xref: /openbmc/linux/drivers/vhost/vhost.c (revision 5927145e)
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 		if (poll->wqh)
216 			remove_wait_queue(poll->wqh, &poll->wait);
217 		ret = -EINVAL;
218 	}
219 
220 	return ret;
221 }
222 EXPORT_SYMBOL_GPL(vhost_poll_start);
223 
224 /* Stop polling a file. After this function returns, it becomes safe to drop the
225  * file reference. You must also flush afterwards. */
226 void vhost_poll_stop(struct vhost_poll *poll)
227 {
228 	if (poll->wqh) {
229 		remove_wait_queue(poll->wqh, &poll->wait);
230 		poll->wqh = NULL;
231 	}
232 }
233 EXPORT_SYMBOL_GPL(vhost_poll_stop);
234 
235 void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work)
236 {
237 	struct vhost_flush_struct flush;
238 
239 	if (dev->worker) {
240 		init_completion(&flush.wait_event);
241 		vhost_work_init(&flush.work, vhost_flush_work);
242 
243 		vhost_work_queue(dev, &flush.work);
244 		wait_for_completion(&flush.wait_event);
245 	}
246 }
247 EXPORT_SYMBOL_GPL(vhost_work_flush);
248 
249 /* Flush any work that has been scheduled. When calling this, don't hold any
250  * locks that are also used by the callback. */
251 void vhost_poll_flush(struct vhost_poll *poll)
252 {
253 	vhost_work_flush(poll->dev, &poll->work);
254 }
255 EXPORT_SYMBOL_GPL(vhost_poll_flush);
256 
257 void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work)
258 {
259 	if (!dev->worker)
260 		return;
261 
262 	if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) {
263 		/* We can only add the work to the list after we're
264 		 * sure it was not in the list.
265 		 * test_and_set_bit() implies a memory barrier.
266 		 */
267 		llist_add(&work->node, &dev->work_list);
268 		wake_up_process(dev->worker);
269 	}
270 }
271 EXPORT_SYMBOL_GPL(vhost_work_queue);
272 
273 /* A lockless hint for busy polling code to exit the loop */
274 bool vhost_has_work(struct vhost_dev *dev)
275 {
276 	return !llist_empty(&dev->work_list);
277 }
278 EXPORT_SYMBOL_GPL(vhost_has_work);
279 
280 void vhost_poll_queue(struct vhost_poll *poll)
281 {
282 	vhost_work_queue(poll->dev, &poll->work);
283 }
284 EXPORT_SYMBOL_GPL(vhost_poll_queue);
285 
286 static void __vhost_vq_meta_reset(struct vhost_virtqueue *vq)
287 {
288 	int j;
289 
290 	for (j = 0; j < VHOST_NUM_ADDRS; j++)
291 		vq->meta_iotlb[j] = NULL;
292 }
293 
294 static void vhost_vq_meta_reset(struct vhost_dev *d)
295 {
296 	int i;
297 
298 	for (i = 0; i < d->nvqs; ++i)
299 		__vhost_vq_meta_reset(d->vqs[i]);
300 }
301 
302 static void vhost_vq_reset(struct vhost_dev *dev,
303 			   struct vhost_virtqueue *vq)
304 {
305 	vq->num = 1;
306 	vq->desc = NULL;
307 	vq->avail = NULL;
308 	vq->used = NULL;
309 	vq->last_avail_idx = 0;
310 	vq->avail_idx = 0;
311 	vq->last_used_idx = 0;
312 	vq->signalled_used = 0;
313 	vq->signalled_used_valid = false;
314 	vq->used_flags = 0;
315 	vq->log_used = false;
316 	vq->log_addr = -1ull;
317 	vq->private_data = NULL;
318 	vq->acked_features = 0;
319 	vq->log_base = NULL;
320 	vq->error_ctx = NULL;
321 	vq->kick = NULL;
322 	vq->call_ctx = NULL;
323 	vq->log_ctx = NULL;
324 	vhost_reset_is_le(vq);
325 	vhost_disable_cross_endian(vq);
326 	vq->busyloop_timeout = 0;
327 	vq->umem = NULL;
328 	vq->iotlb = NULL;
329 	__vhost_vq_meta_reset(vq);
330 }
331 
332 static int vhost_worker(void *data)
333 {
334 	struct vhost_dev *dev = data;
335 	struct vhost_work *work, *work_next;
336 	struct llist_node *node;
337 	mm_segment_t oldfs = get_fs();
338 
339 	set_fs(USER_DS);
340 	use_mm(dev->mm);
341 
342 	for (;;) {
343 		/* mb paired w/ kthread_stop */
344 		set_current_state(TASK_INTERRUPTIBLE);
345 
346 		if (kthread_should_stop()) {
347 			__set_current_state(TASK_RUNNING);
348 			break;
349 		}
350 
351 		node = llist_del_all(&dev->work_list);
352 		if (!node)
353 			schedule();
354 
355 		node = llist_reverse_order(node);
356 		/* make sure flag is seen after deletion */
357 		smp_wmb();
358 		llist_for_each_entry_safe(work, work_next, node, node) {
359 			clear_bit(VHOST_WORK_QUEUED, &work->flags);
360 			__set_current_state(TASK_RUNNING);
361 			work->fn(work);
362 			if (need_resched())
363 				schedule();
364 		}
365 	}
366 	unuse_mm(dev->mm);
367 	set_fs(oldfs);
368 	return 0;
369 }
370 
371 static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq)
372 {
373 	kfree(vq->indirect);
374 	vq->indirect = NULL;
375 	kfree(vq->log);
376 	vq->log = NULL;
377 	kfree(vq->heads);
378 	vq->heads = NULL;
379 }
380 
381 /* Helper to allocate iovec buffers for all vqs. */
382 static long vhost_dev_alloc_iovecs(struct vhost_dev *dev)
383 {
384 	struct vhost_virtqueue *vq;
385 	int i;
386 
387 	for (i = 0; i < dev->nvqs; ++i) {
388 		vq = dev->vqs[i];
389 		vq->indirect = kmalloc(sizeof *vq->indirect * UIO_MAXIOV,
390 				       GFP_KERNEL);
391 		vq->log = kmalloc(sizeof *vq->log * UIO_MAXIOV, GFP_KERNEL);
392 		vq->heads = kmalloc(sizeof *vq->heads * UIO_MAXIOV, GFP_KERNEL);
393 		if (!vq->indirect || !vq->log || !vq->heads)
394 			goto err_nomem;
395 	}
396 	return 0;
397 
398 err_nomem:
399 	for (; i >= 0; --i)
400 		vhost_vq_free_iovecs(dev->vqs[i]);
401 	return -ENOMEM;
402 }
403 
404 static void vhost_dev_free_iovecs(struct vhost_dev *dev)
405 {
406 	int i;
407 
408 	for (i = 0; i < dev->nvqs; ++i)
409 		vhost_vq_free_iovecs(dev->vqs[i]);
410 }
411 
412 void vhost_dev_init(struct vhost_dev *dev,
413 		    struct vhost_virtqueue **vqs, int nvqs)
414 {
415 	struct vhost_virtqueue *vq;
416 	int i;
417 
418 	dev->vqs = vqs;
419 	dev->nvqs = nvqs;
420 	mutex_init(&dev->mutex);
421 	dev->log_ctx = NULL;
422 	dev->umem = NULL;
423 	dev->iotlb = NULL;
424 	dev->mm = NULL;
425 	dev->worker = NULL;
426 	init_llist_head(&dev->work_list);
427 	init_waitqueue_head(&dev->wait);
428 	INIT_LIST_HEAD(&dev->read_list);
429 	INIT_LIST_HEAD(&dev->pending_list);
430 	spin_lock_init(&dev->iotlb_lock);
431 
432 
433 	for (i = 0; i < dev->nvqs; ++i) {
434 		vq = dev->vqs[i];
435 		vq->log = NULL;
436 		vq->indirect = NULL;
437 		vq->heads = NULL;
438 		vq->dev = dev;
439 		mutex_init(&vq->mutex);
440 		vhost_vq_reset(dev, vq);
441 		if (vq->handle_kick)
442 			vhost_poll_init(&vq->poll, vq->handle_kick,
443 					EPOLLIN, dev);
444 	}
445 }
446 EXPORT_SYMBOL_GPL(vhost_dev_init);
447 
448 /* Caller should have device mutex */
449 long vhost_dev_check_owner(struct vhost_dev *dev)
450 {
451 	/* Are you the owner? If not, I don't think you mean to do that */
452 	return dev->mm == current->mm ? 0 : -EPERM;
453 }
454 EXPORT_SYMBOL_GPL(vhost_dev_check_owner);
455 
456 struct vhost_attach_cgroups_struct {
457 	struct vhost_work work;
458 	struct task_struct *owner;
459 	int ret;
460 };
461 
462 static void vhost_attach_cgroups_work(struct vhost_work *work)
463 {
464 	struct vhost_attach_cgroups_struct *s;
465 
466 	s = container_of(work, struct vhost_attach_cgroups_struct, work);
467 	s->ret = cgroup_attach_task_all(s->owner, current);
468 }
469 
470 static int vhost_attach_cgroups(struct vhost_dev *dev)
471 {
472 	struct vhost_attach_cgroups_struct attach;
473 
474 	attach.owner = current;
475 	vhost_work_init(&attach.work, vhost_attach_cgroups_work);
476 	vhost_work_queue(dev, &attach.work);
477 	vhost_work_flush(dev, &attach.work);
478 	return attach.ret;
479 }
480 
481 /* Caller should have device mutex */
482 bool vhost_dev_has_owner(struct vhost_dev *dev)
483 {
484 	return dev->mm;
485 }
486 EXPORT_SYMBOL_GPL(vhost_dev_has_owner);
487 
488 /* Caller should have device mutex */
489 long vhost_dev_set_owner(struct vhost_dev *dev)
490 {
491 	struct task_struct *worker;
492 	int err;
493 
494 	/* Is there an owner already? */
495 	if (vhost_dev_has_owner(dev)) {
496 		err = -EBUSY;
497 		goto err_mm;
498 	}
499 
500 	/* No owner, become one */
501 	dev->mm = get_task_mm(current);
502 	worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid);
503 	if (IS_ERR(worker)) {
504 		err = PTR_ERR(worker);
505 		goto err_worker;
506 	}
507 
508 	dev->worker = worker;
509 	wake_up_process(worker);	/* avoid contributing to loadavg */
510 
511 	err = vhost_attach_cgroups(dev);
512 	if (err)
513 		goto err_cgroup;
514 
515 	err = vhost_dev_alloc_iovecs(dev);
516 	if (err)
517 		goto err_cgroup;
518 
519 	return 0;
520 err_cgroup:
521 	kthread_stop(worker);
522 	dev->worker = NULL;
523 err_worker:
524 	if (dev->mm)
525 		mmput(dev->mm);
526 	dev->mm = NULL;
527 err_mm:
528 	return err;
529 }
530 EXPORT_SYMBOL_GPL(vhost_dev_set_owner);
531 
532 struct vhost_umem *vhost_dev_reset_owner_prepare(void)
533 {
534 	return kvzalloc(sizeof(struct vhost_umem), GFP_KERNEL);
535 }
536 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare);
537 
538 /* Caller should have device mutex */
539 void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_umem *umem)
540 {
541 	int i;
542 
543 	vhost_dev_cleanup(dev);
544 
545 	/* Restore memory to default empty mapping. */
546 	INIT_LIST_HEAD(&umem->umem_list);
547 	dev->umem = umem;
548 	/* We don't need VQ locks below since vhost_dev_cleanup makes sure
549 	 * VQs aren't running.
