, msg->iova,
1012 				     msg->iova + msg->size - 1);
1013 		break;
1014 	default:
1015 		ret = -EINVAL;
1016 		break;
1017 	}
1018 
1019 	vhost_dev_unlock_vqs(dev);
1020 	mutex_unlock(&dev->mutex);
1021 
1022 	return ret;
1023 }
1024 ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
1025 			     struct iov_iter *from)
1026 {
1027 	struct vhost_msg_node node;
1028 	unsigned size = sizeof(struct vhost_msg);
1029 	size_t ret;
1030 	int err;
1031 
1032 	if (iov_iter_count(from) < size)
1033 		return 0;
1034 	ret = copy_from_iter(&node.msg, size, from);
1035 	if (ret != size)
1036 		goto done;
1037 
1038 	switch (node.msg.type) {
1039 	case VHOST_IOTLB_MSG:
1040 		err = vhost_process_iotlb_msg(dev, &node.msg.iotlb);
1041 		if (err)
1042 			ret = err;
1043 		break;
1044 	default:
1045 		ret = -EINVAL;
1046 		break;
1047 	}
1048 
1049 done:
1050 	return ret;
1051 }
1052 EXPORT_SYMBOL(vhost_chr_write_iter);
1053 
1054 __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
1055 			    poll_table *wait)
1056 {
1057 	__poll_t mask = 0;
1058 
1059 	poll_wait(file, &dev->wait, wait);
1060 
1061 	if (!list_empty(&dev->read_list))
1062 		mask |= EPOLLIN | EPOLLRDNORM;
1063 
1064 	return mask;
1065 }
1066 EXPORT_SYMBOL(vhost_chr_poll);
1067 
1068 ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
1069 			    int noblock)
1070 {
1071 	DEFINE_WAIT(wait);
1072 	struct vhost_msg_node *node;
1073 	ssize_t ret = 0;
1074 	unsigned size = sizeof(struct vhost_msg);
1075 
1076 	if (iov_iter_count(to) < size)
1077 		return 0;
1078 
1079 	while (1) {
1080 		if (!noblock)
1081 			prepare_to_wait(&dev->wait, &wait,
1082 					TASK_INTERRUPTIBLE);
1083 
1084 		node = vhost_dequeue_msg(dev, &dev->read_list);
1085 		if (node)
1086 			break;
1087 		if (noblock) {
1088 			ret = -EAGAIN;
1089 			break;
1090 		}
1091 		if (signal_pending(current)) {
1092 			ret = -ERESTARTSYS;
1093 			break;
1094 		}
1095 		if (!dev->iotlb) {
1096 			ret = -EBADFD;
1097 			break;
1098 		}
1099 
1100 		schedule();
1101 	}
1102 
1103 	if (!noblock)
1104 		finish_wait(&dev->wait, &wait);
1105 
1106 	if (node) {
1107 		ret = copy_to_iter(&node->msg, size, to);
1108 
1109 		if (ret != size || node->msg.type != VHOST_IOTLB_MISS) {
1110 			kfree(node);
1111 			return ret;
1112 		}
1113 
1114 		vhost_enqueue_msg(dev, &dev->pending_list, node);
1115 	}
1116 
1117 	return ret;
1118 }
1119 EXPORT_SYMBOL_GPL(vhost_chr_read_iter);
1120 
1121 static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access)
1122 {
1123 	struct vhost_dev *dev = vq->dev;
1124 	struct vhost_msg_node *node;
1125 	struct vhost_iotlb_msg *msg;
1126 
1127 	node = vhost_new_msg(vq, VHOST_IOTLB_MISS);
1128 	if (!node)
1129 		return -ENOMEM;
1130 
1131 	msg = &node->msg.iotlb;
1132 	msg->type = VHOST_IOTLB_MISS;
1133 	msg->iova = iova;
1134 	msg->perm = access;
1135 
1136 	vhost_enqueue_msg(dev, &dev->read_list, node);
1137 
1138 	return 0;
1139 }
1140 
1141 static bool vq_access_ok(struct vhost_virtqueue *vq, unsigned int num,
1142 			 struct vring_desc __user *desc,
1143 			 struct vring_avail __user *avail,
1144 			 struct vring_used __user *used)
1145 
1146 {
1147 	size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
1148 
1149 	return access_ok(VERIFY_READ, desc, num * sizeof *desc) &&
1150 	       access_ok(VERIFY_READ, avail,
1151 			 sizeof *avail + num * sizeof *avail->ring + s) &&
1152 	       access_ok(VERIFY_WRITE, used,
1153 			sizeof *used + num * sizeof *used->ring + s);
1154 }
1155 
1156 static void vhost_vq_meta_update(struct vhost_virtqueue *vq,
1157 				 const struct vhost_umem_node *node,
1158 				 int type)
1159 {
1160 	int access = (type == VHOST_ADDR_USED) ?
1161 		     VHOST_ACCESS_WO : VHOST_ACCESS_RO;
1162 
1163 	if (likely(node->perm & access))
1164 		vq->meta_iotlb[type] = node;
1165 }
1166 
1167 static bool iotlb_access_ok(struct vhost_virtqueue *vq,
1168 			    int access, u64 addr, u64 len, int type)
1169 {
1170 	const struct vhost_umem_node *node;
1171 	struct vhost_umem *umem = vq->iotlb;
1172 	u64 s = 0, size, orig_addr = addr, last = addr + len - 1;
1173 
1174 	if (vhost_vq_meta_fetch(vq, addr, len, type))
1175 		return true;
1176 
1177 	while (len > s) {
1178 		node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1179 							   addr,
1180 							   last);
1181 		if (node == NULL || node->start > addr) {
1182 			vhost_iotlb_miss(vq, addr, access);
1183 			return false;
1184 		} else if (!(node->perm & access)) {
1185 			/* Report the possible access violation by
1186 			 * request another translation from userspace.
