1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (C) 2009 Red Hat, Inc.
3  * Copyright (C) 2006 Rusty Russell IBM Corporation
4  *
5  * Author: Michael S. Tsirkin <mst@redhat.com>
6  *
7  * Inspiration, some code, and most witty comments come from
8  * Documentation/virtual/lguest/lguest.c, by Rusty Russell
9  *
10  * Generic code for virtio server in host kernel.
11  */
12 
13 #include <linux/eventfd.h>
14 #include <linux/vhost.h>
15 #include <linux/uio.h>
16 #include <linux/mm.h>
17 #include <linux/miscdevice.h>
18 #include <linux/mutex.h>
19 #include <linux/poll.h>
20 #include <linux/file.h>
21 #include <linux/highmem.h>
22 #include <linux/slab.h>
23 #include <linux/vmalloc.h>
24 #include <linux/kthread.h>
25 #include <linux/cgroup.h>
26 #include <linux/module.h>
27 #include <linux/sort.h>
28 #include <linux/sched/mm.h>
29 #include <linux/sched/signal.h>
30 #include <linux/interval_tree_generic.h>
31 #include <linux/nospec.h>
32 #include <linux/kcov.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 #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
vhost_disable_cross_endian(struct vhost_virtqueue * vq)53 static void vhost_disable_cross_endian(struct vhost_virtqueue *vq)
54 {
55 	vq->user_be = !virtio_legacy_is_little_endian();
56 }
57 
vhost_enable_cross_endian_big(struct vhost_virtqueue * vq)58 static void vhost_enable_cross_endian_big(struct vhost_virtqueue *vq)
59 {
60 	vq->user_be = true;
61 }
62 
vhost_enable_cross_endian_little(struct vhost_virtqueue * vq)63 static void vhost_enable_cross_endian_little(struct vhost_virtqueue *vq)
64 {
65 	vq->user_be = false;
66 }
67 
vhost_set_vring_endian(struct vhost_virtqueue * vq,int __user * argp)68 static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp)
69 {
70 	struct vhost_vring_state s;
71 
72 	if (vq->private_data)
73 		return -EBUSY;
74 
75 	if (copy_from_user(&s, argp, sizeof(s)))
76 		return -EFAULT;
77 
78 	if (s.num != VHOST_VRING_LITTLE_ENDIAN &&
79 	    s.num != VHOST_VRING_BIG_ENDIAN)
80 		return -EINVAL;
81 
82 	if (s.num == VHOST_VRING_BIG_ENDIAN)
83 		vhost_enable_cross_endian_big(vq);
84 	else
85 		vhost_enable_cross_endian_little(vq);
86 
87 	return 0;
88 }
89 
vhost_get_vring_endian(struct vhost_virtqueue * vq,u32 idx,int __user * argp)90 static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx,
91 				   int __user *argp)
92 {
93 	struct vhost_vring_state s = {
94 		.index = idx,
95 		.num = vq->user_be
96 	};
97 
98 	if (copy_to_user(argp, &s, sizeof(s)))
99 		return -EFAULT;
100 
101 	return 0;
102 }
103 
vhost_init_is_le(struct vhost_virtqueue * vq)104 static void vhost_init_is_le(struct vhost_virtqueue *vq)
105 {
106 	/* Note for legacy virtio: user_be is initialized at reset time
107 	 * according to the host endianness. If userspace does not set an
108 	 * explicit endianness, the default behavior is native endian, as
109 	 * expected by legacy virtio.
110 	 */
111 	vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) || !vq->user_be;
112 }
113 #else
vhost_disable_cross_endian(struct vhost_virtqueue * vq)114 static void vhost_disable_cross_endian(struct vhost_virtqueue *vq)
115 {
116 }
117 
vhost_set_vring_endian(struct vhost_virtqueue * vq,int __user * argp)118 static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp)
119 {
120 	return -ENOIOCTLCMD;
121 }
122 
vhost_get_vring_endian(struct vhost_virtqueue * vq,u32 idx,int __user * argp)123 static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx,
124 				   int __user *argp)
125 {
126 	return -ENOIOCTLCMD;
127 }
128 
vhost_init_is_le(struct vhost_virtqueue * vq)129 static void vhost_init_is_le(struct vhost_virtqueue *vq)
130 {
131 	vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1)
132 		|| virtio_legacy_is_little_endian();
133 }
134 #endif /* CONFIG_VHOST_CROSS_ENDIAN_LEGACY */
135 
vhost_reset_is_le(struct vhost_virtqueue * vq)136 static void vhost_reset_is_le(struct vhost_virtqueue *vq)
137 {
138 	vhost_init_is_le(vq);
139 }
140 
141 struct vhost_flush_struct {
142 	struct vhost_work work;
143 	struct completion wait_event;
144 };
145 
vhost_flush_work(struct vhost_work * work)146 static void vhost_flush_work(struct vhost_work *work)
147 {
148 	struct vhost_flush_struct *s;
149 
150 	s = container_of(work, struct vhost_flush_struct, work);
151 	complete(&s->wait_event);
152 }
153 
vhost_poll_func(struct file * file,wait_queue_head_t * wqh,poll_table * pt)154 static void vhost_poll_func(struct file *file, wait_queue_head_t *wqh,
155 			    poll_table *pt)
156 {
157 	struct vhost_poll *poll;
158 
159 	poll = container_of(pt, struct vhost_poll, table);
160 	poll->wqh = wqh;
161 	add_wait_queue(wqh, &poll->wait);
162 }
163 
vhost_poll_wakeup(wait_queue_entry_t * wait,unsigned mode,int sync,void * key)164 static int vhost_poll_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync,
165 			     void *key)
166 {
167 	struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait);
168 	struct vhost_work *work = &poll->work;
169 
170 	if (!(key_to_poll(key) & poll->mask))
171 		return 0;
172 
173 	if (!poll->dev->use_worker)
174 		work->fn(work);
175 	else
176 		vhost_poll_queue(poll);
177 
178 	return 0;
179 }
180 
vhost_work_init(struct vhost_work * work,vhost_work_fn_t fn)181 void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn)
182 {
183 	clear_bit(VHOST_WORK_QUEUED, &work->flags);
184 	work->fn = fn;
185 }
186 EXPORT_SYMBOL_GPL(vhost_work_init);
187 
188 /* Init poll structure */
vhost_poll_init(struct vhost_poll * poll,vhost_work_fn_t fn,__poll_t mask,struct vhost_dev * dev)189 void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
190 		     __poll_t mask, struct vhost_dev *dev)
191 {
192 	init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup);
193 	init_poll_funcptr(&poll->table, vhost_poll_func);
194 	poll->mask = mask;
195 	poll->dev = dev;
196 	poll->wqh = NULL;
197 
198 	vhost_work_init(&poll->work, fn);
199 }
200 EXPORT_SYMBOL_GPL(vhost_poll_init);
201 
202 /* Start polling a file. We add ourselves to file's wait queue. The caller must
203  * keep a reference to a file until after vhost_poll_stop is called. */
vhost_poll_start(struct vhost_poll * poll,struct file * file)204 int vhost_poll_start(struct vhost_poll *poll, struct file *file)
205 {
206 	__poll_t mask;
207 
208 	if (poll->wqh)
209 		return 0;
210 
211 	mask = vfs_poll(file, &poll->table);
212 	if (mask)
213 		vhost_poll_wakeup(&poll->wait, 0, 0, poll_to_key(mask));
214 	if (mask & EPOLLERR) {
215 		vhost_poll_stop(poll);
216 		return -EINVAL;
217 	}
218 
219 	return 0;
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. */
vhost_poll_stop(struct vhost_poll * poll)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 
vhost_dev_flush(struct vhost_dev * dev)234 void vhost_dev_flush(struct vhost_dev *dev)
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_dev_flush);
247 
vhost_work_queue(struct vhost_dev * dev,struct vhost_work * work)248 void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work)
249 {
250 	if (!dev->worker)
251 		return;
252 
253 	if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) {
254 		/* We can only add the work to the list after we're
255 		 * sure it was not in the list.
256 		 * test_and_set_bit() implies a memory barrier.
257 		 */
258 		llist_add(&work->node, &dev->work_list);
259 		wake_up_process(dev->worker);
260 	}
261 }
262 EXPORT_SYMBOL_GPL(vhost_work_queue);
263 
264 /* A lockless hint for busy polling code to exit the loop */
vhost_has_work(struct vhost_dev * dev)265 bool vhost_has_work(struct vhost_dev *dev)
266 {
267 	return !llist_empty(&dev->work_list);
268 }
269 EXPORT_SYMBOL_GPL(vhost_has_work);
270 
vhost_poll_queue(struct vhost_poll * poll)271 void vhost_poll_queue(struct vhost_poll *poll)
272 {
273 	vhost_work_queue(poll->dev, &poll->work);
274 }
275 EXPORT_SYMBOL_GPL(vhost_poll_queue);
276 
__vhost_vq_meta_reset(struct vhost_virtqueue * vq)277 static void __vhost_vq_meta_reset(struct vhost_virtqueue *vq)
278 {
279 	int j;
280 
281 	for (j = 0; j < VHOST_NUM_ADDRS; j++)
282 		vq->meta_iotlb[j] = NULL;
283 }
284 
vhost_vq_meta_reset(struct vhost_dev * d)285 static void vhost_vq_meta_reset(struct vhost_dev *d)
286 {
287 	int i;
288 
289 	for (i = 0; i < d->nvqs; ++i)
290 		__vhost_vq_meta_reset(d->vqs[i]);
291 }
292 
vhost_vring_call_reset(struct vhost_vring_call * call_ctx)293 static void vhost_vring_call_reset(struct vhost_vring_call *call_ctx)
294 {
295 	call_ctx->ctx = NULL;
296 	memset(&call_ctx->producer, 0x0, sizeof(struct irq_bypass_producer));
297 }
298 
vhost_vq_is_setup(struct vhost_virtqueue * vq)299 bool vhost_vq_is_setup(struct vhost_virtqueue *vq)
300 {
301 	return vq->avail && vq->desc && vq->used && vhost_vq_access_ok(vq);
302 }
303 EXPORT_SYMBOL_GPL(vhost_vq_is_setup);
304 
vhost_vq_reset(struct vhost_dev * dev,struct vhost_virtqueue * vq)305 static void vhost_vq_reset(struct vhost_dev *dev,
306 			   struct vhost_virtqueue *vq)
307 {
308 	vq->num = 1;
309 	vq->desc = NULL;
310 	vq->avail = NULL;
311 	vq->used = NULL;
312 	vq->last_avail_idx = 0;
313 	vq->avail_idx = 0;
314 	vq->last_used_idx = 0;
315 	vq->signalled_used = 0;
316 	vq->signalled_used_valid = false;
317 	vq->used_flags = 0;
318 	vq->log_used = false;
319 	vq->log_addr = -1ull;
320 	vq->private_data = NULL;
321 	vq->acked_features = 0;
322 	vq->acked_backend_features = 0;
323 	vq->log_base = NULL;
324 	vq->error_ctx = NULL;
325 	vq->kick = NULL;
326 	vq->log_ctx = NULL;