550 	 */
551 	for (i = 0; i < dev->nvqs; ++i)
552 		dev->vqs[i]->umem = umem;
553 }
554 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner);
555 
556 void vhost_dev_stop(struct vhost_dev *dev)
557 {
558 	int i;
559 
560 	for (i = 0; i < dev->nvqs; ++i) {
561 		if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) {
562 			vhost_poll_stop(&dev->vqs[i]->poll);
563 			vhost_poll_flush(&dev->vqs[i]->poll);
564 		}
565 	}
566 }
567 EXPORT_SYMBOL_GPL(vhost_dev_stop);
568 
569 static void vhost_umem_free(struct vhost_umem *umem,
570 			    struct vhost_umem_node *node)
571 {
572 	vhost_umem_interval_tree_remove(node, &umem->umem_tree);
573 	list_del(&node->link);
574 	kfree(node);
575 	umem->numem--;
576 }
577 
578 static void vhost_umem_clean(struct vhost_umem *umem)
579 {
580 	struct vhost_umem_node *node, *tmp;
581 
582 	if (!umem)
583 		return;
584 
585 	list_for_each_entry_safe(node, tmp, &umem->umem_list, link)
586 		vhost_umem_free(umem, node);
587 
588 	kvfree(umem);
589 }
590 
591 static void vhost_clear_msg(struct vhost_dev *dev)
592 {
593 	struct vhost_msg_node *node, *n;
594 
595 	spin_lock(&dev->iotlb_lock);
596 
597 	list_for_each_entry_safe(node, n, &dev->read_list, node) {
598 		list_del(&node->node);
599 		kfree(node);
600 	}
601 
602 	list_for_each_entry_safe(node, n, &dev->pending_list, node) {
603 		list_del(&node->node);
604 		kfree(node);
605 	}
606 
607 	spin_unlock(&dev->iotlb_lock);
608 }
609 
610 void vhost_dev_cleanup(struct vhost_dev *dev)
611 {
612 	int i;
613 
614 	for (i = 0; i < dev->nvqs; ++i) {
615 		if (dev->vqs[i]->error_ctx)
616 			eventfd_ctx_put(dev->vqs[i]->error_ctx);
617 		if (dev->vqs[i]->kick)
618 			fput(dev->vqs[i]->kick);
619 		if (dev->vqs[i]->call_ctx)
620 			eventfd_ctx_put(dev->vqs[i]->call_ctx);
621 		vhost_vq_reset(dev, dev->vqs[i]);
622 	}
623 	vhost_dev_free_iovecs(dev);
624 	if (dev->log_ctx)
625 		eventfd_ctx_put(dev->log_ctx);
626 	dev->log_ctx = NULL;
627 	/* No one will access memory at this point */
628 	vhost_umem_clean(dev->umem);
629 	dev->umem = NULL;
630 	vhost_umem_clean(dev->iotlb);
631 	dev->iotlb = NULL;
632 	vhost_clear_msg(dev);
633 	wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
634 	WARN_ON(!llist_empty(&dev->work_list));
635 	if (dev->worker) {
636 		kthread_stop(dev->worker);
637 		dev->worker = NULL;
638 	}
639 	if (dev->mm)
640 		mmput(dev->mm);
641 	dev->mm = NULL;
642 }
643 EXPORT_SYMBOL_GPL(vhost_dev_cleanup);
644 
645 static int log_access_ok(void __user *log_base, u64 addr, unsigned long sz)
646 {
647 	u64 a = addr / VHOST_PAGE_SIZE / 8;
648 
649 	/* Make sure 64 bit math will not overflow. */
650 	if (a > ULONG_MAX - (unsigned long)log_base ||
651 	    a + (unsigned long)log_base > ULONG_MAX)
652 		return 0;
653 
654 	return access_ok(VERIFY_WRITE, log_base + a,
655 			 (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8);
656 }
657 
658 static bool vhost_overflow(u64 uaddr, u64 size)
659 {
660 	/* Make sure 64 bit math will not overflow. */
661 	return uaddr > ULONG_MAX || size > ULONG_MAX || uaddr > ULONG_MAX - size;
662 }
663 
664 /* Caller should have vq mutex and device mutex. */
665 static int vq_memory_access_ok(void __user *log_base, struct vhost_umem *umem,
666 			       int log_all)
667 {
668 	struct vhost_umem_node *node;
669 
670 	if (!umem)
671 		return 0;
672 
673 	list_for_each_entry(node, &umem->umem_list, link) {
674 		unsigned long a = node->userspace_addr;
675 
676 		if (vhost_overflow(node->userspace_addr, node->size))
677 			return 0;
678 
679 
680 		if (!access_ok(VERIFY_WRITE, (void __user *)a,
681 				    node->size))
682 			return 0;
683 		else if (log_all && !log_access_ok(log_base,
684 						   node->start,
685 						   node->size))
686 			return 0;
687 	}
688 	return 1;
689 }
690 
691 static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq,
692 					       u64 addr, unsigned int size,
693 					       int type)
694 {
695 	const struct vhost_umem_node *node = vq->meta_iotlb[type];
696 
697 	if (!node)
698 		return NULL;
699 
700 	return (void *)(uintptr_t)(node->userspace_addr + addr - node->start);
701 }
702 
703 /* Can we switch to this memory table? */
704 /* Caller should have device mutex but not vq mutex */
705 static int memory_access_ok(struct vhost_dev *d, struct vhost_umem *umem,
706 			    int log_all)
707 {
708 	int i;
709 
710 	for (i = 0; i < d->nvqs; ++i) {
711 		int ok;
712 		bool log;
713 
714 		mutex_lock(&d->vqs[i]->mutex);
715 		log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL);
716 		/* If ring is inactive, will check when it's enabled. */
717 		if (d->vqs[i]->private_data)
718 			ok = vq_memory_access_ok(d->vqs[i]->log_base,
719 						 umem, log);
720 		else
721 			ok = 1;
722 		mutex_unlock(&d->vqs[i]->mutex);
723 		if (!ok)
724 			return 0;
725 	}
726 	return 1;
727 }
728 
729 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
730 			  struct iovec iov[], int iov_size, int access);
731 
732 static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to,
733 			      const void *from, unsigned size)
734 {
735 	int ret;
736 
737 	if (!vq->iotlb)
738 		return __copy_to_user(to, from, size);
739 	else {
740 		/* This function should be called after iotlb
741 		 * prefetch, which means we're sure that all vq
742 		 * could be access through iotlb. So -EAGAIN should
743 		 * not happen in this case.
744 		 */
745 		struct iov_iter t;
746 		void __user *uaddr = vhost_vq_meta_fetch(vq,
747 				     (u64)(uintptr_t)to, size,
748 				     VHOST_ADDR_DESC);
749 
750 		if (uaddr)
751 			return __copy_to_user(uaddr, from, size);
752 
753 		ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov,
754 				     ARRAY_SIZE(vq->iotlb_iov),
755 				     VHOST_ACCESS_WO);
756 		if (ret < 0)
757 			goto out;
758 		iov_iter_init(&t, WRITE, vq->iotlb_iov, ret, size);
759 		ret = copy_to_iter(from, size, &t);
760 		if (ret == size)
761 			ret = 0;
762 	}
763 out:
764 	return ret;
765 }
766 
767 static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to,
768 				void __user *from, unsigned size)
769 {
770 	int ret;
771 
772 	if (!vq->iotlb)
773 		return __copy_from_user(to, from, size);
774 	else {
775 		/* This function should be called after iotlb
776 		 * prefetch, which means we're sure that vq
777 		 * could be access through iotlb. So -EAGAIN should
778 		 * not happen in this case.