1187 			 */
1188 			return false;
1189 		}
1190 
1191 		size = node->size - addr + node->start;
1192 
1193 		if (orig_addr == addr && size >= len)
1194 			vhost_vq_meta_update(vq, node, type);
1195 
1196 		s += size;
1197 		addr += size;
1198 	}
1199 
1200 	return true;
1201 }
1202 
1203 int vq_iotlb_prefetch(struct vhost_virtqueue *vq)
1204 {
1205 	size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
1206 	unsigned int num = vq->num;
1207 
1208 	if (!vq->iotlb)
1209 		return 1;
1210 
1211 	return iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->desc,
1212 			       num * sizeof(*vq->desc), VHOST_ADDR_DESC) &&
1213 	       iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->avail,
1214 			       sizeof *vq->avail +
1215 			       num * sizeof(*vq->avail->ring) + s,
1216 			       VHOST_ADDR_AVAIL) &&
1217 	       iotlb_access_ok(vq, VHOST_ACCESS_WO, (u64)(uintptr_t)vq->used,
1218 			       sizeof *vq->used +
1219 			       num * sizeof(*vq->used->ring) + s,
1220 			       VHOST_ADDR_USED);
1221 }
1222 EXPORT_SYMBOL_GPL(vq_iotlb_prefetch);
1223 
1224 /* Can we log writes? */
1225 /* Caller should have device mutex but not vq mutex */
1226 bool vhost_log_access_ok(struct vhost_dev *dev)
1227 {
1228 	return memory_access_ok(dev, dev->umem, 1);
1229 }
1230 EXPORT_SYMBOL_GPL(vhost_log_access_ok);
1231 
1232 /* Verify access for write logging. */
1233 /* Caller should have vq mutex and device mutex */
1234 static bool vq_log_access_ok(struct vhost_virtqueue *vq,
1235 			     void __user *log_base)
1236 {
1237 	size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
1238 
1239 	return vq_memory_access_ok(log_base, vq->umem,
1240 				   vhost_has_feature(vq, VHOST_F_LOG_ALL)) &&
1241 		(!vq->log_used || log_access_ok(log_base, vq->log_addr,
1242 					sizeof *vq->used +
1243 					vq->num * sizeof *vq->used->ring + s));
1244 }
1245 
1246 /* Can we start vq? */
1247 /* Caller should have vq mutex and device mutex */
1248 bool vhost_vq_access_ok(struct vhost_virtqueue *vq)
1249 {
1250 	if (!vq_log_access_ok(vq, vq->log_base))
1251 		return false;
1252 
1253 	/* Access validation occurs at prefetch time with IOTLB */
1254 	if (vq->iotlb)
1255 		return true;
1256 
1257 	return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used);
1258 }
1259 EXPORT_SYMBOL_GPL(vhost_vq_access_ok);
1260 
1261 static struct vhost_umem *vhost_umem_alloc(void)
1262 {
1263 	struct vhost_umem *umem = kvzalloc(sizeof(*umem), GFP_KERNEL);
1264 
1265 	if (!umem)
1266 		return NULL;
1267 
1268 	umem->umem_tree = RB_ROOT_CACHED;
1269 	umem->numem = 0;
1270 	INIT_LIST_HEAD(&umem->umem_list);
1271 
1272 	return umem;
1273 }
1274 
1275 static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)
1276 {
1277 	struct vhost_memory mem, *newmem;
1278 	struct vhost_memory_region *region;
1279 	struct vhost_umem *newumem, *oldumem;
1280 	unsigned long size = offsetof(struct vhost_memory, regions);
1281 	int i;
1282 
1283 	if (copy_from_user(&mem, m, size))
1284 		return -EFAULT;
1285 	if (mem.padding)
1286 		return -EOPNOTSUPP;
1287 	if (mem.nregions > max_mem_regions)
1288 		return -E2BIG;
1289 	newmem = kvzalloc(size + mem.nregions * sizeof(*m->regions), GFP_KERNEL);
1290 	if (!newmem)
1291 		return -ENOMEM;
1292 
1293 	memcpy(newmem, &mem, size);
1294 	if (copy_from_user(newmem->regions, m->regions,
1295 			   mem.nregions * sizeof *m->regions)) {
1296 		kvfree(newmem);
1297 		return -EFAULT;
1298 	}
1299 
1300 	newumem = vhost_umem_alloc();
1301 	if (!newumem) {
1302 		kvfree(newmem);
1303 		return -ENOMEM;
1304 	}
1305 
1306 	for (region = newmem->regions;
1307 	     region < newmem->regions + mem.nregions;
1308 	     region++) {
1309 		if (vhost_new_umem_range(newumem,
1310 					 region->guest_phys_addr,
1311 					 region->memory_size,
1312 					 region->guest_phys_addr +
1313 					 region->memory_size - 1,
1314 					 region->userspace_addr,
1315 					 VHOST_ACCESS_RW))
1316 			goto err;
1317 	}
1318 
1319 	if (!memory_access_ok(d, newumem, 0))
1320 		goto err;
1321 
1322 	oldumem = d->umem;
1323 	d->umem = newumem;
1324 
1325 	/* All memory accesses are done under some VQ mutex. */
1326 	for (i = 0; i < d->nvqs; ++i) {
1327 		mutex_lock(&d->vqs[i]->mutex);
1328 		d->vqs[i]->umem = newumem;
1329 		mutex_unlock(&d->vqs[i]->mutex);
1330 	}
1331 
1332 	kvfree(newmem);
1333 	vhost_umem_clean(oldumem);
1334 	return 0;
1335 
1336 err:
1337 	vhost_umem_clean(newumem);
1338 	kvfree(newmem);
1339 	return -EFAULT;
1340 }
1341 
1342 long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
1343 {
1344 	struct file *eventfp, *filep = NULL;
1345 	bool pollstart = false, pollstop = false;
1346 	struct eventfd_ctx *ctx = NULL;
1347 	u32 __user *idxp = argp;
1348 	struct vhost_virtqueue *vq;
1349 	struct vhost_vring_state s;
1350 	struct vhost_vring_file f;
1351 	struct vhost_vring_addr a;
1352 	u32 idx;
1353 	long r;
1354 
1355 	r = get_user(idx, idxp);
1356 	if (r < 0)
1357 		return r;
1358 	if (idx >= d->nvqs)
1359 		return -ENOBUFS;
1360 
1361 	vq = d->vqs[idx];
1362 
1363 	mutex_lock(&vq->mutex);
1364 
1365 	switch (ioctl) {
1366 	case VHOST_SET_VRING_NUM:
1367 		/* Resizing ring with an active backend?
