xref: /linux/drivers/vhost/vhost.c (revision a1ff5a7d78a036d6c2178ee5acd6ba4946243800)
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/module.h>
26 #include <linux/sort.h>
27 #include <linux/sched/mm.h>
28 #include <linux/sched/signal.h>
29 #include <linux/sched/vhost_task.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,struct vhost_virtqueue * vq)189 void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
190 		     __poll_t mask, struct vhost_dev *dev,
191 		     struct vhost_virtqueue *vq)
192 {
193 	init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup);
194 	init_poll_funcptr(&poll->table, vhost_poll_func);
195 	poll->mask = mask;
196 	poll->dev = dev;
197 	poll->wqh = NULL;
198 	poll->vq = vq;
199 
200 	vhost_work_init(&poll->work, fn);
201 }
202 EXPORT_SYMBOL_GPL(vhost_poll_init);
203 
204 /* Start polling a file. We add ourselves to file's wait queue. The caller must
205  * keep a reference to a file until after vhost_poll_stop is called. */
vhost_poll_start(struct vhost_poll * poll,struct file * file)206 int vhost_poll_start(struct vhost_poll *poll, struct file *file)
207 {
208 	__poll_t mask;
209 
210 	if (poll->wqh)
211 		return 0;
212 
213 	mask = vfs_poll(file, &poll->table);
214 	if (mask)
215 		vhost_poll_wakeup(&poll->wait, 0, 0, poll_to_key(mask));
216 	if (mask & EPOLLERR) {
217 		vhost_poll_stop(poll);
218 		return -EINVAL;
219 	}
220 
221 	return 0;
222 }
223 EXPORT_SYMBOL_GPL(vhost_poll_start);
224 
225 /* Stop polling a file. After this function returns, it becomes safe to drop the
226  * file reference. You must also flush afterwards. */
vhost_poll_stop(struct vhost_poll * poll)227 void vhost_poll_stop(struct vhost_poll *poll)
228 {
229 	if (poll->wqh) {
230 		remove_wait_queue(poll->wqh, &poll->wait);
231 		poll->wqh = NULL;
232 	}
233 }
234 EXPORT_SYMBOL_GPL(vhost_poll_stop);
235 
vhost_worker_queue(struct vhost_worker * worker,struct vhost_work * work)236 static void vhost_worker_queue(struct vhost_worker *worker,
237 			       struct vhost_work *work)
238 {
239 	if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) {
240 		/* We can only add the work to the list after we're
241 		 * sure it was not in the list.
242 		 * test_and_set_bit() implies a memory barrier.
243 		 */
244 		llist_add(&work->node, &worker->work_list);
245 		vhost_task_wake(worker->vtsk);
246 	}
247 }
248 
vhost_vq_work_queue(struct vhost_virtqueue * vq,struct vhost_work * work)249 bool vhost_vq_work_queue(struct vhost_virtqueue *vq, struct vhost_work *work)
250 {
251 	struct vhost_worker *worker;
252 	bool queued = false;
253 
254 	rcu_read_lock();
255 	worker = rcu_dereference(vq->worker);
256 	if (worker) {
257 		queued = true;
258 		vhost_worker_queue(worker, work);
259 	}
260 	rcu_read_unlock();
261 
262 	return queued;
263 }
264 EXPORT_SYMBOL_GPL(vhost_vq_work_queue);
265 
266 /**
267  * __vhost_worker_flush - flush a worker
268  * @worker: worker to flush
269  *
270  * The worker's flush_mutex must be held.
271  */
__vhost_worker_flush(struct vhost_worker * worker)272 static void __vhost_worker_flush(struct vhost_worker *worker)
273 {
274 	struct vhost_flush_struct flush;
275 
276 	if (!worker->attachment_cnt || worker->killed)
277 		return;
278 
279 	init_completion(&flush.wait_event);
280 	vhost_work_init(&flush.work, vhost_flush_work);
281 
282 	vhost_worker_queue(worker, &flush.work);
283 	/*
284 	 * Drop mutex in case our worker is killed and it needs to take the
285 	 * mutex to force cleanup.