779 		 */
780 		void __user *uaddr = vhost_vq_meta_fetch(vq,
781 				     (u64)(uintptr_t)from, size,
782 				     VHOST_ADDR_DESC);
783 		struct iov_iter f;
784 
785 		if (uaddr)
786 			return __copy_from_user(to, uaddr, size);
787 
788 		ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov,
789 				     ARRAY_SIZE(vq->iotlb_iov),
790 				     VHOST_ACCESS_RO);
791 		if (ret < 0) {
792 			vq_err(vq, "IOTLB translation failure: uaddr "
793 			       "%p size 0x%llx\n", from,
794 			       (unsigned long long) size);
795 			goto out;
796 		}
797 		iov_iter_init(&f, READ, vq->iotlb_iov, ret, size);
798 		ret = copy_from_iter(to, size, &f);
799 		if (ret == size)
800 			ret = 0;
801 	}
802 
803 out:
804 	return ret;
805 }
806 
807 static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq,
808 					  void __user *addr, unsigned int size,
809 					  int type)
810 {
811 	int ret;
812 
813 	ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov,
814 			     ARRAY_SIZE(vq->iotlb_iov),
815 			     VHOST_ACCESS_RO);
816 	if (ret < 0) {
817 		vq_err(vq, "IOTLB translation failure: uaddr "
818 			"%p size 0x%llx\n", addr,
819 			(unsigned long long) size);
820 		return NULL;
821 	}
822 
823 	if (ret != 1 || vq->iotlb_iov[0].iov_len != size) {
824 		vq_err(vq, "Non atomic userspace memory access: uaddr "
825 			"%p size 0x%llx\n", addr,
826 			(unsigned long long) size);
827 		return NULL;
828 	}
829 
830 	return vq->iotlb_iov[0].iov_base;
831 }
832 
833 /* This function should be called after iotlb
834  * prefetch, which means we're sure that vq
835  * could be access through iotlb. So -EAGAIN should
836  * not happen in this case.
837  */
838 static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq,
839 					    void *addr, unsigned int size,
840 					    int type)
841 {
842 	void __user *uaddr = vhost_vq_meta_fetch(vq,
843 			     (u64)(uintptr_t)addr, size, type);
844 	if (uaddr)
845 		return uaddr;
846 
847 	return __vhost_get_user_slow(vq, addr, size, type);
848 }
849 
850 #define vhost_put_user(vq, x, ptr)		\
851 ({ \
852 	int ret = -EFAULT; \
853 	if (!vq->iotlb) { \
854 		ret = __put_user(x, ptr); \
855 	} else { \
856 		__typeof__(ptr) to = \
857 			(__typeof__(ptr)) __vhost_get_user(vq, ptr,	\
858 					  sizeof(*ptr), VHOST_ADDR_USED); \
859 		if (to != NULL) \
860 			ret = __put_user(x, to); \
861 		else \
862 			ret = -EFAULT;	\
863 	} \
864 	ret; \
865 })
866 
867 #define vhost_get_user(vq, x, ptr, type)		\
868 ({ \
869 	int ret; \
870 	if (!vq->iotlb) { \
871 		ret = __get_user(x, ptr); \
872 	} else { \
873 		__typeof__(ptr) from = \
874 			(__typeof__(ptr)) __vhost_get_user(vq, ptr, \
875 							   sizeof(*ptr), \
876 							   type); \
877 		if (from != NULL) \
878 			ret = __get_user(x, from); \
879 		else \
880 			ret = -EFAULT; \
881 	} \
882 	ret; \
883 })
884 
885 #define vhost_get_avail(vq, x, ptr) \
886 	vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL)
887 
888 #define vhost_get_used(vq, x, ptr) \
889 	vhost_get_user(vq, x, ptr, VHOST_ADDR_USED)
890 
891 static void vhost_dev_lock_vqs(struct vhost_dev *d)
892 {
893 	int i = 0;
894 	for (i = 0; i < d->nvqs; ++i)
895 		mutex_lock_nested(&d->vqs[i]->mutex, i);
896 }
897 
898 static void vhost_dev_unlock_vqs(struct vhost_dev *d)
899 {
900 	int i = 0;
901 	for (i = 0; i < d->nvqs; ++i)
902 		mutex_unlock(&d->vqs[i]->mutex);
903 }
904 
905 static int vhost_new_umem_range(struct vhost_umem *umem,
906 				u64 start, u64 size, u64 end,
907 				u64 userspace_addr, int perm)
908 {
909 	struct vhost_umem_node *tmp, *node = kmalloc(sizeof(*node), GFP_ATOMIC);
910 
911 	if (!node)
912 		return -ENOMEM;
913 
914 	if (umem->numem == max_iotlb_entries) {
915 		tmp = list_first_entry(&umem->umem_list, typeof(*tmp), link);
916 		vhost_umem_free(umem, tmp);
917 	}
918 
919 	node->start = start;
920 	node->size = size;
921 	node->last = end;
922 	node->userspace_addr = userspace_addr;
923 	node->perm = perm;
924 	INIT_LIST_HEAD(&node->link);
925 	list_add_tail(&node->link, &umem->umem_list);
926 	vhost_umem_interval_tree_insert(node, &umem->umem_tree);
927 	umem->numem++;
928 
929 	return 0;
930 }
931 
932 static void vhost_del_umem_range(struct vhost_umem *umem,
933 				 u64 start, u64 end)
934 {
935 	struct vhost_umem_node *node;
936 
937 	while ((node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
938 							   start, end)))
939 		vhost_umem_free(umem, node);
940 }
941 
942 static void vhost_iotlb_notify_vq(struct vhost_dev *d,
943 				  struct vhost_iotlb_msg *msg)
944 {
945 	struct vhost_msg_node *node, *n;
946 
947 	spin_lock(&d->iotlb_lock);
948 
949 	list_for_each_entry_safe(node, n, &d->pending_list, node) {
950 		struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb;
951 		if (msg->iova <= vq_msg->iova &&
952 		    msg->iova + msg->size - 1 > vq_msg->iova &&
953 		    vq_msg->type == VHOST_IOTLB_MISS) {
954 			vhost_poll_queue(&node->vq->poll);
955 			list_del(&node->node);
956 			kfree(node);
957 		}
958 	}
959 
960 	spin_unlock(&d->iotlb_lock);
961 }
962 
963 static int umem_access_ok(u64 uaddr, u64 size, int access)
964 {
965 	unsigned long a = uaddr;
966 
967 	/* Make sure 64 bit math will not overflow. */
968 	if (vhost_overflow(uaddr, size))
969 		return -EFAULT;
970 
971 	if ((access & VHOST_ACCESS_RO) &&
972 	    !access_ok(VERIFY_READ, (void __user *)a, size))
973 		return -EFAULT;
974 	if ((access & VHOST_ACCESS_WO) &&
975 	    !access_ok(VERIFY_WRITE, (void __user *)a, size))
976 		return -EFAULT;
977 	return 0;
978 }
979 
980 static int vhost_process_iotlb_msg(struct vhost_dev *dev,
981 				   struct vhost_iotlb_msg *msg)
982 {
983 	int ret = 0;
984 
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 	return ret;
1021 }
1022 ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
1023 			     struct iov_iter *from)
1024 {
1025 	struct vhost_msg_node node;
1026 	unsigned size = sizeof(struct vhost_msg);
1027 	size_t ret;
1028 	int err;
1029 
1030 	if (iov_iter_count(from) < size)
1031 		return 0;
1032 	ret = copy_from_iter(&node.msg, size, from);
1033 	if (ret != size)
1034 		goto done;
1035 
1036 	switch (node.msg.type) {
1037 	case VHOST_IOTLB_MSG:
1038 		err = vhost_process_iotlb_msg(dev, &node.msg.iotlb);
1039 		if (err)
1040 			ret = err;
1041 		break;
1042 	default:
1043 		ret = -EINVAL;
1044 		break;
1045 	}
1046 
1047 done:
1048 	return ret;
1049 }
1050 EXPORT_SYMBOL(vhost_chr_write_iter);
1051 
1052 __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
1053 			    poll_table *wait)
1054 {
1055 	__poll_t mask = 0;
1056 
1057 	poll_wait(file, &dev->wait, wait);
1058 
1059 	if (!list_empty(&dev->read_list))
1060 		mask |= EPOLLIN | EPOLLRDNORM;
1061 
1062 	return mask;
1063 }
1064 EXPORT_SYMBOL(vhost_chr_poll);
1065 
1066 ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
1067 			    int noblock)
1068 {
1069 	DEFINE_WAIT(wait);
1070 	struct vhost_msg_node *node;
1071 	ssize_t ret = 0;
1072 	unsigned size = sizeof(struct vhost_msg);
1073 
1074 	if (iov_iter_count(to) < size)
1075 		return 0;
1076 
1077 	while (1) {
1078 		if (!noblock)
1079 			prepare_to_wait(&dev->wait, &wait,
1080 					TASK_INTERRUPTIBLE);
1081 
1082 		node = vhost_dequeue_msg(dev, &dev->read_list);
1083 		if (node)
1084 			break;
1085 		if (noblock) {
1086 			ret = -EAGAIN;
1087 			break;
1088 		}
1089 		if (signal_pending(current)) {
1090 			ret = -ERESTARTSYS;
1091 			break;
1092 		}
1093 		if (!dev->iotlb) {
1094 			ret = -EBADFD;
1095 			break;
1096 		}
1097 
1098 		schedule();
1099 	}
1100 
1101 	if (!noblock)
1102 		finish_wait(&dev->wait, &wait);
1103 
1104 	if (node) {
1105 		ret = copy_to_iter(&node->msg, size, to);
1106 
1107 		if (ret != size || node->msg.type != VHOST_IOTLB_MISS) {
1108 			kfree(node);
1109 			return ret;
1110 		}
1111 
1112 		vhost_enqueue_msg(dev, &dev->pending_list, node);
1113 	}
1114 
1115 	return ret;
1116 }
1117 EXPORT_SYMBOL_GPL(vhost_chr_read_iter);
1118 
1119 static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access)
1120 {
1121 	struct vhost_dev *dev = vq->dev;
1122 	struct vhost_msg_node *node;
1123 	struct vhost_iotlb_msg *msg;
1124 
1125 	node = vhost_new_msg(vq, VHOST_IOTLB_MISS);
1126 	if (!node)
1127 		return -ENOMEM;
1128 
1129 	msg = &node->msg.iotlb;
1130 	msg->type = VHOST_IOTLB_MISS;
1131 	msg->iova = iova;
1132 	msg->perm = access;
1133 
1134 	vhost_enqueue_msg(dev, &dev->read_list, node);
1135 
1136 	return 0;
1137 }
1138 
1139 static int vq_access_ok(struct vhost_virtqueue *vq, unsigned int num,
1140 			struct vring_desc __user *desc,
1141 			struct vring_avail __user *avail,
1142 			struct vring_used __user *used)
1143 
1144 {
1145 	size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
1146 
1147 	return access_ok(VERIFY_READ, desc, num * sizeof *desc) &&
1148 	       access_ok(VERIFY_READ, avail,
1149 			 sizeof *avail + num * sizeof *avail->ring + s) &&
1150 	       access_ok(VERIFY_WRITE, used,
1151 			sizeof *used + num * sizeof *used->ring + s);
1152 }
1153 
1154 static void vhost_vq_meta_update(struct vhost_virtqueue *vq,
1155 				 const struct vhost_umem_node *node,
1156 				 int type)
1157 {
1158 	int access = (type == VHOST_ADDR_USED) ?