1368 		 * You don't want to do that. */
1369 		if (vq->private_data) {
1370 			r = -EBUSY;
1371 			break;
1372 		}
1373 		if (copy_from_user(&s, argp, sizeof s)) {
1374 			r = -EFAULT;
1375 			break;
1376 		}
1377 		if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) {
1378 			r = -EINVAL;
1379 			break;
1380 		}
1381 		vq->num = s.num;
1382 		break;
1383 	case VHOST_SET_VRING_BASE:
1384 		/* Moving base with an active backend?
1385 		 * You don't want to do that. */
1386 		if (vq->private_data) {
1387 			r = -EBUSY;
1388 			break;
1389 		}
1390 		if (copy_from_user(&s, argp, sizeof s)) {
1391 			r = -EFAULT;
1392 			break;
1393 		}
1394 		if (s.num > 0xffff) {
1395 			r = -EINVAL;
1396 			break;
1397 		}
1398 		vq->last_avail_idx = s.num;
1399 		/* Forget the cached index value. */
1400 		vq->avail_idx = vq->last_avail_idx;
1401 		break;
1402 	case VHOST_GET_VRING_BASE:
1403 		s.index = idx;
1404 		s.num = vq->last_avail_idx;
1405 		if (copy_to_user(argp, &s, sizeof s))
1406 			r = -EFAULT;
1407 		break;
1408 	case VHOST_SET_VRING_ADDR:
1409 		if (copy_from_user(&a, argp, sizeof a)) {
1410 			r = -EFAULT;
1411 			break;
1412 		}
1413 		if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) {
1414 			r = -EOPNOTSUPP;
1415 			break;
1416 		}
1417 		/* For 32bit, verify that the top 32bits of the user
1418 		   data are set to zero. */
1419 		if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr ||
1420 		    (u64)(unsigned long)a.used_user_addr != a.used_user_addr ||
1421 		    (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) {
1422 			r = -EFAULT;
1423 			break;
1424 		}
1425 
1426 		/* Make sure it's safe to cast pointers to vring types. */
1427 		BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE);
1428 		BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE);
1429 		if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) ||
1430 		    (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) ||
1431 		    (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1))) {
1432 			r = -EINVAL;
1433 			break;
1434 		}
1435 
1436 		/* We only verify access here if backend is configured.
1437 		 * If it is not, we don't as size might not have been setup.
1438 		 * We will verify when backend is configured. */
1439 		if (vq->private_data) {
1440 			if (!vq_access_ok(vq, vq->num,
1441 				(void __user *)(unsigned long)a.desc_user_addr,
1442 				(void __user *)(unsigned long)a.avail_user_addr,
1443 				(void __user *)(unsigned long)a.used_user_addr)) {
1444 				r = -EINVAL;
1445 				break;
1446 			}
1447 
1448 			/* Also validate log access for used ring if enabled. */
1449 			if ((a.flags & (0x1 << VHOST_VRING_F_LOG)) &&
1450 			    !log_access_ok(vq->log_base, a.log_guest_addr,
1451 					   sizeof *vq->used +
1452 					   vq->num * sizeof *vq->used->ring)) {
1453 				r = -EINVAL;
1454 				break;
1455 			}
1456 		}
1457 
1458 		vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG));
1459 		vq->desc = (void __user *)(unsigned long)a.desc_user_addr;
1460 		vq->avail = (void __user *)(unsigned long)a.avail_user_addr;
1461 		vq->log_addr = a.log_guest_addr;
1462 		vq->used = (void __user *)(unsigned long)a.used_user_addr;
1463 		break;
1464 	case VHOST_SET_VRING_KICK:
1465 		if (copy_from_user(&f, argp, sizeof f)) {
1466 			r = -EFAULT;
1467 			break;
1468 		}
1469 		eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd);
1470 		if (IS_ERR(eventfp)) {
1471 			r = PTR_ERR(eventfp);
1472 			break;
1473 		}
1474 		if (eventfp != vq->kick) {
1475 			pollstop = (filep = vq->kick) != NULL;
1476 			pollstart = (vq->kick = eventfp) != NULL;
1477 		} else
1478 			filep = eventfp;
1479 		break;
1480 	case VHOST_SET_VRING_CALL:
1481 		if (copy_from_user(&f, argp, sizeof f)) {
1482 			r = -EFAULT;
1483 			break;
1484 		}
1485 		ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd);
1486 		if (IS_ERR(ctx)) {
1487 			r = PTR_ERR(ctx);
1488 			break;
1489 		}
1490 		swap(ctx, vq->call_ctx);
1491 		break;
1492 	case VHOST_SET_VRING_ERR:
1493 		if (copy_from_user(&f, argp, sizeof f)) {
1494 			r = -EFAULT;
1495 			break;
1496 		}
1497 		ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd);
1498 		if (IS_ERR(ctx)) {
1499 			r = PTR_ERR(ctx);
1500 			break;
1501 		}
1502 		swap(ctx, vq->error_ctx);
1503 		break;
1504 	case VHOST_SET_VRING_ENDIAN:
1505 		r = vhost_set_vring_endian(vq, argp);
1506 		break;
1507 	case VHOST_GET_VRING_ENDIAN:
1508 		r = vhost_get_vring_endian(vq, idx, argp);
1509 		break;
1510 	case VHOST_SET_VRING_BUSYLOOP_TIMEOUT:
1511 		if (copy_from_user(&s, argp, sizeof(s))) {
1512 			r = -EFAULT;
1513 			break;
1514 		}
1515 		vq->busyloop_timeout = s.num;
1516 		break;
1517 	case VHOST_GET_VRING_BUSYLOOP_TIMEOUT:
1518 		s.index = idx;
1519 		s.num = vq->busyloop_timeout;
1520 		if (copy_to_user(argp, &s, sizeof(s)))
1521 			r = -EFAULT;
1522 		break;
1523 	default:
1524 		r = -ENOIOCTLCMD;
1525 	}
1526 
1527 	if (pollstop && vq->handle_kick)
1528 		vhost_poll_stop(&vq->poll);
1529 
1530 	if (!IS_ERR_OR_NULL(ctx))
1531 		eventfd_ctx_put(ctx);
1532 	if (filep)
1533 		fput(filep);
1534 
1535 	if (pollstart && vq->handle_kick)
1536 		r = vhost_poll_start(&vq->poll, vq->kick);
1537 
1538 	mutex_unlock(&vq->mutex);
1539 
1540 	if (pollstop && vq->handle_kick)
1541 		vhost_poll_flush(&vq->poll);
1542 	return r;
1543 }