327 	vhost_disable_cross_endian(vq);
328 	vhost_reset_is_le(vq);
329 	vq->busyloop_timeout = 0;
330 	vq->umem = NULL;
331 	vq->iotlb = NULL;
332 	vhost_vring_call_reset(&vq->call_ctx);
333 	__vhost_vq_meta_reset(vq);
334 }
335 
vhost_worker(void * data)336 static int vhost_worker(void *data)
337 {
338 	struct vhost_dev *dev = data;
339 	struct vhost_work *work, *work_next;
340 	struct llist_node *node;
341 
342 	kthread_use_mm(dev->mm);
343 
344 	for (;;) {
345 		/* mb paired w/ kthread_stop */
346 		set_current_state(TASK_INTERRUPTIBLE);
347 
348 		if (kthread_should_stop()) {
349 			__set_current_state(TASK_RUNNING);
350 			break;
351 		}
352 
353 		node = llist_del_all(&dev->work_list);
354 		if (!node)
355 			schedule();
356 
357 		node = llist_reverse_order(node);
358 		/* make sure flag is seen after deletion */
359 		smp_wmb();
360 		llist_for_each_entry_safe(work, work_next, node, node) {
361 			clear_bit(VHOST_WORK_QUEUED, &work->flags);
362 			__set_current_state(TASK_RUNNING);
363 			kcov_remote_start_common(dev->kcov_handle);
364 			work->fn(work);
365 			kcov_remote_stop();
366 			if (need_resched())
367 				schedule();
368 		}
369 	}
370 	kthread_unuse_mm(dev->mm);
371 	return 0;
372 }
373 
vhost_vq_free_iovecs(struct vhost_virtqueue * vq)374 static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq)
375 {
376 	kfree(vq->indirect);
377 	vq->indirect = NULL;
378 	kfree(vq->log);
379 	vq->log = NULL;
380 	kfree(vq->heads);
381 	vq->heads = NULL;
382 }
383 
384 /* Helper to allocate iovec buffers for all vqs. */
vhost_dev_alloc_iovecs(struct vhost_dev * dev)385 static long vhost_dev_alloc_iovecs(struct vhost_dev *dev)
386 {
387 	struct vhost_virtqueue *vq;
388 	int i;
389 
390 	for (i = 0; i < dev->nvqs; ++i) {
391 		vq = dev->vqs[i];
392 		vq->indirect = kmalloc_array(UIO_MAXIOV,
393 					     sizeof(*vq->indirect),
394 					     GFP_KERNEL);
395 		vq->log = kmalloc_array(dev->iov_limit, sizeof(*vq->log),
396 					GFP_KERNEL);
397 		vq->heads = kmalloc_array(dev->iov_limit, sizeof(*vq->heads),
398 					  GFP_KERNEL);
399 		if (!vq->indirect || !vq->log || !vq->heads)
400 			goto err_nomem;
401 	}
402 	return 0;
403 
404 err_nomem:
405 	for (; i >= 0; --i)
406 		vhost_vq_free_iovecs(dev->vqs[i]);
407 	return -ENOMEM;
408 }
409 
vhost_dev_free_iovecs(struct vhost_dev * dev)410 static void vhost_dev_free_iovecs(struct vhost_dev *dev)
411 {
412 	int i;
413 
414 	for (i = 0; i < dev->nvqs; ++i)
415 		vhost_vq_free_iovecs(dev->vqs[i]);
416 }
417 
vhost_exceeds_weight(struct vhost_virtqueue * vq,int pkts,int total_len)418 bool vhost_exceeds_weight(struct vhost_virtqueue *vq,
419 			  int pkts, int total_len)
420 {
421 	struct vhost_dev *dev = vq->dev;
422 
423 	if ((dev->byte_weight && total_len >= dev->byte_weight) ||
424 	    pkts >= dev->weight) {
425 		vhost_poll_queue(&vq->poll);
426 		return true;
427 	}
428 
429 	return false;
430 }
431 EXPORT_SYMBOL_GPL(vhost_exceeds_weight);
432 
vhost_get_avail_size(struct vhost_virtqueue * vq,unsigned int num)433 static size_t vhost_get_avail_size(struct vhost_virtqueue *vq,
434 				   unsigned int num)
435 {
436 	size_t event __maybe_unused =
437 	       vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
438 
439 	return sizeof(*vq->avail) +
440 	       sizeof(*vq->avail->ring) * num + event;
441 }
442 
vhost_get_used_size(struct vhost_virtqueue * vq,unsigned int num)443 static size_t vhost_get_used_size(struct vhost_virtqueue *vq,
444 				  unsigned int num)
445 {
446 	size_t event __maybe_unused =
447 	       vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
448 
449 	return sizeof(*vq->used) +
450 	       sizeof(*vq->used->ring) * num + event;
451 }
452 
vhost_get_desc_size(struct vhost_virtqueue * vq,unsigned int num)453 static size_t vhost_get_desc_size(struct vhost_virtqueue *vq,
454 				  unsigned int num)
455 {
456 	return sizeof(*vq->desc) * num;
457 }
458 
vhost_dev_init(struct vhost_dev * dev,struct vhost_virtqueue ** vqs,int nvqs,int iov_limit,int weight,int byte_weight,bool use_worker,int (* msg_handler)(struct vhost_dev * dev,u32 asid,struct vhost_iotlb_msg * msg))459 void vhost_dev_init(struct vhost_dev *dev,
460 		    struct vhost_virtqueue **vqs, int nvqs,
461 		    int iov_limit, int weight, int byte_weight,
462 		    bool use_worker,
463 		    int (*msg_handler)(struct vhost_dev *dev, u32 asid,
464 				       struct vhost_iotlb_msg *msg))
465 {
466 	struct vhost_virtqueue *vq;
467 	int i;
468 
469 	dev->vqs = vqs;
470 	dev->nvqs = nvqs;
471 	mutex_init(&dev->mutex);
472 	dev->log_ctx = NULL;
473 	dev->umem = NULL;
474 	dev->iotlb = NULL;
475 	dev->mm = NULL;
476 	dev->worker = NULL;
477 	dev->iov_limit = iov_limit;
478 	dev->weight = weight;
479 	dev->byte_weight = byte_weight;
480 	dev->use_worker = use_worker;
481 	dev->msg_handler = msg_handler;
482 	init_llist_head(&dev->work_list);
483 	init_waitqueue_head(&dev->wait);
484 	INIT_LIST_HEAD(&dev->read_list);
485 	INIT_LIST_HEAD(&dev->pending_list);
486 	spin_lock_init(&dev->iotlb_lock);
487 
488 
489 	for (i = 0; i < dev->nvqs; ++i) {
490 		vq = dev->vqs[i];
491 		vq->log = NULL;
492 		vq->indirect = NULL;
493 		vq->heads = NULL;
494 		vq->dev = dev;
495 		mutex_init(&vq->mutex);
496 		vhost_vq_reset(dev, vq);
497 		if (vq->handle_kick)
498 			vhost_poll_init(&vq->poll, vq->handle_kick,
499 					EPOLLIN, dev);
500 	}
501 }
502 EXPORT_SYMBOL_GPL(vhost_dev_init);
503 
504 /* Caller should have device mutex */
vhost_dev_check_owner(struct vhost_dev * dev)505 long vhost_dev_check_owner(struct vhost_dev *dev)
506 {
507 	/* Are you the owner? If not, I don't think you mean to do that */
508 	return dev->mm == current->mm ? 0 : -EPERM;
509 }
510 EXPORT_SYMBOL_GPL(vhost_dev_check_owner);
511 
512 struct vhost_attach_cgroups_struct {
513 	struct vhost_work work;
514 	struct task_struct *owner;
515 	int ret;
516 };
517 
vhost_attach_cgroups_work(struct vhost_work * work)518 static void vhost_attach_cgroups_work(struct vhost_work *work)
519 {
520 	struct vhost_attach_cgroups_struct *s;
521 
522 	s = container_of(work, struct vhost_attach_cgroups_struct, work);
523 	s->ret = cgroup_attach_task_all(s->owner, current);
524 }
525 
vhost_attach_cgroups(struct vhost_dev * dev)526 static int vhost_attach_cgroups(struct vhost_dev *dev)
527 {
528 	struct vhost_attach_cgroups_struct attach;
529 
530 	attach.owner = current;
531 	vhost_work_init(&attach.work, vhost_attach_cgroups_work);
532 	vhost_work_queue(dev, &attach.work);
533 	vhost_dev_flush(dev);
534 	return attach.ret;
535 }
536 
537 /* Caller should have device mutex */
vhost_dev_has_owner(struct vhost_dev * dev)538 bool vhost_dev_has_owner(struct vhost_dev *dev)
539 {
540 	return dev->mm;
541 }
542 EXPORT_SYMBOL_GPL(vhost_dev_has_owner);
543 
vhost_attach_mm(struct vhost_dev * dev)544 static void vhost_attach_mm(struct vhost_dev *dev)
545 {
546 	/* No owner, become one */
547 	if (dev->use_worker) {
548 		dev->mm = get_task_mm(current);
549 	} else {
550 		/* vDPA device does not use worker thead, so there's
551 		 * no need to hold the address space for mm. This help
552 		 * to avoid deadlock in the case of mmap() which may
553 		 * held the refcnt of the file and depends on release
554 		 * method to remove vma.
555 		 */
556 		dev->mm = current->mm;
557 		mmgrab(dev->mm);
558 	}
559 }
560 
vhost_detach_mm(struct vhost_dev * dev)561 static void vhost_detach_mm(struct vhost_dev *dev)
562 {
563 	if (!dev->mm)
564 		return;
565 
566 	if (dev->use_worker)
567 		mmput(dev->mm);
568 	else
569 		mmdrop(dev->mm);
570 
571 	dev->mm = NULL;
572 }
573 
574 /* Caller should have device mutex */
vhost_dev_set_owner(struct vhost_dev * dev)575 long vhost_dev_set_owner(struct vhost_dev *dev)
576 {
577 	struct task_struct *worker;
578 	int err;
579 
580 	/* Is there an owner already? */
581 	if (vhost_dev_has_owner(dev)) {
582 		err = -EBUSY;
583 		goto err_mm;
584 	}
585 
586 	vhost_attach_mm(dev);
587 
588 	dev->kcov_handle = kcov_common_handle();
589 	if (dev->use_worker) {
590 		worker = kthread_create(vhost_worker, dev,
591 					"vhost-%d", current->pid);
592 		if (IS_ERR(worker)) {
593 			err = PTR_ERR(worker);
594 			goto err_worker;
595 		}
596 
597 		dev->worker = worker;
598 		wake_up_process(worker); /* avoid contributing to loadavg */
599 
600 		err = vhost_attach_cgroups(dev);
601 		if (err)
602 			goto err_cgroup;
603 	}
604 
605 	err = vhost_dev_alloc_iovecs(dev);
606 	if (err)
607 		goto err_cgroup;
608 
609 	return 0;
610 err_cgroup:
611 	if (dev->worker) {
612 		kthread_stop(dev->worker);
613 		dev->worker = NULL;
614 	}
615 err_worker:
616 	vhost_detach_mm(dev);
617 	dev->kcov_handle = 0;
618 err_mm:
619 	return err;
620 }
621 EXPORT_SYMBOL_GPL(vhost_dev_set_owner);
622 
iotlb_alloc(void)623 static struct vhost_iotlb *iotlb_alloc(void)
624 {
625 	return vhost_iotlb_alloc(max_iotlb_entries,
626 				 VHOST_IOTLB_FLAG_RETIRE);
627 }
628 
vhost_dev_reset_owner_prepare(void)629 struct vhost_iotlb *vhost_dev_reset_owner_prepare(void)
630 {
631 	return iotlb_alloc();
632 }
633 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare);
634 
635 /* Caller should have device mutex */
vhost_dev_reset_owner(struct vhost_dev * dev,struct vhost_iotlb * umem)636 void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_iotlb *umem)
637 {
638 	int i;
639 
640 	vhost_dev_cleanup(dev);
641 
642 	dev->umem = umem;
643 	/* We don't need VQ locks below since vhost_dev_cleanup makes sure
644 	 * VQs aren't running.