1159 		     VHOST_ACCESS_WO : VHOST_ACCESS_RO;
1160 
1161 	if (likely(node->perm & access))
1162 		vq->meta_iotlb[type] = node;
1163 }
1164 
1165 static int iotlb_access_ok(struct vhost_virtqueue *vq,
1166 			   int access, u64 addr, u64 len, int type)
1167 {
1168 	const struct vhost_umem_node *node;
1169 	struct vhost_umem *umem = vq->iotlb;
1170 	u64 s = 0, size, orig_addr = addr, last = addr + len - 1;
1171 
1172 	if (vhost_vq_meta_fetch(vq, addr, len, type))
1173 		return true;
1174 
1175 	while (len > s) {
1176 		node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1177 							   addr,
1178 							   last);
1179 		if (node == NULL || node->start > addr) {
1180 			vhost_iotlb_miss(vq, addr, access);
1181 			return false;
1182 		} else if (!(node->perm & access)) {
1183 			/* Report the possible access violation by
1184 			 * request another translation from userspace.
1185 			 */
1186 			return false;
1187 		}
1188 
1189 		size = node->size - addr + node->start;
1190 
1191 		if (orig_addr == addr && size >= len)
1192 			vhost_vq_meta_update(vq, node, type);
1193 
1194 		s += size;
1195 		addr += size;
1196 	}
1197 
1198 	return true;
1199 }
1200 
1201 int vq_iotlb_prefetch(struct vhost_virtqueue *vq)
1202 {
1203 	size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
1204 	unsigned int num = vq->num;
1205 
1206 	if (!vq->iotlb)
1207 		return 1;
1208 
1209 	return iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->desc,
1210 			       num * sizeof(*vq->desc), VHOST_ADDR_DESC) &&
1211 	       iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->avail,
1212 			       sizeof *vq->avail +
1213 			       num * sizeof(*vq->avail->ring) + s,
1214 			       VHOST_ADDR_AVAIL) &&
1215 	       iotlb_access_ok(vq, VHOST_ACCESS_WO, (u64)(uintptr_t)vq->used,
1216 			       sizeof *vq->used +
1217 			       num * sizeof(*vq->used->ring) + s,
1218 			       VHOST_ADDR_USED);
1219 }
1220 EXPORT_SYMBOL_GPL(vq_iotlb_prefetch);
1221 
1222 /* Can we log writes? */
1223 /* Caller should have device mutex but not vq mutex */
1224 int vhost_log_access_ok(struct vhost_dev *dev)
1225 {
1226 	return memory_access_ok(dev, dev->umem, 1);
1227 }
1228 EXPORT_SYMBOL_GPL(vhost_log_access_ok);
1229 
1230 /* Verify access for write logging. */
1231 /* Caller should have vq mutex and device mutex */
1232 static int vq_log_access_ok(struct vhost_virtqueue *vq,
1233 			    void __user *log_base)
1234 {
1235 	size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
1236 
1237 	return vq_memory_access_ok(log_base, vq->umem,
1238 				   vhost_has_feature(vq, VHOST_F_LOG_ALL)) &&
1239 		(!vq->log_used || log_access_ok(log_base, vq->log_addr,
1240 					sizeof *vq->used +
1241 					vq->num * sizeof *vq->used->ring + s));
1242 }
1243 
1244 /* Can we start vq? */
1245 /* Caller should have vq mutex and device mutex */
1246 int vhost_vq_access_ok(struct vhost_virtqueue *vq)
1247 {
1248 	if (vq->iotlb) {
1249 		/* When device IOTLB was used, the access validation
1250 		 * will be validated during prefetching.
1251 		 */
1252 		return 1;
1253 	}
1254 	return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used) &&
1255 		vq_log_access_ok(vq, vq->log_base);
1256 }
1257 EXPORT_SYMBOL_GPL(vhost_vq_access_ok);
1258 
1259 static struct vhost_umem *vhost_umem_alloc(void)
1260 {
1261 	struct vhost_umem *umem = kvzalloc(sizeof(*umem), GFP_KERNEL);
1262 
1263 	if (!umem)
1264 		return NULL;
1265 
1266 	umem->umem_tree = RB_ROOT_CACHED;
1267 	umem->numem = 0;
1268 	INIT_LIST_HEAD(&umem->umem_list);
1269 
1270 	return umem;
1271 }
1272 
1273 static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)
1274 {
1275 	struct vhost_memory mem, *newmem;
1276 	struct vhost_memory_region *region;
1277 	struct vhost_umem *newumem, *oldumem;
1278 	unsigned long size = offsetof(struct vhost_memory, regions);
1279 	int i;
1280 
1281 	if (copy_from_user(&mem, m, size))
1282 		return -EFAULT;
1283 	if (mem.padding)
1284 		return -EOPNOTSUPP;
1285 	if (mem.nregions > max_mem_regions)
1286 		return -E2BIG;
1287 	newmem = kvzalloc(size + mem.nregions * sizeof(*m->regions), GFP_KERNEL);
1288 	if (!newmem)
1289 		return -ENOMEM;
1290 
1291 	memcpy(newmem, &mem, size);
1292 	if (copy_from_user(newmem->regions, m->regions,
1293 			   mem.nregions * sizeof *m->regions)) {
1294 		kvfree(newmem);
1295 		return -EFAULT;
1296 	}
1297 
1298 	newumem = vhost_umem_alloc();
1299 	if (!newumem) {
1300 		kvfree(newmem);
1301 		return -ENOMEM;
1302 	}
1303 
1304 	for (region = newmem->regions;
1305 	     region < newmem->regions + mem.nregions;
1306 	     region++) {
1307 		if (vhost_new_umem_range(newumem,
1308 					 region->guest_phys_addr,
1309 					 region->memory_size,
1310 					 region->guest_phys_addr +
1311 					 region->memory_size - 1,
1312 					 region->userspace_addr,
1313 					 VHOST_ACCESS_RW))
1314 			goto err;
1315 	}
1316 
1317 	if (!memory_access_ok(d, newumem, 0))
1318 		goto err;
1319 
1320 	oldumem = d->umem;
1321 	d->umem = newumem;
1322 
1323 	/* All memory accesses are done under some VQ mutex. */
1324 	for (i = 0; i < d->nvqs; ++i) {
1325 		mutex_lock(&d->vqs[i]->mutex);
1326 		d->vqs[i]->umem = newumem;
1327 		mutex_unlock(&d->vqs[i]->mutex);
1328 	}
1329 
1330 	kvfree(newmem);
1331 	vhost_umem_clean(oldumem);
1332 	return 0;
1333 
1334 err:
1335 	vhost_umem_clean(newumem);
1336 	kvfree(newmem);
1337 	return -EFAULT;
1338 }
1339 
1340 long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp)
1341 {
1342 	struct file *eventfp, *filep = NULL;
1343 	bool pollstart = false, pollstop = false;
1344 	struct eventfd_ctx *ctx = NULL;
1345 	u32 __user *idxp = argp;
1346 	struct vhost_virtqueue *vq;
1347 	struct vhost_vring_state s;
1348 	struct vhost_vring_file f;
1349 	struct vhost_vring_addr a;
1350 	u32 idx;
1351 	long r;
1352 
1353 	r = get_user(idx, idxp);
1354 	if (r < 0)
1355 		return r;
1356 	if (idx >= d->nvqs)
1357 		return -ENOBUFS;
1358 
1359 	vq = d->vqs[idx];
1360 
1361 	mutex_lock(&vq->mutex);
1362 
1363 	switch (ioctl) {
1364 	case VHOST_SET_VRING_NUM:
1365 		/* Resizing ring with an active backend?