1544 EXPORT_SYMBOL_GPL(vhost_vring_ioctl);
1545 
1546 int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled)
1547 {
1548 	struct vhost_umem *niotlb, *oiotlb;
1549 	int i;
1550 
1551 	niotlb = vhost_umem_alloc();
1552 	if (!niotlb)
1553 		return -ENOMEM;
1554 
1555 	oiotlb = d->iotlb;
1556 	d->iotlb = niotlb;
1557 
1558 	for (i = 0; i < d->nvqs; ++i) {
1559 		mutex_lock(&d->vqs[i]->mutex);
1560 		d->vqs[i]->iotlb = niotlb;
1561 		mutex_unlock(&d->vqs[i]->mutex);
1562 	}
1563 
1564 	vhost_umem_clean(oiotlb);
1565 
1566 	return 0;
1567 }
1568 EXPORT_SYMBOL_GPL(vhost_init_device_iotlb);
1569 
1570 /* Caller must have device mutex */
1571 long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
1572 {
1573 	struct eventfd_ctx *ctx;
1574 	u64 p;
1575 	long r;
1576 	int i, fd;
1577 
1578 	/* If you are not the owner, you can become one */
1579 	if (ioctl == VHOST_SET_OWNER) {
1580 		r = vhost_dev_set_owner(d);
1581 		goto done;
1582 	}
1583 
1584 	/* You must be the owner to do anything else */
1585 	r = vhost_dev_check_owner(d);
1586 	if (r)
1587 		goto done;
1588 
1589 	switch (ioctl) {
1590 	case VHOST_SET_MEM_TABLE:
1591 		r = vhost_set_memory(d, argp);
1592 		break;
1593 	case VHOST_SET_LOG_BASE:
1594 		if (copy_from_user(&p, argp, sizeof p)) {
1595 			r = -EFAULT;
1596 			break;
1597 		}
1598 		if ((u64)(unsigned long)p != p) {
1599 			r = -EFAULT;
1600 			break;
1601 		}
1602 		for (i = 0; i < d->nvqs; ++i) {
1603 			struct vhost_virtqueue *vq;
1604 			void __user *base = (void __user *)(unsigned long)p;
1605 			vq = d->vqs[i];
1606 			mutex_lock(&vq->mutex);
1607 			/* If ring is inactive, will check when it's enabled. */
1608 			if (vq->private_data && !vq_log_access_ok(vq, base))
1609 				r = -EFAULT;
1610 			else
1611 				vq->log_base = base;
1612 			mutex_unlock(&vq->mutex);
1613 		}
1614 		break;
1615 	case VHOST_SET_LOG_FD:
1616 		r = get_user(fd, (int __user *)argp);
1617 		if (r < 0)
1618 			break;
1619 		ctx = fd == -1 ? NULL : eventfd_ctx_fdget(fd);
1620 		if (IS_ERR(ctx)) {
1621 			r = PTR_ERR(ctx);
1622 			break;
1623 		}
1624 		swap(ctx, d->log_ctx);
1625 		for (i = 0; i < d->nvqs; ++i) {
1626 			mutex_lock(&d->vqs[i]->mutex);
1627 			d->vqs[i]->log_ctx = d->log_ctx;
1628 			mutex_unlock(&d->vqs[i]->mutex);
1629 		}
1630 		if (ctx)
1631 			eventfd_ctx_put(ctx);
1632 		break;
1633 	default:
1634 		r = -ENOIOCTLCMD;
1635 		break;
1636 	}
1637 done:
1638 	return r;
1639 }
1640 EXPORT_SYMBOL_GPL(vhost_dev_ioctl);
1641 
1642 /* TODO: This is really inefficient.  We need something like get_user()
1643  * (instruction directly accesses the data, with an exception table entry
1644  * returning -EFAULT). See Documentation/x86/exception-tables.txt.
1645  */
1646 static int set_bit_to_user(int nr, void __user *addr)
1647 {
1648 	unsigned long log = (unsigned long)addr;
1649 	struct page *page;
1650 	void *base;
1651 	int bit = nr + (log % PAGE_SIZE) * 8;
1652 	int r;
1653 
1654 	r = get_user_pages_fast(log, 1, 1, &page);
1655 	if (r < 0)
1656 		return r;
1657 	BUG_ON(r != 1);
1658 	base = kmap_atomic(page);
1659 	set_bit(bit, base);
1660 	kunmap_atomic(base);
1661 	set_page_dirty_lock(page);
1662 	put_page(page);
1663 	return 0;
1664 }
1665 
1666 static int log_write(void __user *log_base,
1667 		     u64 write_address, u64 write_length)
1668 {
1669 	u64 write_page = write_address / VHOST_PAGE_SIZE;
1670 	int r;
1671 
1672 	if (!write_length)
1673 		return 0;
1674 	write_length += write_address % VHOST_PAGE_SIZE;
1675 	for (;;) {
1676 		u64 base = (u64)(unsigned long)log_base;
1677 		u64 log = base + write_page / 8;
1678 		int bit = write_page % 8;
1679 		if ((u64)(unsigned long)log != log)
1680 			return -EFAULT;
1681 		r = set_bit_to_user(bit, (void __user *)(unsigned long)log);
1682 		if (r < 0)
1683 			return r;
1684 		if (write_length <= VHOST_PAGE_SIZE)
1685 			break;
1686 		write_length -= VHOST_PAGE_SIZE;
1687 		write_page += 1;
1688 	}
1689 	return r;
1690 }
1691 
1692 int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
1693 		    unsigned int log_num, u64 len)
1694 {
1695 	int i, r;
1696 
1697 	/* Make sure data written is seen before log. */
1698 	smp_wmb();
1699 	for (i = 0; i < log_num; ++i) {
1700 		u64 l = min(log[i].len, len);
1701 		r = log_write(vq->log_base, log[i].addr, l);
1702 		if (r < 0)
1703 			return r;
1704 		len -= l;
1705 		if (!len) {
1706 			if (vq->log_ctx)
1707 				eventfd_signal(vq->log_ctx, 1);
1708 			return 0;
1709 		}
1710 	}
1711 	/* Length written exceeds what we have stored. This is a bug. */
1712 	BUG();
1713 	return 0;
1714 }
1715 EXPORT_SYMBOL_GPL(vhost_log_write);
1716 
1717 static int vhost_update_used_flags(struct vhost_virtqueue *vq)
1718 {
1719 	void __user *used;
1720 	if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags),
1721 			   &vq->used->flags) < 0)
1722 		return -EFAULT;
1723 	if (unlikely(vq->log_used)) {
1724 		/* Make sure the flag is seen before log. */
1725 		smp_wmb();
1726 		/* Log used flag write. */
1727 		used = &vq->used->flags;
1728 		log_write(vq->log_base, vq->log_addr +
1729 			  (used - (void __user *)vq->used),
1730 			  sizeof vq->used->flags);
1731 		if (vq->log_ctx)
1732 			eventfd_signal(vq->log_ctx, 1);
1733 	}
1734 	return 0;
1735 }
1736 
1737 static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event)
1738 {
1739 	if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx),