645 	 */
646 	for (i = 0; i < dev->nvqs; ++i)
647 		dev->vqs[i]->umem = umem;
648 }
649 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner);
650 
vhost_dev_stop(struct vhost_dev * dev)651 void vhost_dev_stop(struct vhost_dev *dev)
652 {
653 	int i;
654 
655 	for (i = 0; i < dev->nvqs; ++i) {
656 		if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick)
657 			vhost_poll_stop(&dev->vqs[i]->poll);
658 	}
659 
660 	vhost_dev_flush(dev);
661 }
662 EXPORT_SYMBOL_GPL(vhost_dev_stop);
663 
vhost_clear_msg(struct vhost_dev * dev)664 static void vhost_clear_msg(struct vhost_dev *dev)
665 {
666 	struct vhost_msg_node *node, *n;
667 
668 	spin_lock(&dev->iotlb_lock);
669 
670 	list_for_each_entry_safe(node, n, &dev->read_list, node) {
671 		list_del(&node->node);
672 		kfree(node);
673 	}
674 
675 	list_for_each_entry_safe(node, n, &dev->pending_list, node) {
676 		list_del(&node->node);
677 		kfree(node);
678 	}
679 
680 	spin_unlock(&dev->iotlb_lock);
681 }
682 
vhost_dev_cleanup(struct vhost_dev * dev)683 void vhost_dev_cleanup(struct vhost_dev *dev)
684 {
685 	int i;
686 
687 	for (i = 0; i < dev->nvqs; ++i) {
688 		if (dev->vqs[i]->error_ctx)
689 			eventfd_ctx_put(dev->vqs[i]->error_ctx);
690 		if (dev->vqs[i]->kick)
691 			fput(dev->vqs[i]->kick);
692 		if (dev->vqs[i]->call_ctx.ctx)
693 			eventfd_ctx_put(dev->vqs[i]->call_ctx.ctx);
694 		vhost_vq_reset(dev, dev->vqs[i]);
695 	}
696 	vhost_dev_free_iovecs(dev);
697 	if (dev->log_ctx)
698 		eventfd_ctx_put(dev->log_ctx);
699 	dev->log_ctx = NULL;
700 	/* No one will access memory at this point */
701 	vhost_iotlb_free(dev->umem);
702 	dev->umem = NULL;
703 	vhost_iotlb_free(dev->iotlb);
704 	dev->iotlb = NULL;
705 	vhost_clear_msg(dev);
706 	wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
707 	WARN_ON(!llist_empty(&dev->work_list));
708 	if (dev->worker) {
709 		kthread_stop(dev->worker);
710 		dev->worker = NULL;
711 		dev->kcov_handle = 0;
712 	}
713 	vhost_detach_mm(dev);
714 }
715 EXPORT_SYMBOL_GPL(vhost_dev_cleanup);
716 
log_access_ok(void __user * log_base,u64 addr,unsigned long sz)717 static bool log_access_ok(void __user *log_base, u64 addr, unsigned long sz)
718 {
719 	u64 a = addr / VHOST_PAGE_SIZE / 8;
720 
721 	/* Make sure 64 bit math will not overflow. */
722 	if (a > ULONG_MAX - (unsigned long)log_base ||
723 	    a + (unsigned long)log_base > ULONG_MAX)
724 		return false;
725 
726 	return access_ok(log_base + a,
727 			 (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8);
728 }
729 
730 /* Make sure 64 bit math will not overflow. */
vhost_overflow(u64 uaddr,u64 size)731 static bool vhost_overflow(u64 uaddr, u64 size)
732 {
733 	if (uaddr > ULONG_MAX || size > ULONG_MAX)
734 		return true;
735 
736 	if (!size)
737 		return false;
738 
739 	return uaddr > ULONG_MAX - size + 1;
740 }
741 
742 /* Caller should have vq mutex and device mutex. */
vq_memory_access_ok(void __user * log_base,struct vhost_iotlb * umem,int log_all)743 static bool vq_memory_access_ok(void __user *log_base, struct vhost_iotlb *umem,
744 				int log_all)
745 {
746 	struct vhost_iotlb_map *map;
747 
748 	if (!umem)
749 		return false;
750 
751 	list_for_each_entry(map, &umem->list, link) {
752 		unsigned long a = map->addr;
753 
754 		if (vhost_overflow(map->addr, map->size))
755 			return false;
756 
757 
758 		if (!access_ok((void __user *)a, map->size))
759 			return false;
760 		else if (log_all && !log_access_ok(log_base,
761 						   map->start,
762 						   map->size))
763 			return false;
764 	}
765 	return true;
766 }
767 
vhost_vq_meta_fetch(struct vhost_virtqueue * vq,u64 addr,unsigned int size,int type)768 static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq,
769 					       u64 addr, unsigned int size,
770 					       int type)
771 {
772 	const struct vhost_iotlb_map *map = vq->meta_iotlb[type];
773 
774 	if (!map)
775 		return NULL;
776 
777 	return (void __user *)(uintptr_t)(map->addr + addr - map->start);
778 }
779 
780 /* Can we switch to this memory table? */
781 /* Caller should have device mutex but not vq mutex */
memory_access_ok(struct vhost_dev * d,struct vhost_iotlb * umem,int log_all)782 static bool memory_access_ok(struct vhost_dev *d, struct vhost_iotlb *umem,
783 			     int log_all)
784 {
785 	int i;
786 
787 	for (i = 0; i < d->nvqs; ++i) {
788 		bool ok;
789 		bool log;
790 
791 		mutex_lock(&d->vqs[i]->mutex);
792 		log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL);
793 		/* If ring is inactive, will check when it's enabled. */
794 		if (d->vqs[i]->private_data)
795 			ok = vq_memory_access_ok(d->vqs[i]->log_base,
796 						 umem, log);
797 		else
798 			ok = true;
799 		mutex_unlock(&d->vqs[i]->mutex);
800 		if (!ok)
801 			return false;
802 	}
803 	return true;
804 }
805 
806 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
807 			  struct iovec iov[], int iov_size, int access);
808 
vhost_copy_to_user(struct vhost_virtqueue * vq,void __user * to,const void * from,unsigned size)809 static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to,
810 			      const void *from, unsigned size)
811 {
812 	int ret;
813 
814 	if (!vq->iotlb)
815 		return __copy_to_user(to, from, size);
816 	else {
817 		/* This function should be called after iotlb
818 		 * prefetch, which means we're sure that all vq
819 		 * could be access through iotlb. So -EAGAIN should
820 		 * not happen in this case.
821 		 */
822 		struct iov_iter t;
823 		void __user *uaddr = vhost_vq_meta_fetch(vq,
824 				     (u64)(uintptr_t)to, size,
825 				     VHOST_ADDR_USED);
826 
827 		if (uaddr)
828 			return __copy_to_user(uaddr, from, size);
829 
830 		ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov,
831 				     ARRAY_SIZE(vq->iotlb_iov),
832 				     VHOST_ACCESS_WO);
833 		if (ret < 0)
834 			goto out;
835 		iov_iter_init(&t, WRITE, vq->iotlb_iov, ret, size);
836 		ret = copy_to_iter(from, size, &t);
837 		if (ret == size)
838 			ret = 0;
839 	}
840 out:
841 	return ret;
842 }
843 
vhost_copy_from_user(struct vhost_virtqueue * vq,void * to,void __user * from,unsigned size)844 static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to,
845 				void __user *from, unsigned size)
846 {
847 	int ret;
848 
849 	if (!vq->iotlb)
850 		return __copy_from_user(to, from, size);
851 	else {
852 		/* This function should be called after iotlb
853 		 * prefetch, which means we're sure that vq
854 		 * could be access through iotlb. So -EAGAIN should
855 		 * not happen in this case.
856 		 */
857 		void __user *uaddr = vhost_vq_meta_fetch(vq,
858 				     (u64)(uintptr_t)from, size,
859 				     VHOST_ADDR_DESC);
860 		struct iov_iter f;
861 
862 		if (uaddr)
863 			return __copy_from_user(to, uaddr, size);
864 
865 		ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov,
866 				     ARRAY_SIZE(vq->iotlb_iov),
867 				     VHOST_ACCESS_RO);
868 		if (ret < 0) {
869 			vq_err(vq, "IOTLB translation failure: uaddr "
870 			       "%p size 0x%llx\n", from,
871 			       (unsigned long long) size);
872 			goto out;
873 		}
874 		iov_iter_init(&f, READ, vq->iotlb_iov, ret, size);
875 		ret = copy_from_iter(to, size, &f);
876 		if (ret == size)
877 			ret = 0;
878 	}
879 
880 out:
881 	return ret;
882 }
883 
__vhost_get_user_slow(struct vhost_virtqueue * vq,void __user * addr,unsigned int size,int type)884 static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq,
885 					  void __user *addr, unsigned int size,
886 					  int type)
887 {
888 	int ret;
889 
890 	ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov,
891 			     ARRAY_SIZE(vq->iotlb_iov),
892 			     VHOST_ACCESS_RO);
893 	if (ret < 0) {
894 		vq_err(vq, "IOTLB translation failure: uaddr "
895 			"%p size 0x%llx\n", addr,
896 			(unsigned long long) size);
897 		return NULL;
898 	}
899 
900 	if (ret != 1 || vq->iotlb_iov[0].iov_len != size) {
901 		vq_err(vq, "Non atomic userspace memory access: uaddr "
902 			"%p size 0x%llx\n", addr,
903 			(unsigned long long) size);
904 		return NULL;
905 	}
906 
907 	return vq->iotlb_iov[0].iov_base;
908 }
909 
910 /* This function should be called after iotlb
911  * prefetch, which means we're sure that vq
912  * could be access through iotlb. So -EAGAIN should
913  * not happen in this case.
914  */
__vhost_get_user(struct vhost_virtqueue * vq,void __user * addr,unsigned int size,int type)915 static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq,
916 					    void __user *addr, unsigned int size,
917 					    int type)
918 {
919 	void __user *uaddr = vhost_vq_meta_fetch(vq,
920 			     (u64)(uintptr_t)addr, size, type);
921 	if (uaddr)
922 		return uaddr;
923 
924 	return __vhost_get_user_slow(vq, addr, size, type);
925 }
926 
927 #define vhost_put_user(vq, x, ptr)		\
928 ({ \
929 	int ret; \
930 	if (!vq->iotlb) { \
931 		ret = __put_user(x, ptr); \
932 	} else { \
933 		__typeof__(ptr) to = \
934 			(__typeof__(ptr)) __vhost_get_user(vq, ptr,	\
935 					  sizeof(*ptr), VHOST_ADDR_USED); \
936 		if (to != NULL) \
937 			ret = __put_user(x, to); \
938 		else \
939 			ret = -EFAULT;	\
940 	} \
941 	ret; \
942 })
943 
vhost_put_avail_event(struct vhost_virtqueue * vq)944 static inline int vhost_put_avail_event(struct vhost_virtqueue *vq)
945 {
946 	return vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx),
947 			      vhost_avail_event(vq));
948 }
949 
vhost_put_used(struct vhost_virtqueue * vq,struct vring_used_elem * head,int idx,int count)950 static inline int vhost_put_used(struct vhost_virtqueue *vq,
951 				 struct vring_used_elem *head, int idx,
952 				 int count)
953 {
954 	return vhost_copy_to_user(vq, vq->used->ring + idx, head,
955 				  count * sizeof(*head));
956 }
957 
vhost_put_used_flags(struct vhost_virtqueue * vq)958 static inline int vhost_put_used_flags(struct vhost_virtqueue *vq)
959 
960 {
961 	return vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags),
962 			      &vq->used->flags);
963 }
964 
vhost_put_used_idx(struct vhost_virtqueue * vq)965 static inline int vhost_put_used_idx(struct vhost_virtqueue *vq)
966 
967 {
968 	return vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx),
969 			      &vq->used->idx);
970 }
971 
972 #define vhost_get_user(vq, x, ptr, type)		\
973 ({ \
974 	int ret; \
975 	if (!vq->iotlb) { \
976 		ret = __get_user(x, ptr); \
977 	} else { \
978 		__typeof__(ptr) from = \
979 			(__typeof__(ptr)) __vhost_get_user(vq, ptr, \
980 							   sizeof(*ptr), \
981 							   type); \
982 		if (from != NULL) \
983 			ret = __get_user(x, from); \
984 		else \
985 			ret = -EFAULT; \
986 	} \
987 	ret; \
988 })
989 
990 #define vhost_get_avail(vq, x, ptr) \
991 	vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL)
992 
993 #define vhost_get_used(vq, x, ptr) \
994 	vhost_get_user(vq, x, ptr, VHOST_ADDR_USED)
995 
vhost_dev_lock_vqs(struct vhost_dev * d)996 static void vhost_dev_lock_vqs(struct vhost_dev *d)
997 {
998 	int i = 0;
999 	for (i = 0; i < d->nvqs; ++i)