1366 		 * You don't want to do that. */
1367 		if (vq->private_data) {
1368 			r = -EBUSY;
1369 			break;
1370 		}
1371 		if (copy_from_user(&s, argp, sizeof s)) {
1372 			r = -EFAULT;
1373 			break;
1374 		}
1375 		if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) {
1376 			r = -EINVAL;
1377 			break;
1378 		}
1379 		vq->num = s.num;
1380 		break;
1381 	case VHOST_SET_VRING_BASE:
1382 		/* Moving base with an active backend?
1383 		 * You don't want to do that. */
1384 		if (vq->private_data) {
1385 			r = -EBUSY;
1386 			break;
1387 		}
1388 		if (copy_from_user(&s, argp, sizeof s)) {
1389 			r = -EFAULT;
1390 			break;
1391 		}
1392 		if (s.num > 0xffff) {
1393 			r = -EINVAL;
1394 			break;
1395 		}
1396 		vq->last_avail_idx = s.num;
1397 		/* Forget the cached index value. */
1398 		vq->avail_idx = vq->last_avail_idx;
1399 		break;
1400 	case VHOST_GET_VRING_BASE:
1401 		s.index = idx;
1402 		s.num = vq->last_avail_idx;
1403 		if (copy_to_user(argp, &s, sizeof s))
1404 			r = -EFAULT;
1405 		break;
1406 	case VHOST_SET_VRING_ADDR:
1407 		if (copy_from_user(&a, argp, sizeof a)) {
1408 			r = -EFAULT;
1409 			break;
1410 		}
1411 		if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) {
1412 			r = -EOPNOTSUPP;
1413 			break;
1414 		}
1415 		/* For 32bit, verify that the top 32bits of the user
1416 		   data are set to zero. */
1417 		if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr ||
1418 		    (u64)(unsigned long)a.used_user_addr != a.used_user_addr ||
1419 		    (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) {
1420 			r = -EFAULT;
1421 			break;
1422 		}
1423 
1424 		/* Make sure it's safe to cast pointers to vring types. */
1425 		BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE);
1426 		BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE);
1427 		if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) ||
1428 		    (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) ||
1429 		    (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1))) {
1430 			r = -EINVAL;
1431 			break;
1432 		}
1433 
1434 		/* We only verify access here if backend is configured.
1435 		 * If it is not, we don't as size might not have been setup.
1436 		 * We will verify when backend is configured. */
1437 		if (vq->private_data) {
1438 			if (!vq_access_ok(vq, vq->num,
1439 				(void __user *)(unsigned long)a.desc_user_addr,
1440 				(void __user *)(unsigned long)a.avail_user_addr,
1441 				(void __user *)(unsigned long)a.used_user_addr)) {
1442 				r = -EINVAL;
1443 				break;
1444 			}
1445 
1446 			/* Also validate log access for used ring if enabled. */
1447 			if ((a.flags & (0x1 << VHOST_VRING_F_LOG)) &&
1448 			    !log_access_ok(vq->log_base, a.log_guest_addr,
1449 					   sizeof *vq->used +
1450 					   vq->num * sizeof *vq->used->ring)) {
1451 				r = -EINVAL;
1452 				break;
1453 			}
1454 		}
1455 
1456 		vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG));
1457 		vq->desc = (void __user *)(unsigned long)a.desc_user_addr;
1458 		vq->avail = (void __user *)(unsigned long)a.avail_user_addr;
1459 		vq->log_addr = a.log_guest_addr;
1460 		vq->used = (void __user *)(unsigned long)a.used_user_addr;
1461 		break;
1462 	case VHOST_SET_VRING_KICK:
1463 		if (copy_from_user(&f, argp, sizeof f)) {
1464 			r = -EFAULT;
1465 			break;
1466 		}
1467 		eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd);
1468 		if (IS_ERR(eventfp)) {
1469 			r = PTR_ERR(eventfp);
1470 			break;
1471 		}
1472 		if (eventfp != vq->kick) {
1473 			pollstop = (filep = vq->kick) != NULL;
1474 			pollstart = (vq->kick = eventfp) != NULL;
1475 		} else
1476 			filep = eventfp;
1477 		break;
1478 	case VHOST_SET_VRING_CALL:
1479 		if (copy_from_user(&f, argp, sizeof f)) {
1480 			r = -EFAULT;
1481 			break;
1482 		}
1483 		ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd);
1484 		if (IS_ERR(ctx)) {
1485 			r = PTR_ERR(ctx);
1486 			break;
1487 		}
1488 		swap(ctx, vq->call_ctx);
1489 		break;
1490 	case VHOST_SET_VRING_ERR:
1491 		if (copy_from_user(&f, argp, sizeof f)) {
1492 			r = -EFAULT;
1493 			break;
1494 		}
1495 		ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd);
1496 		if (IS_ERR(ctx)) {
1497 			r = PTR_ERR(ctx);
1498 			break;
1499 		}
1500 		swap(ctx, vq->error_ctx);
1501 		break;
1502 	case VHOST_SET_VRING_ENDIAN:
1503 		r = vhost_set_vring_endian(vq, argp);
1504 		break;
1505 	case VHOST_GET_VRING_ENDIAN:
1506 		r = vhost_get_vring_endian(vq, idx, argp);
1507 		break;
1508 	case VHOST_SET_VRING_BUSYLOOP_TIMEOUT:
1509 		if (copy_from_user(&s, argp, sizeof(s))) {
1510 			r = -EFAULT;
1511 			break;
1512 		}
1513 		vq->busyloop_timeout = s.num;
1514 		break;
1515 	case VHOST_GET_VRING_BUSYLOOP_TIMEOUT:
1516 		s.index = idx;
1517 		s.num = vq->busyloop_timeout;
1518 		if (copy_to_user(argp, &s, sizeof(s)))
1519 			r = -EFAULT;
1520 		break;
1521 	default:
1522 		r = -ENOIOCTLCMD;
1523 	}
1524 
1525 	if (pollstop && vq->handle_kick)
1526 		vhost_poll_stop(&vq->poll);
1527 
1528 	if (!IS_ERR_OR_NULL(ctx))
1529 		eventfd_ctx_put(ctx);
1530 	if (filep)
1531 		fput(filep);
1532 
1533 	if (pollstart && vq->handle_kick)
1534 		r = vhost_poll_start(&vq->poll, vq->kick);
1535 
1536 	mutex_unlock(&vq->mutex);
1537 
1538 	if (pollstop && vq->handle_kick)
1539 		vhost_poll_flush(&vq->poll);
1540 	return r;
1541 }
1542 EXPORT_SYMBOL_GPL(vhost_vring_ioctl);
1543 
1544 int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled)
1545 {
1546 	struct vhost_umem *niotlb, *oiotlb;
1547 	int i;
1548 
1549 	niotlb = vhost_umem_alloc();
1550 	if (!niotlb)
1551 		return -ENOMEM;
1552 
1553 	oiotlb = d->iotlb;
1554 	d->iotlb = niotlb;
1555 
1556 	for (i = 0; i < d->nvqs; ++i) {
1557 		mutex_lock(&d->vqs[i]->mutex);
1558 		d->vqs[i]->iotlb = niotlb;
1559 		mutex_unlock(&d->vqs[i]->mutex);
1560 	}
1561 
1562 	vhost_umem_clean(oiotlb);
1563 
1564 	return 0;
1565 }
1566 EXPORT_SYMBOL_GPL(vhost_init_device_iotlb);
1567 
1568 /* Caller must have device mutex */
1569 long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
1570 {
1571 	struct eventfd_ctx *ctx;
1572 	u64 p;
1573 	long r;
1574 	int i, fd;
1575 
1576 	/* If you are not the owner, you can become one */
1577 	if (ioctl == VHOST_SET_OWNER) {
1578 		r = vhost_dev_set_owner(d);
1579 		goto done;
1580 	}
1581 
1582 	/* You must be the owner to do anything else */
1583 	r = vhost_dev_check_owner(d);
1584 	if (r)
1585 		goto done;
1586 
1587 	switch (ioctl) {
1588 	case VHOST_SET_MEM_TABLE:
1589 		r = vhost_set_memory(d, argp);
1590 		break;
1591 	case VHOST_SET_LOG_BASE:
1592 		if (copy_from_user(&p, argp, sizeof p)) {
1593 			r = -EFAULT;
1594 			break;
1595 		}
1596 		if ((u64)(unsigned long)p != p) {
1597 			r = -EFAULT;
1598 			break;
1599 		}
1600 		for (i = 0; i < d->nvqs; ++i) {
1601 			struct vhost_virtqueue *vq;
1602 			void __user *base = (void __user *)(unsigned long)p;
1603 			vq = d->vqs[i];
1604 			mutex_lock(&vq->mutex);
1605 			/* If ring is inactive, will check when it's enabled. */
1606 			if (vq->private_data && !vq_log_access_ok(vq, base))
1607 				r = -EFAULT;
1608 			else
1609 				vq->log_base = base;
1610 			mutex_unlock(&vq->mutex);
1611 		}
1612 		break;
1613 	case VHOST_SET_LOG_FD:
1614 		r = get_user(fd, (int __user *)argp);
1615 		if (r < 0)
1616 			break;
1617 		ctx = fd == -1 ? NULL : eventfd_ctx_fdget(fd);
1618 		if (IS_ERR(ctx)) {
1619 			r = PTR_ERR(ctx);
1620 			break;
1621 		}
1622 		swap(ctx, d->log_ctx);
1623 		for (i = 0; i < d->nvqs; ++i) {
1624 			mutex_lock(&d->vqs[i]->mutex);
1625 			d->vqs[i]->log_ctx = d->log_ctx;
1626 			mutex_unlock(&d->vqs[i]->mutex);
1627 		}
1628 		if (ctx)
1629 			eventfd_ctx_put(ctx);
1630 		break;
1631 	default:
1632 		r = -ENOIOCTLCMD;
1633 		break;
1634 	}
1635 done:
1636 	return r;
1637 }
1638 EXPORT_SYMBOL_GPL(vhost_dev_ioctl);
1639 
1640 /* TODO: This is really inefficient.  We need something like get_user()
1641  * (instruction directly accesses the data, with an exception table entry
1642  * returning -EFAULT). See Documentation/x86/exception-tables.txt.