1740 			   vhost_avail_event(vq)))
1741 		return -EFAULT;
1742 	if (unlikely(vq->log_used)) {
1743 		void __user *used;
1744 		/* Make sure the event is seen before log. */
1745 		smp_wmb();
1746 		/* Log avail event write */
1747 		used = vhost_avail_event(vq);
1748 		log_write(vq->log_base, vq->log_addr +
1749 			  (used - (void __user *)vq->used),
1750 			  sizeof *vhost_avail_event(vq));
1751 		if (vq->log_ctx)
1752 			eventfd_signal(vq->log_ctx, 1);
1753 	}
1754 	return 0;
1755 }
1756 
1757 int vhost_vq_init_access(struct vhost_virtqueue *vq)
1758 {
1759 	__virtio16 last_used_idx;
1760 	int r;
1761 	bool is_le = vq->is_le;
1762 
1763 	if (!vq->private_data)
1764 		return 0;
1765 
1766 	vhost_init_is_le(vq);
1767 
1768 	r = vhost_update_used_flags(vq);
1769 	if (r)
1770 		goto err;
1771 	vq->signalled_used_valid = false;
1772 	if (!vq->iotlb &&
1773 	    !access_ok(VERIFY_READ, &vq->used->idx, sizeof vq->used->idx)) {
1774 		r = -EFAULT;
1775 		goto err;
1776 	}
1777 	r = vhost_get_used(vq, last_used_idx, &vq->used->idx);
1778 	if (r) {
1779 		vq_err(vq, "Can't access used idx at %p\n",
1780 		       &vq->used->idx);
1781 		goto err;
1782 	}
1783 	vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx);
1784 	return 0;
1785 
1786 err:
1787 	vq->is_le = is_le;
1788 	return r;
1789 }
1790 EXPORT_SYMBOL_GPL(vhost_vq_init_access);
1791 
1792 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
1793 			  struct iovec iov[], int iov_size, int access)
1794 {
1795 	const struct vhost_umem_node *node;
1796 	struct vhost_dev *dev = vq->dev;
1797 	struct vhost_umem *umem = dev->iotlb ? dev->iotlb : dev->umem;
1798 	struct iovec *_iov;
1799 	u64 s = 0;
1800 	int ret = 0;
1801 
1802 	while ((u64)len > s) {
1803 		u64 size;
1804 		if (unlikely(ret >= iov_size)) {
1805 			ret = -ENOBUFS;
1806 			break;
1807 		}
1808 
1809 		node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1810 							addr, addr + len - 1);
1811 		if (node == NULL || node->start > addr) {
1812 			if (umem != dev->iotlb) {
1813 				ret = -EFAULT;
1814 				break;
1815 			}
1816 			ret = -EAGAIN;
1817 			break;
1818 		} else if (!(node->perm & access)) {
1819 			ret = -EPERM;
1820 			break;
1821 		}
1822 
1823 		_iov = iov + ret;
1824 		size = node->size - addr + node->start;
1825 		_iov->iov_len = min((u64)len - s, size);
1826 		_iov->iov_base = (void __user *)(unsigned long)
1827 			(node->userspace_addr + addr - node->start);
1828 		s += size;
1829 		addr += size;
1830 		++ret;
1831 	}
1832 
1833 	if (ret == -EAGAIN)
1834 		vhost_iotlb_miss(vq, addr, access);
1835 	return ret;
1836 }
1837 
1838 /* Each buffer in the virtqueues is actually a chain of descriptors.  This
1839  * function returns the next descriptor in the chain,
1840  * or -1U if we're at the end. */
1841 static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc)
1842 {
1843 	unsigned int next;
1844 
1845 	/* If this descriptor says it doesn't chain, we're done. */
1846 	if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT)))
1847 		return -1U;
1848 
1849 	/* Check they're not leading us off end of descriptors. */
1850 	next = vhost16_to_cpu(vq, READ_ONCE(desc->next));
1851 	return next;
1852 }
1853 
1854 static int get_indirect(struct vhost_virtqueue *vq,
1855 			struct iovec iov[], unsigned int iov_size,
1856 			unsigned int *out_num, unsigned int *in_num,
1857 			struct vhost_log *log, unsigned int *log_num,
1858 			struct vring_desc *indirect)
1859 {
1860 	struct vring_desc desc;
1861 	unsigned int i = 0, count, found = 0;
1862 	u32 len = vhost32_to_cpu(vq, indirect->len);
1863 	struct iov_iter from;
1864 	int ret, access;
1865 
1866 	/* Sanity check */
1867 	if (unlikely(len % sizeof desc)) {
1868 		vq_err(vq, "Invalid length in indirect descriptor: "
1869 		       "len 0x%llx not multiple of 0x%zx\n",
1870 		       (unsigned long long)len,
1871 		       sizeof desc);
1872 		return -EINVAL;
1873 	}
1874 
1875 	ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect,
1876 			     UIO_MAXIOV, VHOST_ACCESS_RO);
1877 	if (unlikely(ret < 0)) {
1878 		if (ret != -EAGAIN)
1879 			vq_err(vq, "Translation failure %d in indirect.\n", ret);
1880 		return ret;
1881 	}
1882 	iov_iter_init(&from, READ, vq->indirect, ret, len);
1883 
1884 	/* We will use the result as an address to read from, so most
1885 	 * architectures only need a compiler barrier here. */
1886 	read_barrier_depends();
1887 
1888 	count = len / sizeof desc;
1889 	/* Buffers are chained via a 16 bit next field, so
1890 	 * we can have at most 2^16 of these. */
1891 	if (unlikely(count > USHRT_MAX + 1)) {
1892 		vq_err(vq, "Indirect buffer length too big: %d\n",
1893 		       indirect->len);
1894 		return -E2BIG;
1895 	}
1896 
1897 	do {
1898 		unsigned iov_count = *in_num + *out_num;
1899 		if (unlikely(++found > count)) {
1900 			vq_err(vq, "Loop detected: last one at %u "
1901 			       "indirect size %u\n",
1902 			       i, count);
1903 			return -EINVAL;
1904 		}
1905 		if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) {
1906 			vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n",
1907 			       i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
1908 			return -EINVAL;
1909 		}
1910 		if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) {
1911 			vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n",