1000 		mutex_lock_nested(&d->vqs[i]->mutex, i);
1001 }
1002 
vhost_dev_unlock_vqs(struct vhost_dev * d)1003 static void vhost_dev_unlock_vqs(struct vhost_dev *d)
1004 {
1005 	int i = 0;
1006 	for (i = 0; i < d->nvqs; ++i)
1007 		mutex_unlock(&d->vqs[i]->mutex);
1008 }
1009 
vhost_get_avail_idx(struct vhost_virtqueue * vq,__virtio16 * idx)1010 static inline int vhost_get_avail_idx(struct vhost_virtqueue *vq,
1011 				      __virtio16 *idx)
1012 {
1013 	return vhost_get_avail(vq, *idx, &vq->avail->idx);
1014 }
1015 
vhost_get_avail_head(struct vhost_virtqueue * vq,__virtio16 * head,int idx)1016 static inline int vhost_get_avail_head(struct vhost_virtqueue *vq,
1017 				       __virtio16 *head, int idx)
1018 {
1019 	return vhost_get_avail(vq, *head,
1020 			       &vq->avail->ring[idx & (vq->num - 1)]);
1021 }
1022 
vhost_get_avail_flags(struct vhost_virtqueue * vq,__virtio16 * flags)1023 static inline int vhost_get_avail_flags(struct vhost_virtqueue *vq,
1024 					__virtio16 *flags)
1025 {
1026 	return vhost_get_avail(vq, *flags, &vq->avail->flags);
1027 }
1028 
vhost_get_used_event(struct vhost_virtqueue * vq,__virtio16 * event)1029 static inline int vhost_get_used_event(struct vhost_virtqueue *vq,
1030 				       __virtio16 *event)
1031 {
1032 	return vhost_get_avail(vq, *event, vhost_used_event(vq));
1033 }
1034 
vhost_get_used_idx(struct vhost_virtqueue * vq,__virtio16 * idx)1035 static inline int vhost_get_used_idx(struct vhost_virtqueue *vq,
1036 				     __virtio16 *idx)
1037 {
1038 	return vhost_get_used(vq, *idx, &vq->used->idx);
1039 }
1040 
vhost_get_desc(struct vhost_virtqueue * vq,struct vring_desc * desc,int idx)1041 static inline int vhost_get_desc(struct vhost_virtqueue *vq,
1042 				 struct vring_desc *desc, int idx)
1043 {
1044 	return vhost_copy_from_user(vq, desc, vq->desc + idx, sizeof(*desc));
1045 }
1046 
vhost_iotlb_notify_vq(struct vhost_dev * d,struct vhost_iotlb_msg * msg)1047 static void vhost_iotlb_notify_vq(struct vhost_dev *d,
1048 				  struct vhost_iotlb_msg *msg)
1049 {
1050 	struct vhost_msg_node *node, *n;
1051 
1052 	spin_lock(&d->iotlb_lock);
1053 
1054 	list_for_each_entry_safe(node, n, &d->pending_list, node) {
1055 		struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb;
1056 		if (msg->iova <= vq_msg->iova &&
1057 		    msg->iova + msg->size - 1 >= vq_msg->iova &&
1058 		    vq_msg->type == VHOST_IOTLB_MISS) {
1059 			vhost_poll_queue(&node->vq->poll);
1060 			list_del(&node->node);
1061 			kfree(node);
1062 		}
1063 	}
1064 
1065 	spin_unlock(&d->iotlb_lock);
1066 }
1067 
umem_access_ok(u64 uaddr,u64 size,int access)1068 static bool umem_access_ok(u64 uaddr, u64 size, int access)
1069 {
1070 	unsigned long a = uaddr;
1071 
1072 	/* Make sure 64 bit math will not overflow. */
1073 	if (vhost_overflow(uaddr, size))
1074 		return false;
1075 
1076 	if ((access & VHOST_ACCESS_RO) &&
1077 	    !access_ok((void __user *)a, size))
1078 		return false;
1079 	if ((access & VHOST_ACCESS_WO) &&
1080 	    !access_ok((void __user *)a, size))
1081 		return false;
1082 	return true;
1083 }
1084 
vhost_process_iotlb_msg(struct vhost_dev * dev,u32 asid,struct vhost_iotlb_msg * msg)1085 static int vhost_process_iotlb_msg(struct vhost_dev *dev, u32 asid,
1086 				   struct vhost_iotlb_msg *msg)
1087 {
1088 	int ret = 0;
1089 
1090 	if (asid != 0)
1091 		return -EINVAL;
1092 
1093 	mutex_lock(&dev->mutex);
1094 	vhost_dev_lock_vqs(dev);
1095 	switch (msg->type) {
1096 	case VHOST_IOTLB_UPDATE:
1097 		if (!dev->iotlb) {
1098 			ret = -EFAULT;
1099 			break;
1100 		}
1101 		if (!umem_access_ok(msg->uaddr, msg->size, msg->perm)) {
1102 			ret = -EFAULT;
1103 			break;
1104 		}
1105 		vhost_vq_meta_reset(dev);
1106 		if (vhost_iotlb_add_range(dev->iotlb, msg->iova,
1107 					  msg->iova + msg->size - 1,
1108 					  msg->uaddr, msg->perm)) {
1109 			ret = -ENOMEM;
1110 			break;
1111 		}
1112 		vhost_iotlb_notify_vq(dev, msg);
1113 		break;
1114 	case VHOST_IOTLB_INVALIDATE:
1115 		if (!dev->iotlb) {
1116 			ret = -EFAULT;
1117 			break;
1118 		}
1119 		vhost_vq_meta_reset(dev);
1120 		vhost_iotlb_del_range(dev->iotlb, msg->iova,
1121 				      msg->iova + msg->size - 1);
1122 		break;
1123 	default:
1124 		ret = -EINVAL;
1125 		break;
1126 	}
1127 
1128 	vhost_dev_unlock_vqs(dev);
1129 	mutex_unlock(&dev->mutex);
1130 
1131 	return ret;
1132 }
vhost_chr_write_iter(struct vhost_dev * dev,struct iov_iter * from)1133 ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
1134 			     struct iov_iter *from)
1135 {
1136 	struct vhost_iotlb_msg msg;
1137 	size_t offset;
1138 	int type, ret;
1139 	u32 asid = 0;
1140 
1141 	ret = copy_from_iter(&type, sizeof(type), from);
1142 	if (ret != sizeof(type)) {
1143 		ret = -EINVAL;
1144 		goto done;
1145 	}
1146 
1147 	switch (type) {
1148 	case VHOST_IOTLB_MSG:
1149 		/* There maybe a hole after type for V1 message type,
1150 		 * so skip it here.
1151 		 */
1152 		offset = offsetof(struct vhost_msg, iotlb) - sizeof(int);
1153 		break;
1154 	case VHOST_IOTLB_MSG_V2:
1155 		if (vhost_backend_has_feature(dev->vqs[0],
1156 					      VHOST_BACKEND_F_IOTLB_ASID)) {
1157 			ret = copy_from_iter(&asid, sizeof(asid), from);
1158 			if (ret != sizeof(asid)) {
1159 				ret = -EINVAL;
1160 				goto done;
1161 			}
1162 			offset = 0;
1163 		} else
1164 			offset = sizeof(__u32);
1165 		break;
1166 	default:
1167 		ret = -EINVAL;
1168 		goto done;
1169 	}
1170 
1171 	iov_iter_advance(from, offset);
1172 	ret = copy_from_iter(&msg, sizeof(msg), from);
1173 	if (ret != sizeof(msg)) {
1174 		ret = -EINVAL;
1175 		goto done;
1176 	}
1177 
1178 	if ((msg.type == VHOST_IOTLB_UPDATE ||
1179 	     msg.type == VHOST_IOTLB_INVALIDATE) &&
1180 	     msg.size == 0) {
1181 		ret = -EINVAL;
1182 		goto done;
1183 	}
1184 
1185 	if (dev->msg_handler)
1186 		ret = dev->msg_handler(dev, asid, &msg);
1187 	else
1188 		ret = vhost_process_iotlb_msg(dev, asid, &msg);
1189 	if (ret) {
1190 		ret = -EFAULT;
1191 		goto done;
1192 	}
1193 
1194 	ret = (type == VHOST_IOTLB_MSG) ? sizeof(struct vhost_msg) :
1195 	      sizeof(struct vhost_msg_v2);
1196 done:
1197 	return ret;
1198 }
1199 EXPORT_SYMBOL(vhost_chr_write_iter);
1200 
vhost_chr_poll(struct file * file,struct vhost_dev * dev,poll_table * wait)1201 __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
1202 			    poll_table *wait)
1203 {
1204 	__poll_t mask = 0;
1205 
1206 	poll_wait(file, &dev->wait, wait);
1207 
1208 	if (!list_empty(&dev->read_list))
1209 		mask |= EPOLLIN | EPOLLRDNORM;
1210 
1211 	return mask;
1212 }
1213 EXPORT_SYMBOL(vhost_chr_poll);
1214 
vhost_chr_read_iter(struct vhost_dev * dev,struct iov_iter * to,int noblock)1215 ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
1216 			    int noblock)
1217 {
1218 	DEFINE_WAIT(wait);
1219 	struct vhost_msg_node *node;
1220 	ssize_t ret = 0;
1221 	unsigned size = sizeof(struct vhost_msg);
1222 
1223 	if (iov_iter_count(to) < size)
1224 		return 0;
1225 
1226 	while (1) {
1227 		if (!noblock)
1228 			prepare_to_wait(&dev->wait, &wait,
1229 					TASK_INTERRUPTIBLE);
1230 
1231 		node = vhost_dequeue_msg(dev, &dev->read_list);
1232 		if (node)
1233 			break;
1234 		if (noblock) {
1235 			ret = -EAGAIN;
1236 			break;
1237 		}
1238 		if (signal_pending(current)) {
1239 			ret = -ERESTARTSYS;
1240 			break;
1241 		}
1242 		if (!dev->iotlb) {
1243 			ret = -EBADFD;
1244 			break;
1245 		}
1246 
1247 		schedule();
1248 	}
1249 
1250 	if (!noblock)
1251 		finish_wait(&dev->wait, &wait);
1252 
1253 	if (node) {
1254 		struct vhost_iotlb_msg *msg;
1255 		void *start = &node->msg;
1256 
1257 		switch (node->msg.type) {
1258 		case VHOST_IOTLB_MSG:
1259 			size = sizeof(node->msg);
1260 			msg = &node->msg.iotlb;
1261 			break;
1262 		case VHOST_IOTLB_MSG_V2:
1263 			size = sizeof(node->msg_v2);
1264 			msg = &node->msg_v2.iotlb;
1265 			break;
1266 		default:
1267 			BUG();
1268 			break;
1269 		}
1270 
1271 		ret = copy_to_iter(start, size, to);
1272 		if (ret != size || msg->type != VHOST_IOTLB_MISS) {
1273 			kfree(node);
1274 			return ret;
1275 		}
1276 		vhost_enqueue_msg(dev, &dev->pending_list, node);
1277 	}
1278 
1279 	return ret;
1280 }
1281 EXPORT_SYMBOL_GPL(vhost_chr_read_iter);
1282 
vhost_iotlb_miss(struct vhost_virtqueue * vq,u64 iova,int access)1283 static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access)
1284 {
1285 	struct vhost_dev *dev = vq->dev;
1286 	struct vhost_msg_node *node;
1287 	struct vhost_iotlb_msg *msg;
1288 	bool v2 = vhost_backend_has_feature(vq, VHOST_BACKEND_F_IOTLB_MSG_V2);
1289 
1290 	node = vhost_new_msg(vq, v2 ? VHOST_IOTLB_MSG_V2 : VHOST_IOTLB_MSG);
1291 	if (!node)
1292 		return -ENOMEM;
1293 
1294 	if (v2) {
1295 		node->msg_v2.type = VHOST_IOTLB_MSG_V2;
1296 		msg = &node->msg_v2.iotlb;
1297 	} else {
1298 		msg = &node->msg.iotlb;
1299 	}
1300 
1301 	msg->type = VHOST_IOTLB_MISS;
1302 	msg->iova = iova;
1303 	msg->perm = access;
1304 
1305 	vhost_enqueue_msg(dev, &dev->read_list, node);
1306 
1307 	return 0;
1308 }
1309 
vq_access_ok(struct vhost_virtqueue * vq,unsigned int num,vring_desc_t __user * desc,vring_avail_t __user * avail,vring_used_t __user * used)1310 static bool vq_access_ok(struct vhost_virtqueue *vq, unsigned int num,
1311 			 vring_desc_t __user *desc,
1312 			 vring_avail_t __user *avail,
1313 			 vring_used_t __user *used)
1314 
1315 {
1316 	/* If an IOTLB device is present, the vring addresses are
1317 	 * GIOVAs. Access validation occurs at prefetch time. */
1318 	if (vq->iotlb)
1319 		return true;
1320 
1321 	return access_ok(desc, vhost_get_desc_size(vq, num)) &&
1322 	       access_ok(avail, vhost_get_avail_size(vq, num)) &&
1323 	       access_ok(used, vhost_get_used_size(vq, num));
1324 }
1325 
vhost_vq_meta_update(struct vhost_virtqueue * vq,const struct vhost_iotlb_map * map,int type)1326 static void vhost_vq_meta_update(struct vhost_virtqueue *vq,
1327 				 const struct vhost_iotlb_map *map,
1328 				 int type)
1329 {
1330 	int access = (type == VHOST_ADDR_USED) ?
1331 		     VHOST_ACCESS_WO : VHOST_ACCESS_RO;
1332 
1333 	if (likely(map->perm & access))
1334 		vq->meta_iotlb[type] = map;
1335 }
1336 
iotlb_access_ok(struct vhost_virtqueue * vq,int access,u64 addr,u64 len,int type)1337 static bool iotlb_access_ok(struct vhost_virtqueue *vq,
1338 			    int access, u64 addr, u64 len, int type)
1339 {
1340 	const struct vhost_iotlb_map *map;
1341 	struct vhost_iotlb *umem = vq->iotlb;
1342 	u64 s = 0, size, orig_addr = addr, last = addr + len - 1;
1343 
1344 	if (vhost_vq_meta_fetch(vq, addr, len, type))
1345 		return true;
1346 
1347 	while (len > s) {
1348 		map = vhost_iotlb_itree_first(umem, addr, last);
1349 		if (map == NULL || map->start > addr) {
1350 			vhost_iotlb_miss(vq, addr, access);
1351 			return false;
1352 		} else if (!(map->perm & access)) {
1353 			/* Report the possible access violation by
1354 			 * request another translation from userspace.