1643  */
1644 static int set_bit_to_user(int nr, void __user *addr)
1645 {
1646 	unsigned long log = (unsigned long)addr;
1647 	struct page *page;
1648 	void *base;
1649 	int bit = nr + (log % PAGE_SIZE) * 8;
1650 	int r;
1651 
1652 	r = get_user_pages_fast(log, 1, 1, &page);
1653 	if (r < 0)
1654 		return r;
1655 	BUG_ON(r != 1);
1656 	base = kmap_atomic(page);
1657 	set_bit(bit, base);
1658 	kunmap_atomic(base);
1659 	set_page_dirty_lock(page);
1660 	put_page(page);
1661 	return 0;
1662 }
1663 
1664 static int log_write(void __user *log_base,
1665 		     u64 write_address, u64 write_length)
1666 {
1667 	u64 write_page = write_address / VHOST_PAGE_SIZE;
1668 	int r;
1669 
1670 	if (!write_length)
1671 		return 0;
1672 	write_length += write_address % VHOST_PAGE_SIZE;
1673 	for (;;) {
1674 		u64 base = (u64)(unsigned long)log_base;
1675 		u64 log = base + write_page / 8;
1676 		int bit = write_page % 8;
1677 		if ((u64)(unsigned long)log != log)
1678 			return -EFAULT;
1679 		r = set_bit_to_user(bit, (void __user *)(unsigned long)log);
1680 		if (r < 0)
1681 			return r;
1682 		if (write_length <= VHOST_PAGE_SIZE)
1683 			break;
1684 		write_length -= VHOST_PAGE_SIZE;
1685 		write_page += 1;
1686 	}
1687 	return r;
1688 }
1689 
1690 int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
1691 		    unsigned int log_num, u64 len)
1692 {
1693 	int i, r;
1694 
1695 	/* Make sure data written is seen before log. */
1696 	smp_wmb();
1697 	for (i = 0; i < log_num; ++i) {
1698 		u64 l = min(log[i].len, len);
1699 		r = log_write(vq->log_base, log[i].addr, l);
1700 		if (r < 0)
1701 			return r;
1702 		len -= l;
1703 		if (!len) {
1704 			if (vq->log_ctx)
1705 				eventfd_signal(vq->log_ctx, 1);
1706 			return 0;
1707 		}
1708 	}
1709 	/* Length written exceeds what we have stored. This is a bug. */
1710 	BUG();
1711 	return 0;
1712 }
1713 EXPORT_SYMBOL_GPL(vhost_log_write);
1714 
1715 static int vhost_update_used_flags(struct vhost_virtqueue *vq)
1716 {
1717 	void __user *used;
1718 	if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags),
1719 			   &vq->used->flags) < 0)
1720 		return -EFAULT;
1721 	if (unlikely(vq->log_used)) {
1722 		/* Make sure the flag is seen before log. */
1723 		smp_wmb();
1724 		/* Log used flag write. */
1725 		used = &vq->used->flags;
1726 		log_write(vq->log_base, vq->log_addr +
1727 			  (used - (void __user *)vq->used),
1728 			  sizeof vq->used->flags);
1729 		if (vq->log_ctx)
1730 			eventfd_signal(vq->log_ctx, 1);
1731 	}
1732 	return 0;
1733 }
1734 
1735 static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event)
1736 {
1737 	if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx),
1738 			   vhost_avail_event(vq)))
1739 		return -EFAULT;
1740 	if (unlikely(vq->log_used)) {
1741 		void __user *used;
1742 		/* Make sure the event is seen before log. */
1743 		smp_wmb();
1744 		/* Log avail event write */
1745 		used = vhost_avail_event(vq);
1746 		log_write(vq->log_base, vq->log_addr +
1747 			  (used - (void __user *)vq->used),
1748 			  sizeof *vhost_avail_event(vq));
1749 		if (vq->log_ctx)
1750 			eventfd_signal(vq->log_ctx, 1);
1751 	}
1752 	return 0;
1753 }
1754 
1755 int vhost_vq_init_access(struct vhost_virtqueue *vq)
1756 {
1757 	__virtio16 last_used_idx;
1758 	int r;
1759 	bool is_le = vq->is_le;
1760 
1761 	if (!vq->private_data)
1762 		return 0;
1763 
1764 	vhost_init_is_le(vq);
1765 
1766 	r = vhost_update_used_flags(vq);
1767 	if (r)
1768 		goto err;
1769 	vq->signalled_used_valid = false;
1770 	if (!vq->iotlb &&
1771 	    !access_ok(VERIFY_READ, &vq->used->idx, sizeof vq->used->idx)) {
1772 		r = -EFAULT;
1773 		goto err;
1774 	}
1775 	r = vhost_get_used(vq, last_used_idx, &vq->used->idx);
1776 	if (r) {
1777 		vq_err(vq, "Can't access used idx at %p\n",
1778 		       &vq->used->idx);
1779 		goto err;
1780 	}
1781 	vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx);
1782 	return 0;
1783 
1784 err:
1785 	vq->is_le = is_le;
1786 	return r;
1787 }
1788 EXPORT_SYMBOL_GPL(vhost_vq_init_access);
1789 
1790 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
1791 			  struct iovec iov[], int iov_size, int access)
1792 {
1793 	const struct vhost_umem_node *node;
1794 	struct vhost_dev *dev = vq->dev;
1795 	struct vhost_umem *umem = dev->iotlb ? dev->iotlb : dev->umem;
1796 	struct iovec *_iov;
1797 	u64 s = 0;
1798 	int ret = 0;
1799 
1800 	while ((u64)len > s) {
1801 		u64 size;
1802 		if (unlikely(ret >= iov_size)) {
1803 			ret = -ENOBUFS;
1804 			break;
1805 		}
1806 
1807 		node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1808 							addr, addr + len - 1);
1809 		if (node == NULL || node->start > addr) {
1810 			if (umem != dev->iotlb) {
1811 				ret = -EFAULT;
1812 				break;
1813 			}
1814 			ret = -EAGAIN;
1815 			break;
1816 		} else if (!(node->perm & access)) {
1817 			ret = -EPERM;
1818 			break;
1819 		}
1820 
1821 		_iov = iov + ret;
1822 		size = node->size - addr + node->start;
1823 		_iov->iov_len = min((u64)len - s, size);
1824 		_iov->iov_base = (void __user *)(unsigned long)
1825 			(node->userspace_addr + addr - node->start);
1826 		s += size;
1827 		addr += size;
1828 		++ret;
1829 	}
1830 
1831 	if (ret == -EAGAIN)
1832 		vhost_iotlb_miss(vq, addr, access);
1833 	return ret;
1834 }
1835 
1836 /* Each buffer in the virtqueues is actually a chain of descriptors.  This
1837  * function returns the next descriptor in the chain,
1838  * or -1U if we're at the end. */
1839 static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc)
1840 {
1841 	unsigned int next;
1842 
1843 	/* If this descriptor says it doesn't chain, we're done. */
1844 	if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT)))
1845 		return -1U;
1846 
1847 	/* Check they're not leading us off end of descriptors. */
1848 	next = vhost16_to_cpu(vq, READ_ONCE(desc->next));
1849 	return next;
1850 }
1851 
1852 static int get_indirect(struct vhost_virtqueue *vq,
1853 			struct iovec iov[], unsigned int iov_size,
1854 			unsigned int *out_num, unsigned int *in_num,
1855 			struct vhost_log *log, unsigned int *log_num,
1856 			struct vring_desc *indirect)
1857 {
1858 	struct vring_desc desc;
1859 	unsigned int i = 0, count, found = 0;
1860 	u32 len = vhost32_to_cpu(vq, indirect->len);
1861 	struct iov_iter from;
1862 	int ret, access;
1863 
1864 	/* Sanity check */
1865 	if (unlikely(len % sizeof desc)) {
1866 		vq_err(vq, "Invalid length in indirect descriptor: "
1867 		       "len 0x%llx not multiple of 0x%zx\n",
1868 		       (unsigned long long)len,
1869 		       sizeof desc);
1870 		return -EINVAL;
1871 	}
1872 
1873 	ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect,
1874 			     UIO_MAXIOV, VHOST_ACCESS_RO);
1875 	if (unlikely(ret < 0)) {
1876 		if (ret != -EAGAIN)
1877 			vq_err(vq, "Translation failure %d in indirect.\n", ret);
1878 		return ret;
1879 	}
1880 	iov_iter_init(&from, READ, vq->indirect, ret, len);
1881 
1882 	/* We will use the result as an address to read from, so most
1883 	 * architectures only need a compiler barrier here. */
1884 	read_barrier_depends();
1885 
1886 	count = len / sizeof desc;
1887 	/* Buffers are chained via a 16 bit next field, so
1888 	 * we can have at most 2^16 of these. */
1889 	if (unlikely(count > USHRT_MAX + 1)) {
1890 		vq_err(vq, "Indirect buffer length too big: %d\n",
1891 		       indirect->len);
1892 		return -E2BIG;
1893 	}
1894 
1895 	do {
1896 		unsigned iov_count = *in_num + *out_num;
1897 		if (unlikely(++found > count)) {
1898 			vq_err(vq, "Loop detected: last one at %u "
1899 			       "indirect size %u\n",
1900 			       i, count);
1901 			return -EINVAL;
1902 		}
1903 		if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) {
1904 			vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n",
1905 			       i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
1906 			return -EINVAL;
1907 		}
1908 		if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) {
1909 			vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n",
1910 			       i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
1911 			return -EINVAL;
1912 		}
1913 
1914 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
1915 			access = VHOST_ACCESS_WO;
1916 		else
1917 			access = VHOST_ACCESS_RO;
1918 
1919 		ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
1920 				     vhost32_to_cpu(vq, desc.len), iov + iov_count,
1921 				     iov_size - iov_count, access);
1922 		if (unlikely(ret < 0)) {
1923 			if (ret != -EAGAIN)
1924 				vq_err(vq, "Translation failure %d indirect idx %d\n",
1925 					ret, i);
1926 			return ret;
1927 		}
1928 		/* If this is an input descriptor, increment that count. */
1929 		if (access == VHOST_ACCESS_WO) {
1930 			*in_num += ret;
1931 			if (unlikely(log)) {
1932 				log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
1933 				log[*log_num].len = vhost32_to_cpu(vq, desc.len);
1934 				++*log_num;
1935 			}
1936 		} else {
1937 			/* If it's an output descriptor, they're all supposed
1938 			 * to come before any input descriptors. */
1939 			if (unlikely(*in_num)) {
1940 				vq_err(vq, "Indirect descriptor "
1941 				       "has out after in: idx %d\n", i);
1942 				return -EINVAL;
1943 			}
1944 			*out_num += ret;
1945 		}
1946 	} while ((i = next_desc(vq, &desc)) != -1);
1947 	return 0;
1948 }
1949 
1950 /* This looks in the virtqueue and for the first available buffer, and converts
1951  * it to an iovec for convenient access.  Since descriptors consist of some
1952  * number of output then some number of input descriptors, it's actually two
1953  * iovecs, but we pack them into one and note how many of each there were.