286 	 */
287 	mutex_unlock(&worker->mutex);
288 	wait_for_completion(&flush.wait_event);
289 	mutex_lock(&worker->mutex);
290 }
291 
vhost_worker_flush(struct vhost_worker * worker)292 static void vhost_worker_flush(struct vhost_worker *worker)
293 {
294 	mutex_lock(&worker->mutex);
295 	__vhost_worker_flush(worker);
296 	mutex_unlock(&worker->mutex);
297 }
298 
vhost_dev_flush(struct vhost_dev * dev)299 void vhost_dev_flush(struct vhost_dev *dev)
300 {
301 	struct vhost_worker *worker;
302 	unsigned long i;
303 
304 	xa_for_each(&dev->worker_xa, i, worker)
305 		vhost_worker_flush(worker);
306 }
307 EXPORT_SYMBOL_GPL(vhost_dev_flush);
308 
309 /* A lockless hint for busy polling code to exit the loop */
vhost_vq_has_work(struct vhost_virtqueue * vq)310 bool vhost_vq_has_work(struct vhost_virtqueue *vq)
311 {
312 	struct vhost_worker *worker;
313 	bool has_work = false;
314 
315 	rcu_read_lock();
316 	worker = rcu_dereference(vq->worker);
317 	if (worker && !llist_empty(&worker->work_list))
318 		has_work = true;
319 	rcu_read_unlock();
320 
321 	return has_work;
322 }
323 EXPORT_SYMBOL_GPL(vhost_vq_has_work);
324 
vhost_poll_queue(struct vhost_poll * poll)325 void vhost_poll_queue(struct vhost_poll *poll)
326 {
327 	vhost_vq_work_queue(poll->vq, &poll->work);
328 }
329 EXPORT_SYMBOL_GPL(vhost_poll_queue);
330 
__vhost_vq_meta_reset(struct vhost_virtqueue * vq)331 static void __vhost_vq_meta_reset(struct vhost_virtqueue *vq)
332 {
333 	int j;
334 
335 	for (j = 0; j < VHOST_NUM_ADDRS; j++)
336 		vq->meta_iotlb[j] = NULL;
337 }
338 
vhost_vq_meta_reset(struct vhost_dev * d)339 static void vhost_vq_meta_reset(struct vhost_dev *d)
340 {
341 	int i;
342 
343 	for (i = 0; i < d->nvqs; ++i)
344 		__vhost_vq_meta_reset(d->vqs[i]);
345 }
346 
vhost_vring_call_reset(struct vhost_vring_call * call_ctx)347 static void vhost_vring_call_reset(struct vhost_vring_call *call_ctx)
348 {
349 	call_ctx->ctx = NULL;
350 	memset(&call_ctx->producer, 0x0, sizeof(struct irq_bypass_producer));
351 }
352 
vhost_vq_is_setup(struct vhost_virtqueue * vq)353 bool vhost_vq_is_setup(struct vhost_virtqueue *vq)
354 {
355 	return vq->avail && vq->desc && vq->used && vhost_vq_access_ok(vq);
356 }
357 EXPORT_SYMBOL_GPL(vhost_vq_is_setup);
358 
vhost_vq_reset(struct vhost_dev * dev,struct vhost_virtqueue * vq)359 static void vhost_vq_reset(struct vhost_dev *dev,
360 			   struct vhost_virtqueue *vq)
361 {
362 	vq->num = 1;
363 	vq->desc = NULL;
364 	vq->avail = NULL;
365 	vq->used = NULL;
366 	vq->last_avail_idx = 0;
367 	vq->avail_idx = 0;
368 	vq->last_used_idx = 0;
369 	vq->signalled_used = 0;
370 	vq->signalled_used_valid = false;
371 	vq->used_flags = 0;
372 	vq->log_used = false;
373 	vq->log_addr = -1ull;
374 	vq->private_data = NULL;