1355 			 */
1356 			return false;
1357 		}
1358 
1359 		size = map->size - addr + map->start;
1360 
1361 		if (orig_addr == addr && size >= len)
1362 			vhost_vq_meta_update(vq, map, type);
1363 
1364 		s += size;
1365 		addr += size;
1366 	}
1367 
1368 	return true;
1369 }
1370 
vq_meta_prefetch(struct vhost_virtqueue * vq)1371 int vq_meta_prefetch(struct vhost_virtqueue *vq)
1372 {
1373 	unsigned int num = vq->num;
1374 
1375 	if (!vq->iotlb)
1376 		return 1;
1377 
1378 	return iotlb_access_ok(vq, VHOST_MAP_RO, (u64)(uintptr_t)vq->desc,
1379 			       vhost_get_desc_size(vq, num), VHOST_ADDR_DESC) &&
1380 	       iotlb_access_ok(vq, VHOST_MAP_RO, (u64)(uintptr_t)vq->avail,
1381 			       vhost_get_avail_size(vq, num),
1382 			       VHOST_ADDR_AVAIL) &&
1383 	       iotlb_access_ok(vq, VHOST_MAP_WO, (u64)(uintptr_t)vq->used,
1384 			       vhost_get_used_size(vq, num), VHOST_ADDR_USED);
1385 }
1386 EXPORT_SYMBOL_GPL(vq_meta_prefetch);
1387 
1388 /* Can we log writes? */
1389 /* Caller should have device mutex but not vq mutex */
vhost_log_access_ok(struct vhost_dev * dev)1390 bool vhost_log_access_ok(struct vhost_dev *dev)
1391 {
1392 	return memory_access_ok(dev, dev->umem, 1);
1393 }
1394 EXPORT_SYMBOL_GPL(vhost_log_access_ok);
1395 
vq_log_used_access_ok(struct vhost_virtqueue * vq,void __user * log_base,bool log_used,u64 log_addr)1396 static bool vq_log_used_access_ok(struct vhost_virtqueue *vq,
1397 				  void __user *log_base,
1398 				  bool log_used,
1399 				  u64 log_addr)
1400 {
1401 	/* If an IOTLB device is present, log_addr is a GIOVA that
1402 	 * will never be logged by log_used(). */
1403 	if (vq->iotlb)
1404 		return true;
1405 
1406 	return !log_used || log_access_ok(log_base, log_addr,
1407 					  vhost_get_used_size(vq, vq->num));
1408 }
1409 
1410 /* Verify access for write logging. */
1411 /* Caller should have vq mutex and device mutex */
vq_log_access_ok(struct vhost_virtqueue * vq,void __user * log_base)1412 static bool vq_log_access_ok(struct vhost_virtqueue *vq,
1413 			     void __user *log_base)
1414 {
1415 	return vq_memory_access_ok(log_base, vq->umem,
1416 				   vhost_has_feature(vq, VHOST_F_LOG_ALL)) &&
1417 		vq_log_used_access_ok(vq, log_base, vq->log_used, vq->log_addr);
1418 }
1419 
1420 /* Can we start vq? */
1421 /* Caller should have vq mutex and device mutex */
vhost_vq_access_ok(struct vhost_virtqueue * vq)1422 bool vhost_vq_access_ok(struct vhost_virtqueue *vq)
1423 {
1424 	if (!vq_log_access_ok(vq, vq->log_base))
1425 		return false;
1426 
1427 	return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used);
1428 }
1429 EXPORT_SYMBOL_GPL(vhost_vq_access_ok);
1430 
vhost_set_memory(struct vhost_dev * d,struct vhost_memory __user * m)1431 static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)
1432 {
1433 	struct vhost_memory mem, *newmem;
1434 	struct vhost_memory_region *region;
1435 	struct vhost_iotlb *newumem, *oldumem;
1436 	unsigned long size = offsetof(struct vhost_memory, regions);
1437 	int i;
1438 
1439 	if (copy_from_user(&mem, m, size))
1440 		return -EFAULT;
1441 	if (mem.padding)
1442 		return -EOPNOTSUPP;
1443 	if (mem.nregions > max_mem_regions)
1444 		return -E2BIG;
1445 	newmem = kvzalloc(struct_size(newmem, regions, mem.nregions),
1446 			GFP_KERNEL);
1447 	if (!newmem)
1448 		return -ENOMEM;
1449 
1450 	memcpy(newmem, &mem, size);
1451 	if (copy_from_user(newmem->regions, m->regions,
1452 			   flex_array_size(newmem, regions, mem.nregions))) {
1453 		kvfree(newmem);
1454 		return -EFAULT;
1455 	}
1456 
1457 	newumem = iotlb_alloc();
1458 	if (!newumem) {
1459 		kvfree(newmem);
1460 		return -ENOMEM;
1461 	}
1462 
1463 	for (region = newmem->regions;
1464 	     region < newmem->regions + mem.nregions;
1465 	     region++) {
1466 		if (vhost_iotlb_add_range(newumem,
1467 					  region->guest_phys_addr,
1468 					  region->guest_phys_addr +
1469 					  region->memory_size - 1,
1470 					  region->userspace_addr,
1471 					  VHOST_MAP_RW))
1472 			goto err;
1473 	}
1474 
1475 	if (!memory_access_ok(d, newumem, 0))
1476 		goto err;
1477 
1478 	oldumem = d->umem;
1479 	d->umem = newumem;
1480 
1481 	/* All memory accesses are done under some VQ mutex. */
1482 	for (i = 0; i < d->nvqs; ++i) {
1483 		mutex_lock(&d->vqs[i]->mutex);
1484 		d->vqs[i]->umem = newumem;
1485 		mutex_unlock(&d->vqs[i]->mutex);
1486 	}
1487 
1488 	kvfree(newmem);
1489 	vhost_iotlb_free(oldumem);
1490 	return 0;
1491 
1492 err:
1493 	vhost_iotlb_free(newumem);
1494 	kvfree(newmem);
1495 	return -EFAULT;
1496 }
1497 
vhost_vring_set_num(struct vhost_dev * d,struct vhost_virtqueue * vq,void __user * argp)1498 static long vhost_vring_set_num(struct vhost_dev *d,
1499 				struct vhost_virtqueue *vq,
1500 				void __user *argp)
1501 {
1502 	struct vhost_vring_state s;
1503 
1504 	/* Resizing ring with an active backend?
1505 	 * You don't want to do that. */
1506 	if (vq->private_data)
1507 		return -EBUSY;
1508 
1509 	if (copy_from_user(&s, argp, sizeof s))
1510 		return -EFAULT;
1511 
1512 	if (!s.num || s.num > 0xffff || (s.num & (s.num - 1)))
1513 		return -EINVAL;
1514 	vq->num = s.num;
1515 
1516 	return 0;
1517 }
1518 
vhost_vring_set_addr(struct vhost_dev * d,struct vhost_virtqueue * vq,void __user * argp)1519 static long vhost_vring_set_addr(struct vhost_dev *d,
1520 				 struct vhost_virtqueue *vq,
1521 				 void __user *argp)
1522 {
1523 	struct vhost_vring_addr a;
1524 
1525 	if (copy_from_user(&a, argp, sizeof a))
1526 		return -EFAULT;
1527 	if (a.flags & ~(0x1 << VHOST_VRING_F_LOG))
1528 		return -EOPNOTSUPP;
1529 
1530 	/* For 32bit, verify that the top 32bits of the user
1531 	   data are set to zero. */
1532 	if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr ||
1533 	    (u64)(unsigned long)a.used_user_addr != a.used_user_addr ||
1534 	    (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr)
1535 		return -EFAULT;
1536 
1537 	/* Make sure it's safe to cast pointers to vring types. */
1538 	BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE);
1539 	BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE);
1540 	if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) ||
1541 	    (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) ||
1542 	    (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1)))
1543 		return -EINVAL;
1544 
1545 	/* We only verify access here if backend is configured.
1546 	 * If it is not, we don't as size might not have been setup.
1547 	 * We will verify when backend is configured. */
1548 	if (vq->private_data) {
1549 		if (!vq_access_ok(vq, vq->num,
1550 			(void __user *)(unsigned long)a.desc_user_addr,
1551 			(void __user *)(unsigned long)a.avail_user_addr,
1552 			(void __user *)(unsigned long)a.used_user_addr))
1553 			return -EINVAL;
1554 
1555 		/* Also validate log access for used ring if enabled. */
1556 		if (!vq_log_used_access_ok(vq, vq->log_base,
1557 				a.flags & (0x1 << VHOST_VRING_F_LOG),
1558 				a.log_guest_addr))
1559 			return -EINVAL;
1560 	}
1561 
1562 	vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG));
1563 	vq->desc = (void __user *)(unsigned long)a.desc_user_addr;
1564 	vq->avail = (void __user *)(unsigned long)a.avail_user_addr;
1565 	vq->log_addr = a.log_guest_addr;
1566 	vq->used = (void __user *)(unsigned long)a.used_user_addr;
1567 
1568 	return 0;
1569 }
1570 
vhost_vring_set_num_addr(struct vhost_dev * d,struct vhost_virtqueue * vq,unsigned int ioctl,void __user * argp)1571 static long vhost_vring_set_num_addr(struct vhost_dev *d,
1572 				     struct vhost_virtqueue *vq,
1573 				     unsigned int ioctl,
1574 				     void __user *argp)
1575 {
1576 	long r;
1577 
1578 	mutex_lock(&vq->mutex);
1579 
1580 	switch (ioctl) {
1581 	case VHOST_SET_VRING_NUM:
1582 		r = vhost_vring_set_num(d, vq, argp);
1583 		break;
1584 	case VHOST_SET_VRING_ADDR:
1585 		r = vhost_vring_set_addr(d, vq, argp);
1586 		break;
1587 	default:
1588 		BUG();
1589 	}
1590 
1591 	mutex_unlock(&vq->mutex);
1592 
1593 	return r;
1594 }
vhost_vring_ioctl(struct vhost_dev * d,unsigned int ioctl,void __user * argp)1595 long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
1596 {
1597 	struct file *eventfp, *filep = NULL;
1598 	bool pollstart = false, pollstop = false;
1599 	struct eventfd_ctx *ctx = NULL;
1600 	u32 __user *idxp = argp;
1601 	struct vhost_virtqueue *vq;
1602 	struct vhost_vring_state s;
1603 	struct vhost_vring_file f;
1604 	u32 idx;
1605 	long r;
1606 
1607 	r = get_user(idx, idxp);
1608 	if (r < 0)
1609 		return r;
1610 	if (idx >= d->nvqs)
1611 		return -ENOBUFS;
1612 
1613 	idx = array_index_nospec(idx, d->nvqs);
1614 	vq = d->vqs[idx];
1615 
1616 	if (ioctl == VHOST_SET_VRING_NUM ||
1617 	    ioctl == VHOST_SET_VRING_ADDR) {
1618 		return vhost_vring_set_num_addr(d, vq, ioctl, argp);
1619 	}
1620 
1621 	mutex_lock(&vq->mutex);
1622 
1623 	switch (ioctl) {
1624 	case VHOST_SET_VRING_BASE:
1625 		/* Moving base with an active backend?