1954  *
1955  * This function returns the descriptor number found, or vq->num (which is
1956  * never a valid descriptor number) if none was found.  A negative code is
1957  * returned on error. */
1958 int vhost_get_vq_desc(struct vhost_virtqueue *vq,
1959 		      struct iovec iov[], unsigned int iov_size,
1960 		      unsigned int *out_num, unsigned int *in_num,
1961 		      struct vhost_log *log, unsigned int *log_num)
1962 {
1963 	struct vring_desc desc;
1964 	unsigned int i, head, found = 0;
1965 	u16 last_avail_idx;
1966 	__virtio16 avail_idx;
1967 	__virtio16 ring_head;
1968 	int ret, access;
1969 
1970 	/* Check it isn't doing very strange things with descriptor numbers. */
1971 	last_avail_idx = vq->last_avail_idx;
1972 
1973 	if (vq->avail_idx == vq->last_avail_idx) {
1974 		if (unlikely(vhost_get_avail(vq, avail_idx, &vq->avail->idx))) {
1975 			vq_err(vq, "Failed to access avail idx at %p\n",
1976 				&vq->avail->idx);
1977 			return -EFAULT;
1978 		}
1979 		vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
1980 
1981 		if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) {
1982 			vq_err(vq, "Guest moved used index from %u to %u",
1983 				last_avail_idx, vq->avail_idx);
1984 			return -EFAULT;
1985 		}
1986 
1987 		/* If there's nothing new since last we looked, return
1988 		 * invalid.
1989 		 */
1990 		if (vq->avail_idx == last_avail_idx)
1991 			return vq->num;
1992 
1993 		/* Only get avail ring entries after they have been
1994 		 * exposed by guest.
1995 		 */
1996 		smp_rmb();
1997 	}
1998 
1999 	/* Grab the next descriptor number they're advertising, and increment
2000 	 * the index we've seen. */
2001 	if (unlikely(vhost_get_avail(vq, ring_head,
2002 		     &vq->avail->ring[last_avail_idx & (vq->num - 1)]))) {
2003 		vq_err(vq, "Failed to read head: idx %d address %p\n",
2004 		       last_avail_idx,
2005 		       &vq->avail->ring[last_avail_idx % vq->num]);
2006 		return -EFAULT;
2007 	}
2008 
2009 	head = vhost16_to_cpu(vq, ring_head);
2010 
2011 	/* If their number is silly, that's an error. */
2012 	if (unlikely(head >= vq->num)) {
2013 		vq_err(vq, "Guest says index %u > %u is available",
2014 		       head, vq->num);
2015 		return -EINVAL;
2016 	}
2017 
2018 	/* When we start there are none of either input nor output. */
2019 	*out_num = *in_num = 0;
2020 	if (unlikely(log))
2021 		*log_num = 0;
2022 
2023 	i = head;
2024 	do {
2025 		unsigned iov_count = *in_num + *out_num;
2026 		if (unlikely(i >= vq->num)) {
2027 			vq_err(vq, "Desc index is %u > %u, head = %u",
2028 			       i, vq->num, head);
2029 			return -EINVAL;
2030 		}
2031 		if (unlikely(++found > vq->num)) {
2032 			vq_err(vq, "Loop detected: last one at %u "
2033 			       "vq size %u head %u\n",
2034 			       i, vq->num, head);
2035 			return -EINVAL;
2036 		}
2037 		ret = vhost_copy_from_user(vq, &desc, vq->desc + i,
2038 					   sizeof desc);
2039 		if (unlikely(ret)) {
2040 			vq_err(vq, "Failed to get descriptor: idx %d addr %p\n",
2041 			       i, vq->desc + i);
2042 			return -EFAULT;
2043 		}
2044 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) {
2045 			ret = get_indirect(vq, iov, iov_size,
2046 					   out_num, in_num,
2047 					   log, log_num, &desc);
2048 			if (unlikely(ret < 0)) {
2049 				if (ret != -EAGAIN)
2050 					vq_err(vq, "Failure detected "
2051 						"in indirect descriptor at idx %d\n", i);
2052 				return ret;
2053 			}
2054 			continue;
2055 		}
2056 
2057 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2058 			access = VHOST_ACCESS_WO;
2059 		else
2060 			access = VHOST_ACCESS_RO;
2061 		ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2062 				     vhost32_to_cpu(vq, desc.len), iov + iov_count,
2063 				     iov_size - iov_count, access);
2064 		if (unlikely(ret < 0)) {
2065 			if (ret != -EAGAIN)
2066 				vq_err(vq, "Translation failure %d descriptor idx %d\n",
2067 					ret, i);
2068 			return ret;
2069 		}
2070 		if (access == VHOST_ACCESS_WO) {
2071 			/* If this is an input descriptor,
2072 			 * increment that count. */
2073 			*in_num += ret;
2074 			if (unlikely(log)) {
2075 				log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2076 				log[*log_num].len = vhost32_to_cpu(vq, desc.len);
2077 				++*log_num;
2078 			}
2079 		} else {
2080 			/* If it's an output descriptor, they're all supposed
2081 			 * to come before any input descriptors. */
2082 			if (unlikely(*in_num)) {
2083 				vq_err(vq, "Descriptor has out after in: "
2084 				       "idx %d\n", i);
2085 				return -EINVAL;
2086 			}
2087 			*out_num += ret;
2088 		}
2089 	} while ((i = next_desc(vq, &desc)) != -1);
2090 
2091 	/* On success, increment avail index. */
2092 	vq->last_avail_idx++;
2093 
2094 	/* Assume notifications from guest are disabled at this point,
2095 	 * if they aren't we would need to update avail_event index. */
2096 	BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY));
2097 	return head;
2098 }
2099 EXPORT_SYMBOL_GPL(vhost_get_vq_desc);
2100 
2101 /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */
2102 void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n)
2103 {
2104 	vq->last_avail_idx -= n;
2105 }
2106 EXPORT_SYMBOL_GPL(vhost_discard_vq_desc);
2107 
2108 /* After we've used one of their buffers, we tell them about it.  We'll then
2109  * want to notify the guest, using eventfd. */
2110 int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len)
2111 {
2112 	struct vring_used_elem heads = {
2113 		cpu_to_vhost32(vq, head),
2114 		cpu_to_vhost32(vq, len)
2115 	};
2116 
2117 	return vhost_add_used_n(vq, &heads, 1);
2118 }
2119 EXPORT_SYMBOL_GPL(vhost_add_used);
2120 
2121 static int __vhost_add_used_n(struct vhost_virtqueue *vq,
2122 			    struct vring_used_elem *heads,
2123 			    unsigned count)
2124 {
2125 	struct vring_used_elem __user *used;
2126 	u16 old, new;
2127 	int start;
2128 
2129 	start = vq->last_used_idx & (vq->num - 1);
2130 	used = vq->used->ring + start;
2131 	if (count == 1) {
2132 		if (vhost_put_user(vq, heads[0].id, &used->id)) {
2133 			vq_err(vq, "Failed to write used id");
2134 			return -EFAULT;
2135 		}
2136 		if (vhost_put_user(vq, heads[0].len, &used->len)) {
2137 			vq_err(vq, "Failed to write used len");
2138 			return -EFAULT;
2139 		}
2140 	} else if (vhost_copy_to_user(vq, used, heads, count * sizeof *used)) {
2141 		vq_err(vq, "Failed to write used");
2142 		return -EFAULT;
2143 	}
2144 	if (unlikely(vq->log_used)) {
2145 		/* Make sure data is seen before log. */
2146 		smp_wmb();
2147 		/* Log used ring entry write. */
2148 		log_write(vq->log_base,
2149 			  vq->log_addr +
2150 			   ((void __user *)used - (void __user *)vq->used),
2151 			  count * sizeof *used);
2152 	}
2153 	old = vq->last_used_idx;
2154 	new = (vq->last_used_idx += count);
2155 	/* If the driver never bothers to signal in a very long while,
2156 	 * used index might wrap around. If that happens, invalidate
2157 	 * signalled_used index we stored. TODO: make sure driver
2158 	 * signals at least once in 2^16 and remove this. */
2159 	if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old)))
2160 		vq->signalled_used_valid = false;
2161 	return 0;
2162 }
2163 