375 	vq->acked_features = 0;
376 	vq->acked_backend_features = 0;
377 	vq->log_base = NULL;
378 	vq->error_ctx = NULL;
379 	vq->kick = NULL;
380 	vq->log_ctx = NULL;
381 	vhost_disable_cross_endian(vq);
382 	vhost_reset_is_le(vq);
383 	vq->busyloop_timeout = 0;
384 	vq->umem = NULL;
385 	vq->iotlb = NULL;
386 	rcu_assign_pointer(vq->worker, NULL);
387 	vhost_vring_call_reset(&vq->call_ctx);
388 	__vhost_vq_meta_reset(vq);
389 }
390 
vhost_run_work_list(void * data)391 static bool vhost_run_work_list(void *data)
392 {
393 	struct vhost_worker *worker = data;
394 	struct vhost_work *work, *work_next;
395 	struct llist_node *node;
396 
397 	node = llist_del_all(&worker->work_list);
398 	if (node) {
399 		__set_current_state(TASK_RUNNING);
400 
401 		node = llist_reverse_order(node);
402 		/* make sure flag is seen after deletion */
403 		smp_wmb();
404 		llist_for_each_entry_safe(work, work_next, node, node) {
405 			clear_bit(VHOST_WORK_QUEUED, &work->flags);
406 			kcov_remote_start_common(worker->kcov_handle);
407 			work->fn(work);
408 			kcov_remote_stop();
409 			cond_resched();
410 		}
411 	}
412 
413 	return !!node;
414 }
415 
vhost_worker_killed(void * data)416 static void vhost_worker_killed(void *data)
417 {
418 	struct vhost_worker *worker = data;
419 	struct vhost_dev *dev = worker->dev;
420 	struct vhost_virtqueue *vq;
421 	int i, attach_cnt = 0;
422 
423 	mutex_lock(&worker->mutex);
424 	worker->killed = true;
425 
426 	for (i = 0; i < dev->nvqs; i++) {
427 		vq = dev->vqs[i];
428 
429 		mutex_lock(&vq->mutex);
430 		if (worker ==
431 		    rcu_dereference_check(vq->worker,
432 					  lockdep_is_held(&vq->mutex))) {
433 			rcu_assign_pointer(vq->worker, NULL);
434 			attach_cnt++;
435 		}
436 		mutex_unlock(&vq->mutex);
437 	}
438 
439 	worker->attachment_cnt -= attach_cnt;
440 	if (attach_cnt)
441 		synchronize_rcu();
442 	/*
443 	 * Finish vhost_worker_flush calls and any other works that snuck in
444 	 * before the synchronize_rcu.
445 	 */
446 	vhost_run_work_list(worker);
447 	mutex_unlock(&worker->mutex);
448 }
449 
vhost_vq_free_iovecs(struct vhost_virtqueue * vq)450 static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq)
451 {
452 	kfree(vq->indirect);
453 	vq->indirect = NULL;
454 	kfree(vq->log);
455 	vq->log = NULL;
456 	kfree(vq->heads);
457 	vq->heads = NULL;
458 }
459 
460 /* Helper to allocate iovec buffers for all vqs. */
vhost_dev_alloc_iovecs(struct vhost_dev * dev)461 static long vhost_dev_alloc_iovecs(struct vhost_dev *dev)
462 {
463 	struct vhost_virtqueue *vq;
464 	int i;
465 
466 	for (i = 0; i < dev->nvqs; ++i) {
467 		vq = dev->vqs[i];
468 		vq->indirect = kmalloc_array(UIO_MAXIOV,
469 					     sizeof(*vq->indirect),
470 					     GFP_KERNEL);