1626 		 * You don't want to do that. */
1627 		if (vq->private_data) {
1628 			r = -EBUSY;
1629 			break;
1630 		}
1631 		if (copy_from_user(&s, argp, sizeof s)) {
1632 			r = -EFAULT;
1633 			break;
1634 		}
1635 		if (s.num > 0xffff) {
1636 			r = -EINVAL;
1637 			break;
1638 		}
1639 		vq->last_avail_idx = s.num;
1640 		/* Forget the cached index value. */
1641 		vq->avail_idx = vq->last_avail_idx;
1642 		break;
1643 	case VHOST_GET_VRING_BASE:
1644 		s.index = idx;
1645 		s.num = vq->last_avail_idx;
1646 		if (copy_to_user(argp, &s, sizeof s))
1647 			r = -EFAULT;
1648 		break;
1649 	case VHOST_SET_VRING_KICK:
1650 		if (copy_from_user(&f, argp, sizeof f)) {
1651 			r = -EFAULT;
1652 			break;
1653 		}
1654 		eventfp = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_fget(f.fd);
1655 		if (IS_ERR(eventfp)) {
1656 			r = PTR_ERR(eventfp);
1657 			break;
1658 		}
1659 		if (eventfp != vq->kick) {
1660 			pollstop = (filep = vq->kick) != NULL;
1661 			pollstart = (vq->kick = eventfp) != NULL;
1662 		} else
1663 			filep = eventfp;
1664 		break;
1665 	case VHOST_SET_VRING_CALL:
1666 		if (copy_from_user(&f, argp, sizeof f)) {
1667 			r = -EFAULT;
1668 			break;
1669 		}
1670 		ctx = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(f.fd);
1671 		if (IS_ERR(ctx)) {
1672 			r = PTR_ERR(ctx);
1673 			break;
1674 		}
1675 
1676 		swap(ctx, vq->call_ctx.ctx);
1677 		break;
1678 	case VHOST_SET_VRING_ERR:
1679 		if (copy_from_user(&f, argp, sizeof f)) {
1680 			r = -EFAULT;
1681 			break;
1682 		}
1683 		ctx = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(f.fd);
1684 		if (IS_ERR(ctx)) {
1685 			r = PTR_ERR(ctx);
1686 			break;
1687 		}
1688 		swap(ctx, vq->error_ctx);
1689 		break;
1690 	case VHOST_SET_VRING_ENDIAN:
1691 		r = vhost_set_vring_endian(vq, argp);
1692 		break;
1693 	case VHOST_GET_VRING_ENDIAN:
1694 		r = vhost_get_vring_endian(vq, idx, argp);
1695 		break;
1696 	case VHOST_SET_VRING_BUSYLOOP_TIMEOUT:
1697 		if (copy_from_user(&s, argp, sizeof(s))) {
1698 			r = -EFAULT;
1699 			break;
1700 		}
1701 		vq->busyloop_timeout = s.num;
1702 		break;
1703 	case VHOST_GET_VRING_BUSYLOOP_TIMEOUT:
1704 		s.index = idx;
1705 		s.num = vq->busyloop_timeout;
1706 		if (copy_to_user(argp, &s, sizeof(s)))
1707 			r = -EFAULT;
1708 		break;
1709 	default:
1710 		r = -ENOIOCTLCMD;
1711 	}
1712 
1713 	if (pollstop && vq->handle_kick)
1714 		vhost_poll_stop(&vq->poll);
1715 
1716 	if (!IS_ERR_OR_NULL(ctx))
1717 		eventfd_ctx_put(ctx);
1718 	if (filep)
1719 		fput(filep);
1720 
1721 	if (pollstart && vq->handle_kick)
1722 		r = vhost_poll_start(&vq->poll, vq->kick);
1723 
1724 	mutex_unlock(&vq->mutex);
1725 
1726 	if (pollstop && vq->handle_kick)
1727 		vhost_dev_flush(vq->poll.dev);
1728 	return r;
1729 }
1730 EXPORT_SYMBOL_GPL(vhost_vring_ioctl);
1731 
vhost_init_device_iotlb(struct vhost_dev * d,bool enabled)1732 int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled)
1733 {
1734 	struct vhost_iotlb *niotlb, *oiotlb;
1735 	int i;
1736 
1737 	niotlb = iotlb_alloc();
1738 	if (!niotlb)
1739 		return -ENOMEM;
1740 
1741 	oiotlb = d->iotlb;
1742 	d->iotlb = niotlb;
1743 
1744 	for (i = 0; i < d->nvqs; ++i) {
1745 		struct vhost_virtqueue *vq = d->vqs[i];
1746 
1747 		mutex_lock(&vq->mutex);
1748 		vq->iotlb = niotlb;
1749 		__vhost_vq_meta_reset(vq);
1750 		mutex_unlock(&vq->mutex);
1751 	}
1752 
1753 	vhost_iotlb_free(oiotlb);
1754 
1755 	return 0;
1756 }
1757 EXPORT_SYMBOL_GPL(vhost_init_device_iotlb);
1758 
1759 /* Caller must have device mutex */
vhost_dev_ioctl(struct vhost_dev * d,unsigned int ioctl,void __user * argp)1760 long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
1761 {
1762 	struct eventfd_ctx *ctx;
1763 	u64 p;
1764 	long r;
1765 	int i, fd;
1766 
1767 	/* If you are not the owner, you can become one */
1768 	if (ioctl == VHOST_SET_OWNER) {
1769 		r = vhost_dev_set_owner(d);
1770 		goto done;
1771 	}
1772 
1773 	/* You must be the owner to do anything else */
1774 	r = vhost_dev_check_owner(d);
1775 	if (r)
1776 		goto done;
1777 
1778 	switch (ioctl) {
1779 	case VHOST_SET_MEM_TABLE:
1780 		r = vhost_set_memory(d, argp);
1781 		break;
1782 	case VHOST_SET_LOG_BASE:
1783 		if (copy_from_user(&p, argp, sizeof p)) {
1784 			r = -EFAULT;
1785 			break;
1786 		}
1787 		if ((u64)(unsigned long)p != p) {
1788 			r = -EFAULT;
1789 			break;
1790 		}
1791 		for (i = 0; i < d->nvqs; ++i) {
1792 			struct vhost_virtqueue *vq;
1793 			void __user *base = (void __user *)(unsigned long)p;
1794 			vq = d->vqs[i];
1795 			mutex_lock(&vq->mutex);
1796 			/* If ring is inactive, will check when it's enabled. */
1797 			if (vq->private_data && !vq_log_access_ok(vq, base))
1798 				r = -EFAULT;
1799 			else
1800 				vq->log_base = base;
1801 			mutex_unlock(&vq->mutex);
1802 		}
1803 		break;
1804 	case VHOST_SET_LOG_FD:
1805 		r = get_user(fd, (int __user *)argp);
1806 		if (r < 0)
1807 			break;
1808 		ctx = fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(fd);
1809 		if (IS_ERR(ctx)) {
1810 			r = PTR_ERR(ctx);
1811 			break;
1812 		}
1813 		swap(ctx, d->log_ctx);
1814 		for (i = 0; i < d->nvqs; ++i) {
1815 			mutex_lock(&d->vqs[i]->mutex);
1816 			d->vqs[i]->log_ctx = d->log_ctx;
1817 			mutex_unlock(&d->vqs[i]->mutex);
1818 		}
1819 		if (ctx)
1820 			eventfd_ctx_put(ctx);
1821 		break;
1822 	default:
1823 		r = -ENOIOCTLCMD;
1824 		break;
1825 	}
1826 done:
1827 	return r;
1828 }
1829 EXPORT_SYMBOL_GPL(vhost_dev_ioctl);
1830 
1831 /* TODO: This is really inefficient.  We need something like get_user()
1832  * (instruction directly accesses the data, with an exception table entry
1833  * returning -EFAULT). See Documentation/x86/exception-tables.rst.
1834  */
set_bit_to_user(int nr,void __user * addr)1835 static int set_bit_to_user(int nr, void __user *addr)
1836 {
1837 	unsigned long log = (unsigned long)addr;
1838 	struct page *page;
1839 	void *base;
1840 	int bit = nr + (log % PAGE_SIZE) * 8;
1841 	int r;
1842 
1843 	r = pin_user_pages_fast(log, 1, FOLL_WRITE, &page);
1844 	if (r < 0)
1845 		return r;
1846 	BUG_ON(r != 1);
1847 	base = kmap_atomic(page);
1848 	set_bit(bit, base);
1849 	kunmap_atomic(base);
1850 	unpin_user_pages_dirty_lock(&page, 1, true);
1851 	return 0;
1852 }
1853 
log_write(void __user * log_base,u64 write_address,u64 write_length)1854 static int log_write(void __user *log_base,
1855 		     u64 write_address, u64 write_length)
1856 {
1857 	u64 write_page = write_address / VHOST_PAGE_SIZE;
1858 	int r;
1859 
1860 	if (!write_length)
1861 		return 0;
1862 	write_length += write_address % VHOST_PAGE_SIZE;
1863 	for (;;) {
1864 		u64 base = (u64)(unsigned long)log_base;
1865 		u64 log = base + write_page / 8;
1866 		int bit = write_page % 8;
1867 		if ((u64)(unsigned long)log != log)
1868 			return -EFAULT;
1869 		r = set_bit_to_user(bit, (void __user *)(unsigned long)log);
1870 		if (r < 0)
1871 			return r;
1872 		if (write_length <= VHOST_PAGE_SIZE)
1873 			break;
1874 		write_length -= VHOST_PAGE_SIZE;
1875 		write_page += 1;
1876 	}
1877 	return r;
1878 }
1879 
log_write_hva(struct vhost_virtqueue * vq,u64 hva,u64 len)1880 static int log_write_hva(struct vhost_virtqueue *vq, u64 hva, u64 len)
1881 {
1882 	struct vhost_iotlb *umem = vq->umem;
1883 	struct vhost_iotlb_map *u;
1884 	u64 start, end, l, min;
1885 	int r;
1886 	bool hit = false;
1887 
1888 	while (len) {
1889 		min = len;
1890 		/* More than one GPAs can be mapped into a single HVA. So
1891 		 * iterate all possible umems here to be safe.
1892 		 */
1893 		list_for_each_entry(u, &umem->list, link) {
1894 			if (u->addr > hva - 1 + len ||
1895 			    u->addr - 1 + u->size < hva)
1896 				continue;
1897 			start = max(u->addr, hva);
1898 			end = min(u->addr - 1 + u->size, hva - 1 + len);
1899 			l = end - start + 1;
1900 			r = log_write(vq->log_base,
1901 				      u->start + start - u->addr,
1902 				      l);
1903 			if (r < 0)
1904 				return r;
1905 			hit = true;
1906 			min = min(l, min);
1907 		}
1908 
1909 		if (!hit)
1910 			return -EFAULT;
1911 
1912 		len -= min;
1913 		hva += min;
1914 	}
1915 
1916 	return 0;
1917 }
1918 
log_used(struct vhost_virtqueue * vq,u64 used_offset,u64 len)1919 static int log_used(struct vhost_virtqueue *vq, u64 used_offset, u64 len)
1920 {
1921 	struct iovec *iov = vq->log_iov;
1922 	int i, ret;
1923 
1924 	if (!vq->iotlb)
1925 		return log_write(vq->log_base, vq->log_addr + used_offset, len);
1926 
1927 	ret = translate_desc(vq, (uintptr_t)vq->used + used_offset,
1928 			     len, iov, 64, VHOST_ACCESS_WO);
1929 	if (ret < 0)
1930 		return ret;
1931 
1932 	for (i = 0; i < ret; i++) {
1933 		ret = log_write_hva(vq,	(uintptr_t)iov[i].iov_base,
1934 				    iov[i].iov_len);
1935 		if (ret)
1936 			return ret;
1937 	}
1938 
1939 	return 0;
1940 }
1941 
vhost_log_write(struct vhost_virtqueue * vq,struct vhost_log * log,unsigned int log_num,u64 len,struct iovec * iov,int count)1942 int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
1943 		    unsigned int log_num, u64 len, struct iovec *iov, int count)
1944 {
1945 	int i, r;
1946 
1947 	/* Make sure data written is seen before log. */
1948 	smp_wmb();
1949 
1950 	if (vq->iotlb) {
1951 		for (i = 0; i < count; i++) {
1952 			r = log_write_hva(vq, (uintptr_t)iov[i].iov_base,
1953 					  iov[i].iov_len);
1954 			if (r < 0)
1955 				return r;
1956 		}
1957 		return 0;
1958 	}
1959 
1960 	for (i = 0; i < log_num; ++i) {
1961 		u64 l = min(log[i].len, len);
1962 		r = log_write(vq->log_base, log[i].addr, l);
1963 		if (r < 0)
1964 			return r;
1965 		len -= l;
1966 		if (!len) {
1967 			if (vq->log_ctx)
1968 				eventfd_signal(vq->log_ctx, 1);
1969 			return 0;
1970 		}