1912 			       i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
1913 			return -EINVAL;
1914 		}
1915 
1916 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
1917 			access = VHOST_ACCESS_WO;
1918 		else
1919 			access = VHOST_ACCESS_RO;
1920 
1921 		ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
1922 				     vhost32_to_cpu(vq, desc.len), iov + iov_count,
1923 				     iov_size - iov_count, access);
1924 		if (unlikely(ret < 0)) {
1925 			if (ret != -EAGAIN)
1926 				vq_err(vq, "Translation failure %d indirect idx %d\n",
1927 					ret, i);
1928 			return ret;
1929 		}
1930 		/* If this is an input descriptor, increment that count. */
1931 		if (access == VHOST_ACCESS_WO) {
1932 			*in_num += ret;
1933 			if (unlikely(log)) {
1934 				log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
1935 				log[*log_num].len = vhost32_to_cpu(vq, desc.len);
1936 				++*log_num;
1937 			}
1938 		} else {
1939 			/* If it's an output descriptor, they're all supposed
1940 			 * to come before any input descriptors. */
1941 			if (unlikely(*in_num)) {
1942 				vq_err(vq, "Indirect descriptor "
1943 				       "has out after in: idx %d\n", i);
1944 				return -EINVAL;
1945 			}
1946 			*out_num += ret;
1947 		}
1948 	} while ((i = next_desc(vq, &desc)) != -1);
1949 	return 0;
1950 }
1951 
1952 /* This looks in the virtqueue and for the first available buffer, and converts
1953  * it to an iovec for convenient access.  Since descriptors consist of some
1954  * number of output then some number of input descriptors, it's actually two
1955  * iovecs, but we pack them into one and note how many of each there were.
1956  *
1957  * This function returns the descriptor number found, or vq->num (which is
1958  * never a valid descriptor number) if none was found.  A negative code is
1959  * returned on error. */
1960 int vhost_get_vq_desc(struct vhost_virtqueue *vq,
1961 		      struct iovec iov[], unsigned int iov_size,
1962 		      unsigned int *out_num, unsigned int *in_num,
1963 		      struct vhost_log *log, unsigned int *log_num)
1964 {
1965 	struct vring_desc desc;
1966 	unsigned int i, head, found = 0;
1967 	u16 last_avail_idx;
1968 	__virtio16 avail_idx;
1969 	__virtio16 ring_head;
1970 	int ret, access;
1971 
1972 	/* Check it isn't doing very strange things with descriptor numbers. */
1973 	last_avail_idx = vq->last_avail_idx;
1974 
1975 	if (vq->avail_idx == vq->last_avail_idx) {
1976 		if (unlikely(vhost_get_avail(vq, avail_idx, &vq->avail->idx))) {
1977 			vq_err(vq, "Failed to access avail idx at %p\n",
1978 				&vq->avail->idx);
1979 			return -EFAULT;
1980 		}
1981 		vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
1982 
1983 		if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) {
1984 			vq_err(vq, "Guest moved used index from %u to %u",
1985 				last_avail_idx, vq->avail_idx);
1986 			return -EFAULT;
1987 		}
1988 
1989 		/* If there's nothing new since last we looked, return
1990 		 * invalid.
1991 		 */
1992 		if (vq->avail_idx == last_avail_idx)
1993 			return vq->num;
1994 
1995 		/* Only get avail ring entries after they have been
1996 		 * exposed by guest.
1997 		 */
1998 		smp_rmb();
1999 	}
2000 
2001 	/* Grab the next descriptor number they're advertising, and increment
2002 	 * the index we've seen. */
2003 	if (unlikely(vhost_get_avail(vq, ring_head,
2004 		     &vq->avail->ring[last_avail_idx & (vq->num - 1)]))) {
2005 		vq_err(vq, "Failed to read head: idx %d address %p\n",
2006 		       last_avail_idx,
2007 		       &vq->avail->ring[last_avail_idx % vq->num]);
2008 		return -EFAULT;
2009 	}
2010 
2011 	head = vhost16_to_cpu(vq, ring_head);
2012 
2013 	/* If their number is silly, that's an error. */
2014 	if (unlikely(head >= vq->num)) {
2015 		vq_err(vq, "Guest says index %u > %u is available",
2016 		       head, vq->num);
2017 		return -EINVAL;
2018 	}
2019 
2020 	/* When we start there are none of either input nor output. */
2021 	*out_num = *in_num = 0;
2022 	if (unlikely(log))
2023 		*log_num = 0;
2024 
2025 	i = head;
2026 	do {
2027 		unsigned iov_count = *in_num + *out_num;
2028 		if (unlikely(i >= vq->num)) {
2029 			vq_err(vq, "Desc index is %u > %u, head = %u",
2030 			       i, vq->num, head);
2031 			return -EINVAL;
2032 		}
2033 		if (unlikely(++found > vq->num)) {
2034 			vq_err(vq, "Loop detected: last one at %u "
2035 			       "vq size %u head %u\n",
2036 			       i, vq->num, head);
2037 			return -EINVAL;
2038 		}
2039 		ret = vhost_copy_from_user(vq, &desc, vq->desc + i,
2040 					   sizeof desc);
2041 		if (unlikely(ret)) {
2042 			vq_err(vq, "Failed to get descriptor: idx %d addr %p\n",
2043 			       i, vq->desc + i);
2044 			return -EFAULT;
2045 		}
2046 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) {
2047 			ret = get_indirect(vq, iov, iov_size,
2048 					   out_num, in_num,
2049 					   log, log_num, &desc);
2050 			if (unlikely(ret < 0)) {
2051 				if (ret != -EAGAIN)
2052 					vq_err(vq, "Failure detected "
2053 						"in indirect descriptor at idx %d\n", i);
2054 				return ret;
2055 			}
2056 			continue;
2057 		}
2058 
2059 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2060 			access = VHOST_ACCESS_WO;
2061 		else
2062 			access = VHOST_ACCESS_RO;
2063 		ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2064 				     vhost32_to_cpu(vq, desc.len), iov + iov_count,