471 		vq->log = kmalloc_array(dev->iov_limit, sizeof(*vq->log),
472 					GFP_KERNEL);
473 		vq->heads = kmalloc_array(dev->iov_limit, sizeof(*vq->heads),
474 					  GFP_KERNEL);
475 		if (!vq->indirect || !vq->log || !vq->heads)
476 			goto err_nomem;
477 	}
478 	return 0;
479 
480 err_nomem:
481 	for (; i >= 0; --i)
482 		vhost_vq_free_iovecs(dev->vqs[i]);
483 	return -ENOMEM;
484 }
485 
vhost_dev_free_iovecs(struct vhost_dev * dev)486 static void vhost_dev_free_iovecs(struct vhost_dev *dev)
487 {
488 	int i;
489 
490 	for (i = 0; i < dev->nvqs; ++i)
491 		vhost_vq_free_iovecs(dev->vqs[i]);
492 }
493 
vhost_exceeds_weight(struct vhost_virtqueue * vq,int pkts,int total_len)494 bool vhost_exceeds_weight(struct vhost_virtqueue *vq,
495 			  int pkts, int total_len)
496 {
497 	struct vhost_dev *dev = vq->dev;
498 
499 	if ((dev->byte_weight && total_len >= dev->byte_weight) ||
500 	    pkts >= dev->weight) {
501 		vhost_poll_queue(&vq->poll);
502 		return true;
503 	}
504 
505 	return false;
506 }
507 EXPORT_SYMBOL_GPL(vhost_exceeds_weight);
508 
vhost_get_avail_size(struct vhost_virtqueue * vq,unsigned int num)509 static size_t vhost_get_avail_size(struct vhost_virtqueue *vq,
510 				   unsigned int num)
511 {
512 	size_t event __maybe_unused =
513 	       vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
514 
515 	return size_add(struct_size(vq->avail, ring, num), event);
516 }
517 
vhost_get_used_size(struct vhost_virtqueue * vq,unsigned int num)518 static size_t vhost_get_used_size(struct vhost_virtqueue *vq,
519 				  unsigned int num)
520 {
521 	size_t event __maybe_unused =
522 	       vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
523 
524 	return size_add(struct_size(vq->used, ring, num), event);
525 }
526 
vhost_get_desc_size(struct vhost_virtqueue * vq,unsigned int num)527 static size_t vhost_get_desc_size(struct vhost_virtqueue *vq,
528 				  unsigned int num)
529 {
530 	return sizeof(*vq->desc) * num;
531 }
532 
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))533 void vhost_dev_init(struct vhost_dev *dev,
534 		    struct vhost_virtqueue **vqs, int nvqs,
535 		    int iov_limit, int weight, int byte_weight,
536 		    bool use_worker,
537 		    int (*msg_handler)(struct vhost_dev *dev, u32 asid,
538 				       struct vhost_iotlb_msg *msg))
539 {
540 	struct vhost_virtqueue *vq;
541 	int i;
542 
543 	dev->vqs = vqs;
544 	dev->nvqs = nvqs;
545 	mutex_init(&dev->mutex);
546 	dev->log_ctx = NULL;
547 	dev->umem = NULL;
548 	dev->iotlb = NULL;
549 	dev->mm = NULL;
550 	dev->iov_limit = iov_limit;
551 	dev->weight = weight;
552 	dev->byte_weight = byte_weight;
553 	dev->use_worker = use_worker;
554 	dev->msg_handler = msg_handler;
555 	init_waitqueue_head(&dev->wait);