1971 	}
1972 	/* Length written exceeds what we have stored. This is a bug. */
1973 	BUG();
1974 	return 0;
1975 }
1976 EXPORT_SYMBOL_GPL(vhost_log_write);
1977 
vhost_update_used_flags(struct vhost_virtqueue * vq)1978 static int vhost_update_used_flags(struct vhost_virtqueue *vq)
1979 {
1980 	void __user *used;
1981 	if (vhost_put_used_flags(vq))
1982 		return -EFAULT;
1983 	if (unlikely(vq->log_used)) {
1984 		/* Make sure the flag is seen before log. */
1985 		smp_wmb();
1986 		/* Log used flag write. */
1987 		used = &vq->used->flags;
1988 		log_used(vq, (used - (void __user *)vq->used),
1989 			 sizeof vq->used->flags);
1990 		if (vq->log_ctx)
1991 			eventfd_signal(vq->log_ctx, 1);
1992 	}
1993 	return 0;
1994 }
1995 
vhost_update_avail_event(struct vhost_virtqueue * vq)1996 static int vhost_update_avail_event(struct vhost_virtqueue *vq)
1997 {
1998 	if (vhost_put_avail_event(vq))
1999 		return -EFAULT;
2000 	if (unlikely(vq->log_used)) {
2001 		void __user *used;
2002 		/* Make sure the event is seen before log. */
2003 		smp_wmb();
2004 		/* Log avail event write */
2005 		used = vhost_avail_event(vq);
2006 		log_used(vq, (used - (void __user *)vq->used),
2007 			 sizeof *vhost_avail_event(vq));
2008 		if (vq->log_ctx)
2009 			eventfd_signal(vq->log_ctx, 1);
2010 	}
2011 	return 0;
2012 }
2013 
vhost_vq_init_access(struct vhost_virtqueue * vq)2014 int vhost_vq_init_access(struct vhost_virtqueue *vq)
2015 {
2016 	__virtio16 last_used_idx;
2017 	int r;
2018 	bool is_le = vq->is_le;
2019 
2020 	if (!vq->private_data)
2021 		return 0;
2022 
2023 	vhost_init_is_le(vq);
2024 
2025 	r = vhost_update_used_flags(vq);
2026 	if (r)
2027 		goto err;
2028 	vq->signalled_used_valid = false;
2029 	if (!vq->iotlb &&
2030 	    !access_ok(&vq->used->idx, sizeof vq->used->idx)) {
2031 		r = -EFAULT;
2032 		goto err;
2033 	}
2034 	r = vhost_get_used_idx(vq, &last_used_idx);
2035 	if (r) {
2036 		vq_err(vq, "Can't access used idx at %p\n",
2037 		       &vq->used->idx);
2038 		goto err;
2039 	}
2040 	vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx);
2041 	return 0;
2042 
2043 err:
2044 	vq->is_le = is_le;
2045 	return r;
2046 }
2047 EXPORT_SYMBOL_GPL(vhost_vq_init_access);
2048 
translate_desc(struct vhost_virtqueue * vq,u64 addr,u32 len,struct iovec iov[],int iov_size,int access)2049 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
2050 			  struct iovec iov[], int iov_size, int access)
2051 {
2052 	const struct vhost_iotlb_map *map;
2053 	struct vhost_dev *dev = vq->dev;
2054 	struct vhost_iotlb *umem = dev->iotlb ? dev->iotlb : dev->umem;
2055 	struct iovec *_iov;
2056 	u64 s = 0, last = addr + len - 1;
2057 	int ret = 0;
2058 
2059 	while ((u64)len > s) {
2060 		u64 size;
2061 		if (unlikely(ret >= iov_size)) {
2062 			ret = -ENOBUFS;
2063 			break;
2064 		}
2065 
2066 		map = vhost_iotlb_itree_first(umem, addr, last);
2067 		if (map == NULL || map->start > addr) {
2068 			if (umem != dev->iotlb) {
2069 				ret = -EFAULT;
2070 				break;
2071 			}
2072 			ret = -EAGAIN;
2073 			break;
2074 		} else if (!(map->perm & access)) {
2075 			ret = -EPERM;
2076 			break;
2077 		}
2078 
2079 		_iov = iov + ret;
2080 		size = map->size - addr + map->start;
2081 		_iov->iov_len = min((u64)len - s, size);
2082 		_iov->iov_base = (void __user *)(unsigned long)
2083 				 (map->addr + addr - map->start);
2084 		s += size;
2085 		addr += size;
2086 		++ret;
2087 	}
2088 
2089 	if (ret == -EAGAIN)
2090 		vhost_iotlb_miss(vq, addr, access);
2091 	return ret;
2092 }
2093 
2094 /* Each buffer in the virtqueues is actually a chain of descriptors.  This
2095  * function returns the next descriptor in the chain,
2096  * or -1U if we're at the end. */
next_desc(struct vhost_virtqueue * vq,struct vring_desc * desc)2097 static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc)
2098 {
2099 	unsigned int next;
2100 
2101 	/* If this descriptor says it doesn't chain, we're done. */
2102 	if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT)))
2103 		return -1U;
2104 
2105 	/* Check they're not leading us off end of descriptors. */
2106 	next = vhost16_to_cpu(vq, READ_ONCE(desc->next));
2107 	return next;
2108 }
2109 
get_indirect(struct vhost_virtqueue * vq,struct iovec iov[],unsigned int iov_size,unsigned int * out_num,unsigned int * in_num,struct vhost_log * log,unsigned int * log_num,struct vring_desc * indirect)2110 static int get_indirect(struct vhost_virtqueue *vq,
2111 			struct iovec iov[], unsigned int iov_size,
2112 			unsigned int *out_num, unsigned int *in_num,
2113 			struct vhost_log *log, unsigned int *log_num,
2114 			struct vring_desc *indirect)
2115 {
2116 	struct vring_desc desc;
2117 	unsigned int i = 0, count, found = 0;
2118 	u32 len = vhost32_to_cpu(vq, indirect->len);
2119 	struct iov_iter from;
2120 	int ret, access;
2121 
2122 	/* Sanity check */
2123 	if (unlikely(len % sizeof desc)) {
2124 		vq_err(vq, "Invalid length in indirect descriptor: "
2125 		       "len 0x%llx not multiple of 0x%zx\n",
2126 		       (unsigned long long)len,
2127 		       sizeof desc);
2128 		return -EINVAL;
2129 	}
2130 
2131 	ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect,
2132 			     UIO_MAXIOV, VHOST_ACCESS_RO);
2133 	if (unlikely(ret < 0)) {
2134 		if (ret != -EAGAIN)
2135 			vq_err(vq, "Translation failure %d in indirect.\n", ret);
2136 		return ret;
2137 	}
2138 	iov_iter_init(&from, READ, vq->indirect, ret, len);
2139 	count = len / sizeof desc;
2140 	/* Buffers are chained via a 16 bit next field, so
2141 	 * we can have at most 2^16 of these. */
2142 	if (unlikely(count > USHRT_MAX + 1)) {
2143 		vq_err(vq, "Indirect buffer length too big: %d\n",
2144 		       indirect->len);
2145 		return -E2BIG;
2146 	}
2147 
2148 	do {
2149 		unsigned iov_count = *in_num + *out_num;
2150 		if (unlikely(++found > count)) {
2151 			vq_err(vq, "Loop detected: last one at %u "
2152 			       "indirect size %u\n",
2153 			       i, count);
2154 			return -EINVAL;
2155 		}
2156 		if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) {
2157 			vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n",
2158 			       i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
2159 			return -EINVAL;
2160 		}
2161 		if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) {
2162 			vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n",
2163 			       i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
2164 			return -EINVAL;
2165 		}
2166 
2167 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2168 			access = VHOST_ACCESS_WO;
2169 		else
2170 			access = VHOST_ACCESS_RO;
2171 
2172 		ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2173 				     vhost32_to_cpu(vq, desc.len), iov + iov_count,
2174 				     iov_size - iov_count, access);
2175 		if (unlikely(ret < 0)) {
2176 			if (ret != -EAGAIN)
2177 				vq_err(vq, "Translation failure %d indirect idx %d\n",
2178 					ret, i);
2179 			return ret;
2180 		}
2181 		/* If this is an input descriptor, increment that count. */
2182 		if (access == VHOST_ACCESS_WO) {
2183 			*in_num += ret;
2184 			if (unlikely(log && ret)) {
2185 				log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2186 				log[*log_num].len = vhost32_to_cpu(vq, desc.len);
2187 				++*log_num;
2188 			}
2189 		} else {
2190 			/* If it's an output descriptor, they're all supposed
2191 			 * to come before any input descriptors. */
2192 			if (unlikely(*in_num)) {
2193 				vq_err(vq, "Indirect descriptor "
2194 				       "has out after in: idx %d\n", i);
2195 				return -EINVAL;
2196 			}
2197 			*out_num += ret;
2198 		}
2199 	} while ((i = next_desc(vq, &desc)) != -1);
2200 	return 0;
2201 }
2202 
2203 /* This looks in the virtqueue and for the first available buffer, and converts
2204  * it to an iovec for convenient access.  Since descriptors consist of some
2205  * number of output then some number of input descriptors, it's actually two
2206  * iovecs, but we pack them into one and note how many of each there were.
2207  *
2208  * This function returns the descriptor number found, or vq->num (which is
2209  * never a valid descriptor number) if none was found.  A negative code is
2210  * returned on error. */
vhost_get_vq_desc(struct vhost_virtqueue * vq,struct iovec iov[],unsigned int iov_size,unsigned int * out_num,unsigned int * in_num,struct vhost_log * log,unsigned int * log_num)2211 int vhost_get_vq_desc(struct vhost_virtqueue *vq,
2212 		      struct iovec iov[], unsigned int iov_size,
2213 		      unsigned int *out_num, unsigned int *in_num,
2214 		      struct vhost_log *log, unsigned int *log_num)
2215 {
2216 	struct vring_desc desc;
2217 	unsigned int i, head, found = 0;
2218 	u16 last_avail_idx;
2219 	__virtio16 avail_idx;
2220 	__virtio16 ring_head;
2221 	int ret, access;
2222 
2223 	/* Check it isn't doing very strange things with descriptor numbers. */
2224 	last_avail_idx = vq->last_avail_idx;
2225 
2226 	if (vq->avail_idx == vq->last_avail_idx) {
2227 		if (unlikely(vhost_get_avail_idx(vq, &avail_idx))) {
2228 			vq_err(vq, "Failed to access avail idx at %p\n",
2229 				&vq->avail->idx);
2230 			return -EFAULT;
2231 		}
2232 		vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
2233 
2234 		if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) {
2235 			vq_err(vq, "Guest moved used index from %u to %u",
2236 				last_avail_idx, vq->avail_idx);
2237 			return -EFAULT;
2238 		}
2239 
2240 		/* If there's nothing new since last we looked, return
2241 		 * invalid.
2242 		 */
2243 		if (vq->avail_idx == last_avail_idx)
2244 			return vq->num;
2245 
2246 		/* Only get avail ring entries after they have been
2247 		 * exposed by guest.