2065 				     iov_size - iov_count, access);
2066 		if (unlikely(ret < 0)) {
2067 			if (ret != -EAGAIN)
2068 				vq_err(vq, "Translation failure %d descriptor idx %d\n",
2069 					ret, i);
2070 			return ret;
2071 		}
2072 		if (access == VHOST_ACCESS_WO) {
2073 			/* If this is an input descriptor,
2074 			 * increment that count. */
2075 			*in_num += ret;
2076 			if (unlikely(log)) {
2077 				log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2078 				log[*log_num].len = vhost32_to_cpu(vq, desc.len);
2079 				++*log_num;
2080 			}
2081 		} else {
2082 			/* If it's an output descriptor, they're all supposed
2083 			 * to come before any input descriptors. */
2084 			if (unlikely(*in_num)) {
2085 				vq_err(vq, "Descriptor has out after in: "
2086 				       "idx %d\n", i);
2087 				return -EINVAL;
2088 			}
2089 			*out_num += ret;
2090 		}
2091 	} while ((i = next_desc(vq, &desc)) != -1);
2092 
2093 	/* On success, increment avail index. */
2094 	vq->last_avail_idx++;
2095 
2096 	/* Assume notifications from guest are disabled at this point,
2097 	 * if they aren't we would need to update avail_event index. */
2098 	BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY));
2099 	return head;
2100 }
2101 EXPORT_SYMBOL_GPL(vhost_get_vq_desc);
2102 
2103 /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */
2104 void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n)
2105 {
2106 	vq->last_avail_idx -= n;
2107 }
2108 EXPORT_SYMBOL_GPL(vhost_discard_vq_desc);
2109 
2110 /* After we've used one of their buffers, we tell them about it.  We'll then
2111  * want to notify the guest, using eventfd. */
2112 int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len)
2113 {
2114 	struct vring_used_elem heads = {
2115 		cpu_to_vhost32(vq, head),
2116 		cpu_to_vhost32(vq, len)
2117 	};
2118 
2119 	return vhost_add_used_n(vq, &heads, 1);
2120 }
2121 EXPORT_SYMBOL_GPL(vhost_add_used);
2122 
2123 static int __vhost_add_used_n(struct vhost_virtqueue *vq,
2124 			    struct vring_used_elem *heads,
2125 			    unsigned count)
2126 {
2127 	struct vring_used_elem __user *used;
2128 	u16 old, new;
2129 	int start;
2130 
2131 	start = vq->last_used_idx & (vq->num - 1);
2132 	used = vq->used->ring + start;
2133 	if (count == 1) {
2134 		if (vhost_put_user(vq, heads[0].id, &used->id)) {
2135 			vq_err(vq, "Failed to write used id");
2136 			return -EFAULT;
2137 		}
2138 		if (vhost_put_user(vq, heads[0].len, &used->len)) {
2139 			vq_err(vq, "Failed to write used len");
2140 			return -EFAULT;
2141 		}
2142 	} else if (vhost_copy_to_user(vq, used, heads, count * sizeof *used)) {
2143 		vq_err(vq, "Failed to write used");
2144 		return -EFAULT;
2145 	}
2146 	if (unlikely(vq->log_used)) {
2147 		/* Make sure data is seen before log. */
2148 		smp_wmb();
2149 		/* Log used ring entry write. */
2150 		log_write(vq->log_base,
2151 			  vq->log_addr +
2152 			   ((void __user *)used - (void __user *)vq->used),
2153 			  count * sizeof *used);
2154 	}
2155 	old = vq->last_used_idx;
2156 	new = (vq->last_used_idx += count);
2157 	/* If the driver never bothers to signal in a very long while,
2158 	 * used index might wrap around. If that happens, invalidate
2159 	 * signalled_used index we stored. TODO: make sure driver
2160 	 * signals at least once in 2^16 and remove this. */
2161 	if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old)))
2162 		vq->signalled_used_valid = false;
2163 	return 0;
2164 }
2165 
2166 /* After we've used one of their buffers, we tell them about it.  We'll then
2167  * want to notify the guest, using eventfd. */
2168 int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads,
2169 		     unsigned count)
2170 {
2171 	int start, n, r;
2172 
2173 	start = vq->last_used_idx & (vq->num - 1);
2174 	n = vq->num - start;
2175 	if (n < count) {
2176 		r = __vhost_add_used_n(vq, heads, n);
2177 		if (r < 0)
2178 			return r;
2179 		heads += n;
2180 		count -= n;
2181 	}
2182 	r = __vhost_add_used_n(vq, heads, count);
2183 
2184 	/* Make sure buffer is written before we update index. */
2185 	smp_wmb();
2186 	if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx),
2187 			   &vq->used->idx)) {
2188 		vq_err(vq, "Failed to increment used idx");
2189 		return -EFAULT;
2190 	}
2191 	if (unlikely(vq->log_used)) {
2192 		/* Log used index update. */
2193 		log_write(vq->log_base,
2194 			  vq->log_addr + offsetof(struct vring_used, idx),
2195 			  sizeof vq->used->idx);
2196 		if (vq->log_ctx)
2197 			eventfd_signal(vq->log_ctx, 1);
2198 	}
2199 	return r;
2200 }
2201 EXPORT_SYMBOL_GPL(vhost_add_used_n);
2202 
2203 static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2204 {
2205 	__u16 old, new;
2206 	__virtio16 event;
2207 	bool v;
2208 	/* Flush out used index updates. This is paired
2209 	 * with the barrier that the Guest executes when enabling
2210 	 * interrupts. */
2211 	smp_mb();
2212 
2213 	if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2214 	    unlikely(vq->avail_idx == vq->last_avail_idx))
2215 		return true;
2216 
2217 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2218 		__virtio16 flags;
2219 		if (vhost_get_avail(vq, flags, &vq->avail->flags)) {
2220 			vq_err(vq, "Failed to get flags");
2221 			return true;
2222 		}
2223 		return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT));