556 	INIT_LIST_HEAD(&dev->read_list);
557 	INIT_LIST_HEAD(&dev->pending_list);
558 	spin_lock_init(&dev->iotlb_lock);
559 	xa_init_flags(&dev->worker_xa, XA_FLAGS_ALLOC);
560 
561 	for (i = 0; i < dev->nvqs; ++i) {
562 		vq = dev->vqs[i];
563 		vq->log = NULL;
564 		vq->indirect = NULL;
565 		vq->heads = NULL;
566 		vq->dev = dev;
567 		mutex_init(&vq->mutex);
568 		vhost_vq_reset(dev, vq);
569 		if (vq->handle_kick)
570 			vhost_poll_init(&vq->poll, vq->handle_kick,
571 					EPOLLIN, dev, vq);
572 	}
573 }
574 EXPORT_SYMBOL_GPL(vhost_dev_init);
575 
576 /* Caller should have device mutex */
vhost_dev_check_owner(struct vhost_dev * dev)577 long vhost_dev_check_owner(struct vhost_dev *dev)
578 {
579 	/* Are you the owner? If not, I don't think you mean to do that */
580 	return dev->mm == current->mm ? 0 : -EPERM;
581 }
582 EXPORT_SYMBOL_GPL(vhost_dev_check_owner);
583 
584 /* Caller should have device mutex */
vhost_dev_has_owner(struct vhost_dev * dev)585 bool vhost_dev_has_owner(struct vhost_dev *dev)
586 {
587 	return dev->mm;
588 }
589 EXPORT_SYMBOL_GPL(vhost_dev_has_owner);
590 
vhost_attach_mm(struct vhost_dev * dev)591 static void vhost_attach_mm(struct vhost_dev *dev)
592 {
593 	/* No owner, become one */
594 	if (dev->use_worker) {
595 		dev->mm = get_task_mm(current);
596 	} else {
597 		/* vDPA device does not use worker thead, so there's
598 		 * no need to hold the address space for mm. This help
599 		 * to avoid deadlock in the case of mmap() which may
600 		 * held the refcnt of the file and depends on release
601 		 * method to remove vma.
602 		 */
603 		dev->mm = current->mm;
604 		mmgrab(dev->mm);
605 	}
606 }
607 
vhost_detach_mm(struct vhost_dev * dev)608 static void vhost_detach_mm(struct vhost_dev *dev)
609 {
610 	if (!dev->mm)
611 		return;
612 
613 	if (dev->use_worker)
614 		mmput(dev->mm);
615 	else
616 		mmdrop(dev->mm);
617 
618 	dev->mm = NULL;
619 }
620 
vhost_worker_destroy(struct vhost_dev * dev,struct vhost_worker * worker)621 static void vhost_worker_destroy(struct vhost_dev *dev,
622 				 struct vhost_worker *worker)
623 {
624 	if (!worker)
625 		return;
626 
627 	WARN_ON(!llist_empty(&worker->work_list));
628 	xa_erase(&dev->worker_xa, worker->id);
629 	vhost_task_stop(worker->vtsk);
630 	kfree(worker);
631 }
632 
vhost_workers_free(struct vhost_dev * dev)633 static void vhost_workers_free(struct vhost_dev *dev)
634 {
635 	struct vhost_worker *worker;
636 	unsigned long i;
637 
638 	if (!dev->use_worker)
639 		return;
640 
641 	for (i = 0; i < dev->nvqs; i++)
642 		rcu_assign_pointer(dev->vqs[i]->worker, NULL);
643 	/*
644 	 * Free the default worker we created and cleanup workers userspace
645 	 * created but couldn't clean up (it forgot or crashed).