2248 		 */
2249 		smp_rmb();
2250 	}
2251 
2252 	/* Grab the next descriptor number they're advertising, and increment
2253 	 * the index we've seen. */
2254 	if (unlikely(vhost_get_avail_head(vq, &ring_head, last_avail_idx))) {
2255 		vq_err(vq, "Failed to read head: idx %d address %p\n",
2256 		       last_avail_idx,
2257 		       &vq->avail->ring[last_avail_idx % vq->num]);
2258 		return -EFAULT;
2259 	}
2260 
2261 	head = vhost16_to_cpu(vq, ring_head);
2262 
2263 	/* If their number is silly, that's an error. */
2264 	if (unlikely(head >= vq->num)) {
2265 		vq_err(vq, "Guest says index %u > %u is available",
2266 		       head, vq->num);
2267 		return -EINVAL;
2268 	}
2269 
2270 	/* When we start there are none of either input nor output. */
2271 	*out_num = *in_num = 0;
2272 	if (unlikely(log))
2273 		*log_num = 0;
2274 
2275 	i = head;
2276 	do {
2277 		unsigned iov_count = *in_num + *out_num;
2278 		if (unlikely(i >= vq->num)) {
2279 			vq_err(vq, "Desc index is %u > %u, head = %u",
2280 			       i, vq->num, head);
2281 			return -EINVAL;
2282 		}
2283 		if (unlikely(++found > vq->num)) {
2284 			vq_err(vq, "Loop detected: last one at %u "
2285 			       "vq size %u head %u\n",
2286 			       i, vq->num, head);
2287 			return -EINVAL;
2288 		}
2289 		ret = vhost_get_desc(vq, &desc, i);
2290 		if (unlikely(ret)) {
2291 			vq_err(vq, "Failed to get descriptor: idx %d addr %p\n",
2292 			       i, vq->desc + i);
2293 			return -EFAULT;
2294 		}
2295 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) {
2296 			ret = get_indirect(vq, iov, iov_size,
2297 					   out_num, in_num,
2298 					   log, log_num, &desc);
2299 			if (unlikely(ret < 0)) {
2300 				if (ret != -EAGAIN)
2301 					vq_err(vq, "Failure detected "
2302 						"in indirect descriptor at idx %d\n", i);
2303 				return ret;
2304 			}
2305 			continue;
2306 		}
2307 
2308 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2309 			access = VHOST_ACCESS_WO;
2310 		else
2311 			access = VHOST_ACCESS_RO;
2312 		ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2313 				     vhost32_to_cpu(vq, desc.len), iov + iov_count,
2314 				     iov_size - iov_count, access);
2315 		if (unlikely(ret < 0)) {
2316 			if (ret != -EAGAIN)
2317 				vq_err(vq, "Translation failure %d descriptor idx %d\n",
2318 					ret, i);
2319 			return ret;
2320 		}
2321 		if (access == VHOST_ACCESS_WO) {
2322 			/* If this is an input descriptor,
2323 			 * increment that count. */
2324 			*in_num += ret;
2325 			if (unlikely(log && ret)) {
2326 				log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2327 				log[*log_num].len = vhost32_to_cpu(vq, desc.len);
2328 				++*log_num;
2329 			}
2330 		} else {
2331 			/* If it's an output descriptor, they're all supposed
2332 			 * to come before any input descriptors. */
2333 			if (unlikely(*in_num)) {
2334 				vq_err(vq, "Descriptor has out after in: "
2335 				       "idx %d\n", i);
2336 				return -EINVAL;
2337 			}
2338 			*out_num += ret;
2339 		}
2340 	} while ((i = next_desc(vq, &desc)) != -1);
2341 
2342 	/* On success, increment avail index. */
2343 	vq->last_avail_idx++;
2344 
2345 	/* Assume notifications from guest are disabled at this point,
2346 	 * if they aren't we would need to update avail_event index. */
2347 	BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY));
2348 	return head;
2349 }
2350 EXPORT_SYMBOL_GPL(vhost_get_vq_desc);
2351 
2352 /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */
vhost_discard_vq_desc(struct vhost_virtqueue * vq,int n)2353 void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n)
2354 {
2355 	vq->last_avail_idx -= n;
2356 }
2357 EXPORT_SYMBOL_GPL(vhost_discard_vq_desc);
2358 
2359 /* After we've used one of their buffers, we tell them about it.  We'll then
2360  * want to notify the guest, using eventfd. */
vhost_add_used(struct vhost_virtqueue * vq,unsigned int head,int len)2361 int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len)
2362 {
2363 	struct vring_used_elem heads = {
2364 		cpu_to_vhost32(vq, head),
2365 		cpu_to_vhost32(vq, len)
2366 	};
2367 
2368 	return vhost_add_used_n(vq, &heads, 1);
2369 }
2370 EXPORT_SYMBOL_GPL(vhost_add_used);
2371 
__vhost_add_used_n(struct vhost_virtqueue * vq,struct vring_used_elem * heads,unsigned count)2372 static int __vhost_add_used_n(struct vhost_virtqueue *vq,
2373 			    struct vring_used_elem *heads,
2374 			    unsigned count)
2375 {
2376 	vring_used_elem_t __user *used;
2377 	u16 old, new;
2378 	int start;
2379 
2380 	start = vq->last_used_idx & (vq->num - 1);
2381 	used = vq->used->ring + start;
2382 	if (vhost_put_used(vq, heads, start, count)) {
2383 		vq_err(vq, "Failed to write used");
2384 		return -EFAULT;
2385 	}
2386 	if (unlikely(vq->log_used)) {
2387 		/* Make sure data is seen before log. */
2388 		smp_wmb();
2389 		/* Log used ring entry write. */
2390 		log_used(vq, ((void __user *)used - (void __user *)vq->used),
2391 			 count * sizeof *used);
2392 	}
2393 	old = vq->last_used_idx;
2394 	new = (vq->last_used_idx += count);
2395 	/* If the driver never bothers to signal in a very long while,
2396 	 * used index might wrap around. If that happens, invalidate
2397 	 * signalled_used index we stored. TODO: make sure driver
2398 	 * signals at least once in 2^16 and remove this. */
2399 	if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old)))
2400 		vq->signalled_used_valid = false;
2401 	return 0;
2402 }
2403 
2404 /* After we've used one of their buffers, we tell them about it.  We'll then
2405  * want to notify the guest, using eventfd. */
vhost_add_used_n(struct vhost_virtqueue * vq,struct vring_used_elem * heads,unsigned count)2406 int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads,
2407 		     unsigned count)
2408 {
2409 	int start, n, r;
2410 
2411 	start = vq->last_used_idx & (vq->num - 1);
2412 	n = vq->num - start;
2413 	if (n < count) {
2414 		r = __vhost_add_used_n(vq, heads, n);
2415 		if (r < 0)
2416 			return r;
2417 		heads += n;
2418 		count -= n;
2419 	}
2420 	r = __vhost_add_used_n(vq, heads, count);
2421 
2422 	/* Make sure buffer is written before we update index. */
2423 	smp_wmb();
2424 	if (vhost_put_used_idx(vq)) {
2425 		vq_err(vq, "Failed to increment used idx");
2426 		return -EFAULT;
2427 	}
2428 	if (unlikely(vq->log_used)) {
2429 		/* Make sure used idx is seen before log. */
2430 		smp_wmb();
2431 		/* Log used index update. */
2432 		log_used(vq, offsetof(struct vring_used, idx),
2433 			 sizeof vq->used->idx);
2434 		if (vq->log_ctx)
2435 			eventfd_signal(vq->log_ctx, 1);
2436 	}
2437 	return r;
2438 }
2439 EXPORT_SYMBOL_GPL(vhost_add_used_n);
2440 
vhost_notify(struct vhost_dev * dev,struct vhost_virtqueue * vq)2441 static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2442 {
2443 	__u16 old, new;
2444 	__virtio16 event;
2445 	bool v;
2446 	/* Flush out used index updates. This is paired
2447 	 * with the barrier that the Guest executes when enabling
2448 	 * interrupts. */
2449 	smp_mb();
2450 
2451 	if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2452 	    unlikely(vq->avail_idx == vq->last_avail_idx))
2453 		return true;
2454 
2455 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2456 		__virtio16 flags;
2457 		if (vhost_get_avail_flags(vq, &flags)) {
2458 			vq_err(vq, "Failed to get flags");
2459 			return true;
2460 		}
2461 		return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT));
2462 	}
2463 	old = vq->signalled_used;
2464 	v = vq->signalled_used_valid;
2465 	new = vq->signalled_used = vq->last_used_idx;
2466 	vq->signalled_used_valid = true;
2467 
2468 	if (unlikely(!v))
2469 		return true;
2470 
2471 	if (vhost_get_used_event(vq, &event)) {
2472 		vq_err(vq, "Failed to get used event idx");
2473 		return true;
2474 	}
2475 	return vring_need_event(vhost16_to_cpu(vq, event), new, old);
2476 }
2477 
2478 /* This actually signals the guest, using eventfd. */
vhost_signal(struct vhost_dev * dev,struct vhost_virtqueue * vq)2479 void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2480 {
2481 	/* Signal the Guest tell them we used something up. */
2482 	if (vq->call_ctx.ctx && vhost_notify(dev, vq))
2483 		eventfd_signal(vq->call_ctx.ctx, 1);
2484 }
2485 EXPORT_SYMBOL_GPL(vhost_signal);
2486 
2487 /* And here's the combo meal deal.  Supersize me! */
vhost_add_used_and_signal(struct vhost_dev * dev,struct vhost_virtqueue * vq,unsigned int head,int len)2488 void vhost_add_used_and_signal(struct vhost_dev *dev,
2489 			       struct vhost_virtqueue *vq,
2490 			       unsigned int head, int len)
2491 {
2492 	vhost_add_used(vq, head, len);
2493 	vhost_signal(dev, vq);
2494 }
2495 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal);
2496 
2497 /* multi-buffer version of vhost_add_used_and_signal */
vhost_add_used_and_signal_n(struct vhost_dev * dev,struct vhost_virtqueue * vq,struct vring_used_elem * heads,unsigned count)2498 void vhost_add_used_and_signal_n(struct vhost_dev *dev,
2499 				 struct vhost_virtqueue *vq,
2500 				 struct vring_used_elem *heads, unsigned count)
2501 {
2502 	vhost_add_used_n(vq, heads, count);
2503 	vhost_signal(dev, vq);
2504 }
2505 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n);
2506 
2507 /* return true if we're sure that avaiable ring is empty */
vhost_vq_avail_empty(struct vhost_dev * dev,struct vhost_virtqueue * vq)2508 bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2509 {
2510 	__virtio16 avail_idx;
2511 	int r;
2512 
2513 	if (vq->avail_idx != vq->last_avail_idx)
2514 		return false;
2515 
2516 	r = vhost_get_avail_idx(vq, &avail_idx);
2517 	if (unlikely(r))
2518 		return false;
2519 	vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
2520 
2521 	return vq->avail_idx == vq->last_avail_idx;
2522 }
2523 EXPORT_SYMBOL_GPL(vhost_vq_avail_empty);
2524 
2525 /* OK, now we need to know about added descriptors. */
vhost_enable_notify(struct vhost_dev * dev,struct vhost_virtqueue * vq)2526 bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2527 {
2528 	__virtio16 avail_idx;
2529 	int r;
2530 
2531 	if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY))
2532 		return false;
2533 	vq->used_flags &= ~VRING_USED_F_NO_NOTIFY;
2534 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2535 		r = vhost_update_used_flags(vq);
2536 		if (r) {
2537 			vq_err(vq, "Failed to enable notification at %p: %d\n",
2538 			       &vq->used->flags, r);
2539 			return false;
2540 		}
2541 	} else {
2542 		r = vhost_update_avail_event(vq);
2543 		if (r) {
2544 			vq_err(vq, "Failed to update avail event index at %p: %d\n",
2545 			       vhost_avail_event(vq), r);
2546 			return false;
2547 		}
2548 	}
2549 	/* They could have slipped one in as we were doing that: make
2550 	 * sure it's written, then check again. */
2551 	smp_mb();
2552 	r = vhost_get_avail_idx(vq, &avail_idx);
2553 	if (r) {
2554 		vq_err(vq, "Failed to check avail idx at %p: %d\n",
2555 		       &vq->avail->idx, r);
2556 		return false;
2557 	}
2558 	vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
2559 
2560 	return vq->avail_idx != vq->last_avail_idx;
2561 }
2562 EXPORT_SYMBOL_GPL(vhost_enable_notify);
2563 
2564 /* We don't need to be notified again. */
vhost_disable_notify(struct vhost_dev * dev,struct vhost_virtqueue * vq)2565 void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2566 {
2567 	int r;
2568 
2569 	if (vq->used_flags & VRING_USED_F_NO_NOTIFY)
2570 		return;
2571 	vq->used_flags |= VRING_USED_F_NO_NOTIFY;
2572 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2573 		r = vhost_update_used_flags(vq);
2574 		if (r)
2575 			vq_err(vq, "Failed to disable notification at %p: %d\n",
2576 			       &vq->used->flags, r);
2577 	}
2578 }
2579 EXPORT_SYMBOL_GPL(vhost_disable_notify);
2580 
2581 /* Create a new message. */
vhost_new_msg(struct vhost_virtqueue * vq,int type)2582 struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type)
2583 {
2584 	struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL);
2585 	if (!node)
2586 		return NULL;
2587 
2588 	/* Make sure all padding within the structure is initialized. */
2589 	memset(&node->msg, 0, sizeof node->msg);
2590 	node->vq = vq;
2591 	node->msg.type = type;
2592 	return node;
2593 }
2594 EXPORT_SYMBOL_GPL(vhost_new_msg);
2595 
vhost_enqueue_msg(struct vhost_dev * dev,struct list_head * head,struct vhost_msg_node * node)2596 void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head,
2597 		       struct vhost_msg_node *node)
2598 {
2599 	spin_lock(&dev->iotlb_lock);
2600 	list_add_tail(&node->node, head);
2601 	spin_unlock(&dev->iotlb_lock);
2602 
2603 	wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
2604 }
2605 EXPORT_SYMBOL_GPL(vhost_enqueue_msg);
2606 
vhost_dequeue_msg(struct vhost_dev * dev,struct list_head * head)2607 struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
2608 					 struct list_head *head)
2609 {
2610 	struct vhost_msg_node *node = NULL;
2611 
2612 	spin_lock(&dev->iotlb_lock);
2613 	if (!list_empty(head)) {
2614 		node = list_first_entry(head, struct vhost_msg_node,
2615 					node);
2616 		list_del(&node->node);
2617 	}
2618 	spin_unlock(&dev->iotlb_lock);
2619 
2620 	return node;
2621 }
2622 EXPORT_SYMBOL_GPL(vhost_dequeue_msg);
2623 
vhost_set_backend_features(struct vhost_dev * dev,u64 features)2624 void vhost_set_backend_features(struct vhost_dev *dev, u64 features)
2625 {
2626 	struct vhost_virtqueue *vq;
2627 	int i;
2628 
2629 	mutex_lock(&dev->mutex);
2630 	for (i = 0; i < dev->nvqs; ++i) {
2631 		vq = dev->vqs[i];
2632 		mutex_lock(&vq->mutex);
2633 		vq->acked_backend_features = features;
2634 		mutex_unlock(&vq->mutex);
2635 	}
2636 	mutex_unlock(&dev->mutex);
2637 }
2638 EXPORT_SYMBOL_GPL(vhost_set_backend_features);
2639 
vhost_init(void)2640 static int __init vhost_init(void)
2641 {
2642 	return 0;
2643 }
2644 
vhost_exit(void)2645 static void __exit vhost_exit(void)
2646 {
2647 }
2648 
2649 module_init(vhost_init);
2650 module_exit(vhost_exit);
2651 
2652 MODULE_VERSION("0.0.1");
2653 MODULE_LICENSE("GPL v2");
2654 MODULE_AUTHOR("Michael S. Tsirkin");
2655 MODULE_DESCRIPTION("Host kernel accelerator for virtio");
2656