2164 /* After we've used one of their buffers, we tell them about it.  We'll then
2165  * want to notify the guest, using eventfd. */
2166 int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads,
2167 		     unsigned count)
2168 {
2169 	int start, n, r;
2170 
2171 	start = vq->last_used_idx & (vq->num - 1);
2172 	n = vq->num - start;
2173 	if (n < count) {
2174 		r = __vhost_add_used_n(vq, heads, n);
2175 		if (r < 0)
2176 			return r;
2177 		heads += n;
2178 		count -= n;
2179 	}
2180 	r = __vhost_add_used_n(vq, heads, count);
2181 
2182 	/* Make sure buffer is written before we update index. */
2183 	smp_wmb();
2184 	if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx),
2185 			   &vq->used->idx)) {
2186 		vq_err(vq, "Failed to increment used idx");
2187 		return -EFAULT;
2188 	}
2189 	if (unlikely(vq->log_used)) {
2190 		/* Log used index update. */
2191 		log_write(vq->log_base,
2192 			  vq->log_addr + offsetof(struct vring_used, idx),
2193 			  sizeof vq->used->idx);
2194 		if (vq->log_ctx)
2195 			eventfd_signal(vq->log_ctx, 1);
2196 	}
2197 	return r;
2198 }
2199 EXPORT_SYMBOL_GPL(vhost_add_used_n);
2200 
2201 static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2202 {
2203 	__u16 old, new;
2204 	__virtio16 event;
2205 	bool v;
2206 	/* Flush out used index updates. This is paired
2207 	 * with the barrier that the Guest executes when enabling
2208 	 * interrupts. */
2209 	smp_mb();
2210 
2211 	if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2212 	    unlikely(vq->avail_idx == vq->last_avail_idx))
2213 		return true;
2214 
2215 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2216 		__virtio16 flags;
2217 		if (vhost_get_avail(vq, flags, &vq->avail->flags)) {
2218 			vq_err(vq, "Failed to get flags");
2219 			return true;
2220 		}
2221 		return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT));
2222 	}
2223 	old = vq->signalled_used;
2224 	v = vq->signalled_used_valid;
2225 	new = vq->signalled_used = vq->last_used_idx;
2226 	vq->signalled_used_valid = true;
2227 
2228 	if (unlikely(!v))
2229 		return true;
2230 
2231 	if (vhost_get_avail(vq, event, vhost_used_event(vq))) {
2232 		vq_err(vq, "Failed to get used event idx");
2233 		return true;
2234 	}
2235 	return vring_need_event(vhost16_to_cpu(vq, event), new, old);
2236 }
2237 
2238 /* This actually signals the guest, using eventfd. */
2239 void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2240 {
2241 	/* Signal the Guest tell them we used something up. */
2242 	if (vq->call_ctx && vhost_notify(dev, vq))
2243 		eventfd_signal(vq->call_ctx, 1);
2244 }
2245 EXPORT_SYMBOL_GPL(vhost_signal);
2246 
2247 /* And here's the combo meal deal.  Supersize me! */
2248 void vhost_add_used_and_signal(struct vhost_dev *dev,
2249 			       struct vhost_virtqueue *vq,
2250 			       unsigned int head, int len)
2251 {
2252 	vhost_add_used(vq, head, len);
2253 	vhost_signal(dev, vq);
2254 }
2255 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal);
2256 
2257 /* multi-buffer version of vhost_add_used_and_signal */
2258 void vhost_add_used_and_signal_n(struct vhost_dev *dev,
2259 				 struct vhost_virtqueue *vq,
2260 				 struct vring_used_elem *heads, unsigned count)
2261 {
2262 	vhost_add_used_n(vq, heads, count);
2263 	vhost_signal(dev, vq);
2264 }
2265 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n);
2266 
2267 /* return true if we're sure that avaiable ring is empty */
2268 bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2269 {
2270 	__virtio16 avail_idx;
2271 	int r;
2272 
2273 	if (vq->avail_idx != vq->last_avail_idx)
2274 		return false;
2275 
2276 	r = vhost_get_avail(vq, avail_idx, &vq->avail->idx);
2277 	if (unlikely(r))
2278 		return false;
2279 	vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
2280 
2281 	return vq->avail_idx == vq->last_avail_idx;
2282 }
2283 EXPORT_SYMBOL_GPL(vhost_vq_avail_empty);
2284 
2285 /* OK, now we need to know about added descriptors. */
2286 bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2287 {
2288 	__virtio16 avail_idx;
2289 	int r;
2290 
2291 	if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY))
2292 		return false;
2293 	vq->used_flags &= ~VRING_USED_F_NO_NOTIFY;
2294 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2295 		r = vhost_update_used_flags(vq);
2296 		if (r) {
2297 			vq_err(vq, "Failed to enable notification at %p: %d\n",
2298 			       &vq->used->flags, r);
2299 			return false;
2300 		}
2301 	} else {
2302 		r = vhost_update_avail_event(vq, vq->avail_idx);
2303 		if (r) {
2304 			vq_err(vq, "Failed to update avail event index at %p: %d\n",
2305 			       vhost_avail_event(vq), r);
2306 			return false;
2307 		}
2308 	}
2309 	/* They could have slipped one in as we were doing that: make
2310 	 * sure it's written, then check again. */
2311 	smp_mb();
2312 	r = vhost_get_avail(vq, avail_idx, &vq->avail->idx);
2313 	if (r) {
2314 		vq_err(vq, "Failed to check avail idx at %p: %d\n",
2315 		       &vq->avail->idx, r);
2316 		return false;
2317 	}
2318 
2319 	return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx;
2320 }
2321 EXPORT_SYMBOL_GPL(vhost_enable_notify);
2322 
2323 /* We don't need to be notified again. */
2324 void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2325 {
2326 	int r;
2327 
2328 	if (vq->used_flags & VRING_USED_F_NO_NOTIFY)
2329 		return;
2330 	vq->used_flags |= VRING_USED_F_NO_NOTIFY;
2331 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2332 		r = vhost_update_used_flags(vq);
2333 		if (r)
2334 			vq_err(vq, "Failed to enable notification at %p: %d\n",
2335 			       &vq->used->flags, r);
2336 	}
2337 }
2338 EXPORT_SYMBOL_GPL(vhost_disable_notify);
2339 
2340 /* Create a new message. */
2341 struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type)
2342 {
2343 	struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL);
2344 	if (!node)
2345 		return NULL;
2346 	node->vq = vq;
2347 	node->msg.type = type;
2348 	return node;
2349 }
2350 EXPORT_SYMBOL_GPL(vhost_new_msg);
2351 
2352 void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head,
2353 		       struct vhost_msg_node *node)
2354 {
2355 	spin_lock(&dev->iotlb_lock);
2356 	list_add_tail(&node->node, head);
2357 	spin_unlock(&dev->iotlb_lock);
2358 
2359 	wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
2360 }
2361 EXPORT_SYMBOL_GPL(vhost_enqueue_msg);
2362 
2363 struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
2364 					 struct list_head *head)
2365 {
2366 	struct vhost_msg_node *node = NULL;
2367 
2368 	spin_lock(&dev->iotlb_lock);
2369 	if (!list_empty(head)) {
2370 		node = list_first_entry(head, struct vhost_msg_node,
2371 					node);
2372 		list_del(&node->node);
2373 	}
2374 	spin_unlock(&dev->iotlb_lock);
2375 
2376 	return node;
2377 }
2378 EXPORT_SYMBOL_GPL(vhost_dequeue_msg);
2379 
2380 
2381 static int __init vhost_init(void)
2382 {
2383 	return 0;
2384 }
2385 
2386 static void __exit vhost_exit(void)
2387 {
2388 }
2389 
2390 module_init(vhost_init);
2391 module_exit(vhost_exit);
2392 
2393 MODULE_VERSION("0.0.1");
2394 MODULE_LICENSE("GPL v2");
2395 MODULE_AUTHOR("Michael S. Tsirkin");
2396 MODULE_DESCRIPTION("Host kernel accelerator for virtio");
2397