646 	 */
647 	xa_for_each(&dev->worker_xa, i, worker)
648 		vhost_worker_destroy(dev, worker);
649 	xa_destroy(&dev->worker_xa);
650 }
651 
vhost_worker_create(struct vhost_dev * dev)652 static struct vhost_worker *vhost_worker_create(struct vhost_dev *dev)
653 {
654 	struct vhost_worker *worker;
655 	struct vhost_task *vtsk;
656 	char name[TASK_COMM_LEN];
657 	int ret;
658 	u32 id;
659 
660 	worker = kzalloc(sizeof(*worker), GFP_KERNEL_ACCOUNT);
661 	if (!worker)
662 		return NULL;
663 
664 	worker->dev = dev;
665 	snprintf(name, sizeof(name), "vhost-%d", current->pid);
666 
667 	vtsk = vhost_task_create(vhost_run_work_list, vhost_worker_killed,
668 				 worker, name);
669 	if (!vtsk)
670 		goto free_worker;
671 
672 	mutex_init(&worker->mutex);
673 	init_llist_head(&worker->work_list);
674 	worker->kcov_handle = kcov_common_handle();
675 	worker->vtsk = vtsk;
676 
677 	vhost_task_start(vtsk);
678 
679 	ret = xa_alloc(&dev->worker_xa, &id, worker, xa_limit_32b, GFP_KERNEL);
680 	if (ret < 0)
681 		goto stop_worker;
682 	worker->id = id;
683 
684 	return worker;
685 
686 stop_worker:
687 	vhost_task_stop(vtsk);
688 free_worker:
689 	kfree(worker);
690 	return NULL;
691 }
692 
693 /* Caller must have device mutex */
__vhost_vq_attach_worker(struct vhost_virtqueue * vq,struct vhost_worker * worker)694 static void __vhost_vq_attach_worker(struct vhost_virtqueue *vq,
695 				     struct vhost_worker *worker)
696 {
697 	struct vhost_worker *old_worker;
698 
699 	mutex_lock(&worker->mutex);
700 	if (worker->killed) {
701 		mutex_unlock(&worker->mutex);
702 		return;
703 	}
704 
705 	mutex_lock(&vq->mutex);
706 
707 	old_worker = rcu_dereference_check(vq->worker,
708 					   lockdep_is_held(&vq->mutex));
709 	rcu_assign_pointer(vq->worker, worker);
710 	worker->attachment_cnt++;
711 
712 	if (!old_worker) {
713 		mutex_unlock(&vq->mutex);
714 		mutex_unlock(&worker->mutex);
715 		return;
716 	}
717 	mutex_unlock(&vq->mutex);
718 	mutex_unlock(&worker->mutex);
719 
720 	/*
721 	 * Take the worker mutex to make sure we see the work queued from
722 	 * device wide flushes which doesn't use RCU for execution.
723 	 */
724 	mutex_lock(&old_worker->mutex);
725 	if (old_worker->killed) {
726 		mutex_unlock(&old_worker->mutex);
727 		return;
728 	}
729 
730 	/*
731 	 * We don't want to call synchronize_rcu for every vq during setup
732 	 * because it will slow down VM startup. If we haven't done
733 	 * VHOST_SET_VRING_KICK and not done the driver specific
734 	 * SET_ENDPOINT/RUNNUNG then we can skip the sync since there will
735 	 * not be any works queued for scsi and net.
736 	 */
737 	mutex_lock(&vq->mutex);
738 	if (!vhost_vq_get_backend(vq) && !vq->kick) {
739 		mutex_unlock(&vq->mutex);
740 
741 		old_worker->attachment_cnt--;
742 		mutex_unlock(&old_worker->mutex);
743 		/*
744 		 * vsock can queue anytime after VHOST_VSOCK_SET_GUEST_CID.
745 		 * Warn if it adds support for multiple workers but forgets to
746 		 * handle the early queueing case.
747 		 */
748 		WARN_ON(!old_worker->attachment_cnt &&
749 			!llist_empty(&old_worker->work_list));
750 		return;
751 	}
752 	mutex_unlock(&vq->mutex);
753 
754 	/* Make sure new vq queue/flush/poll calls see the new worker */
755 	synchronize_rcu();
756 	/* Make sure whatever was queued gets run */
757 	__vhost_worker_flush(old_worker);
758 	old_worker->attachment_cnt--;