2224 	}
2225 	old = vq->signalled_used;
2226 	v = vq->signalled_used_valid;
2227 	new = vq->signalled_used = vq->last_used_idx;
2228 	vq->signalled_used_valid = true;
2229 
2230 	if (unlikely(!v))
2231 		return true;
2232 
2233 	if (vhost_get_avail(vq, event, vhost_used_event(vq))) {
2234 		vq_err(vq, "Failed to get used event idx");
2235 		return true;
2236 	}
2237 	return vring_need_event(vhost16_to_cpu(vq, event), new, old);
2238 }
2239 
2240 /* This actually signals the guest, using eventfd. */
2241 void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2242 {
2243 	/* Signal the Guest tell them we used something up. */
2244 	if (vq->call_ctx && vhost_notify(dev, vq))
2245 		eventfd_signal(vq->call_ctx, 1);
2246 }
2247 EXPORT_SYMBOL_GPL(vhost_signal);
2248 
2249 /* And here's the combo meal deal.  Supersize me! */
2250 void vhost_add_used_and_signal(struct vhost_dev *dev,
2251 			       struct vhost_virtqueue *vq,
2252 			       unsigned int head, int len)
2253 {
2254 	vhost_add_used(vq, head, len);
2255 	vhost_signal(dev, vq);
2256 }
2257 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal);
2258 
2259 /* multi-buffer version of vhost_add_used_and_signal */
2260 void vhost_add_used_and_signal_n(struct vhost_dev *dev,
2261 				 struct vhost_virtqueue *vq,
2262 				 struct vring_used_elem *heads, unsigned count)
2263 {
2264 	vhost_add_used_n(vq, heads, count);
2265 	vhost_signal(dev, vq);
2266 }
2267 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n);
2268 
2269 /* return true if we're sure that avaiable ring is empty */
2270 bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2271 {
2272 	__virtio16 avail_idx;
2273 	int r;
2274 
2275 	if (vq->avail_idx != vq->last_avail_idx)
2276 		return false;
2277 
2278 	r = vhost_get_avail(vq, avail_idx, &vq->avail->idx);
2279 	if (unlikely(r))
2280 		return false;
2281 	vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
2282 
2283 	return vq->avail_idx == vq->last_avail_idx;
2284 }
2285 EXPORT_SYMBOL_GPL(vhost_vq_avail_empty);
2286 
2287 /* OK, now we need to know about added descriptors. */
2288 bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2289 {
2290 	__virtio16 avail_idx;
2291 	int r;
2292 
2293 	if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY))
2294 		return false;
2295 	vq->used_flags &= ~VRING_USED_F_NO_NOTIFY;
2296 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2297 		r = vhost_update_used_flags(vq);
2298 		if (r) {
2299 			vq_err(vq, "Failed to enable notification at %p: %d\n",
2300 			       &vq->used->flags, r);
2301 			return false;
2302 		}
2303 	} else {
2304 		r = vhost_update_avail_event(vq, vq->avail_idx);
2305 		if (r) {
2306 			vq_err(vq, "Failed to update avail event index at %p: %d\n",
2307 			       vhost_avail_event(vq), r);
2308 			return false;
2309 		}
2310 	}
2311 	/* They could have slipped one in as we were doing that: make
2312 	 * sure it's written, then check again. */
2313 	smp_mb();
2314 	r = vhost_get_avail(vq, avail_idx, &vq->avail->idx);
2315 	if (r) {
2316 		vq_err(vq, "Failed to check avail idx at %p: %d\n",
2317 		       &vq->avail->idx, r);
2318 		return false;
2319 	}
2320 
2321 	return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx;
2322 }
2323 EXPORT_SYMBOL_GPL(vhost_enable_notify);
2324 
2325 /* We don't need to be notified again. */
2326 void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2327 {
2328 	int r;
2329 
2330 	if (vq->used_flags & VRING_USED_F_NO_NOTIFY)
2331 		return;
2332 	vq->used_flags |= VRING_USED_F_NO_NOTIFY;
2333 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2334 		r = vhost_update_used_flags(vq);
2335 		if (r)
2336 			vq_err(vq, "Failed to enable notification at %p: %d\n",
2337 			       &vq->used->flags, r);
2338 	}
2339 }
2340 EXPORT_SYMBOL_GPL(vhost_disable_notify);
2341 
2342 /* Create a new message. */
2343 struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type)
2344 {
2345 	struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL);
2346 	if (!node)
2347 		return NULL;
2348 	node->vq = vq;
2349 	node->msg.type = type;
2350 	return node;
2351 }
2352 EXPORT_SYMBOL_GPL(vhost_new_msg);
2353 
2354 void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head,
2355 		       struct vhost_msg_node *node)
2356 {
2357 	spin_lock(&dev->iotlb_lock);
2358 	list_add_tail(&node->node, head);
2359 	spin_unlock(&dev->iotlb_lock);
2360 
2361 	wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
2362 }
2363 EXPORT_SYMBOL_GPL(vhost_enqueue_msg);
2364 
2365 struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
2366 					 struct list_head *head)
2367 {
2368 	struct vhost_msg_node *node = NULL;
2369 
2370 	spin_lock(&dev->iotlb_lock);
2371 	if (!list_empty(head)) {
2372 		node = list_first_entry(head, struct vhost_msg_node,
2373 					node);
2374 		list_del(&node->node);
2375 	}
2376 	spin_unlock(&dev->iotlb_lock);
2377 
2378 	return node;
2379 }
2380 EXPORT_SYMBOL_GPL(vhost_dequeue_msg);
2381 
2382 
2383 static int __init vhost_init(void)
2384 {
2385 	return 0;
2386 }
2387 
2388 static void __exit vhost_exit(void)
2389 {
2390 }
2391 
2392 module_init(vhost_init);
2393 module_exit(vhost_exit);
2394 
2395 MODULE_VERSION("0.0.1");
2396 MODULE_LICENSE("GPL v2");
2397 MODULE_AUTHOR("Michael S. Tsirkin");
2398 MODULE_DESCRIPTION("Host kernel accelerator for virtio");
2399