759 	mutex_unlock(&old_worker->mutex);
760 }
761 
762  /* Caller must have device mutex */
vhost_vq_attach_worker(struct vhost_virtqueue * vq,struct vhost_vring_worker * info)763 static int vhost_vq_attach_worker(struct vhost_virtqueue *vq,
764 				  struct vhost_vring_worker *info)
765 {
766 	unsigned long index = info->worker_id;
767 	struct vhost_dev *dev = vq->dev;
768 	struct vhost_worker *worker;
769 
770 	if (!dev->use_worker)
771 		return -EINVAL;
772 
773 	worker = xa_find(&dev->worker_xa, &index, UINT_MAX, XA_PRESENT);
774 	if (!worker || worker->id != info->worker_id)
775 		return -ENODEV;
776 
777 	__vhost_vq_attach_worker(vq, worker);
778 	return 0;
779 }
780 
781 /* Caller must have device mutex */
vhost_new_worker(struct vhost_dev * dev,struct vhost_worker_state * info)782 static int vhost_new_worker(struct vhost_dev *dev,
783 			    struct vhost_worker_state *info)
784 {
785 	struct vhost_worker *worker;
786 
787 	worker = vhost_worker_create(dev);
788 	if (!worker)
789 		return -ENOMEM;
790 
791 	info->worker_id = worker->id;
792 	return 0;
793 }
794 
795 /* Caller must have device mutex */
vhost_free_worker(struct vhost_dev * dev,struct vhost_worker_state * info)796 static int vhost_free_worker(struct vhost_dev *dev,
797 			     struct vhost_worker_state *info)
798 {
799 	unsigned long index = info->worker_id;
800 	struct vhost_worker *worker;
801 
802 	worker = xa_find(&dev->worker_xa, &index, UINT_MAX, XA_PRESENT);
803 	if (!worker || worker->id != info->worker_id)
804 		return -ENODEV;
805 
806 	mutex_lock(&worker->mutex);
807 	if (worker->attachment_cnt || worker->killed) {
808 		mutex_unlock(&worker->mutex);
809 		return -EBUSY;
810 	}
811 	/*
812 	 * A flush might have raced and snuck in before attachment_cnt was set
813 	 * to zero. Make sure flushes are flushed from the queue before
814 	 * freeing.
815 	 */
816 	__vhost_worker_flush(worker);
817 	mutex_unlock(&worker->mutex);
818 
819 	vhost_worker_destroy(dev, worker);
820 	return 0;
821 }
822 
vhost_get_vq_from_user(struct vhost_dev * dev,void __user * argp,struct vhost_virtqueue ** vq,u32 * id)823 static int vhost_get_vq_from_user(struct vhost_dev *dev, void __user *argp,
824 				  struct vhost_virtqueue **vq, u32 *id)
825 {
826 	u32 __user *idxp = argp;
827 	u32 idx;
828 	long r;
829 
830 	r = get_user(idx, idxp);
831 	if (r < 0)
832 		return r;
833 
834 	if (idx >= dev->nvqs)
835 		return -ENOBUFS;
836 
837 	idx = array_index_nospec(idx, dev->nvqs);
838 
839 	*vq = dev->vqs[idx];
840 	*id = idx;
841 	return 0;
842 }
843 
844 /* Caller must have device mutex */
vhost_worker_ioctl(struct vhost_dev * dev,unsigned int ioctl,void __user * argp)845 long vhost_worker_ioctl(struct vhost_dev *dev, unsigned int ioctl,
846 			void __user *argp)
847 {
848 	struct vhost_vring_worker ring_worker;
849 	struct vhost_worker_state state;
850 	struct vhost_worker *worker;
851 	struct vhost_virtqueue *vq;
852 	long ret;
853 	u32 idx;
854 
855 	if (!dev->use_worker)
856 		return -EINVAL;
857 
858 	if (!vhost_dev_has_owner(dev))
859 		return -EINVAL;
860 
861 	ret = vhost_dev_check_owner(dev);
862 	if (ret)
863 		return ret;
864 
865 	switch (ioctl) {
866 	/* dev worker ioctls */
867 	case VHOST_NEW_WORKER:
868 		ret = vhost_new_worker(dev, &state);
869 		if (!ret && copy_to_user(argp, &state, sizeof(state)))
870 			ret = -EFAULT;
871 		return ret;
872 	case VHOST_FREE_WORKER:
873 		if (copy_from_user(&state, argp, sizeof(state)))
874 			return -EFAULT;
875 		return vhost_free_worker(dev, &state);
876 	/* vring worker ioctls */
877 	case VHOST_ATTACH_VRING_WORKER:
878 	case VHOST_GET_VRING_WORKER:
879 		break;
880 	default:
881 		return -ENOIOCTLCMD;
882 	}
883 
884 	ret = vhost_get_vq_from_user(dev, argp, &vq, &idx);
885 	if (ret)
886 		return ret;
887 
888 	switch (ioctl) {
889 	case VHOST_ATTACH_VRING_WORKER:
890 		if (copy_from_user(&ring_worker, argp, sizeof(ring_worker))) {
891 			ret = -EFAULT;
892 			break;
893 		}
894 
895 		ret = vhost_vq_attach_worker(vq, &ring_worker);
896 		break;
897 	case VHOST_GET_VRING_WORKER:
898 		worker = rcu_dereference_check(vq->worker,
899 					       lockdep_is_held(&dev->mutex));
900 		if (!worker) {
901 			ret = -EINVAL;
902 			break;
903 		}
904 
905 		ring_worker.index = idx;
906 		ring_worker.worker_id = worker->id;
907 
908 		if (copy_to_user(argp, &ring_worker, sizeof(ring_worker)))
909 			ret = -EFAULT;
910 		break;
911 	default:
912 		ret = -ENOIOCTLCMD;
913 		break;
914 	}
915 
916 	return ret;
917 }
918 EXPORT_SYMBOL_GPL(vhost_worker_ioctl);
919 
920 /* Caller should have device mutex */
vhost_dev_set_owner(struct vhost_dev * dev)921 long vhost_dev_set_owner(struct vhost_dev *dev)
922 {
923 	struct vhost_worker *worker;
924 	int err, i;
925 
926 	/* Is there an owner already? */
927 	if (vhost_dev_has_owner(dev)) {
928 		err = -EBUSY;
929 		goto err_mm;
930 	}
931 
932 	vhost_attach_mm(dev);
933 
934 	err = vhost_dev_alloc_iovecs(dev);
935 	if (err)
936 		goto err_iovecs;
937 
938 	if (dev->use_worker) {
939 		/*
940 		 * This should be done last, because vsock can queue work
941 		 * before VHOST_SET_OWNER so it simplifies the failure path
942 		 * below since we don't have to worry about vsock queueing
943 		 * while we free the worker.
944 		 */
945 		worker = vhost_worker_create(dev);
946 		if (!worker) {
947 			err = -ENOMEM;
948 			goto err_worker;
949 		}
950 
951 		for (i = 0; i < dev->nvqs; i++)
952 			__vhost_vq_attach_worker(dev->vqs[i], worker);
953 	}
954 
955 	return 0;
956 
957 err_worker:
958 	vhost_dev_free_iovecs(dev);
959 err_iovecs:
960 	vhost_detach_mm(dev);
961 err_mm:
962 	return err;
963 }
964 EXPORT_SYMBOL_GPL(vhost_dev_set_owner);
965 
iotlb_alloc(void)966 static struct vhost_iotlb *iotlb_alloc(void)
967 {
968 	return vhost_iotlb_alloc(max_iotlb_entries,
969 				 VHOST_IOTLB_FLAG_RETIRE);
970 }
971 
vhost_dev_reset_owner_prepare(void)972 struct vhost_iotlb *vhost_dev_reset_owner_prepare(void)
973 {
974 	return iotlb_alloc();
975 }
976 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare);
977 
978 /* Caller should have device mutex */
vhost_dev_reset_owner(struct vhost_dev * dev,struct vhost_iotlb * umem)979 void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_iotlb *umem)
980 {
981 	int i;
982 
983 	vhost_dev_cleanup(dev);
984 
985 	dev->umem = umem;
986 	/* We don't need VQ locks below since vhost_dev_cleanup makes sure
987 	 * VQs aren't running.
988 	 */
989 	for (i = 0; i < dev->nvqs; ++i)
990 		dev->vqs[i]->umem = umem;
991 }
992 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner);
993 
vhost_dev_stop(struct vhost_dev * dev)994 void vhost_dev_stop(struct vhost_dev *dev)
995 {
996 	int i;
997 
998 	for (i = 0; i < dev->nvqs; ++i) {
999 		if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick)
1000 			vhost_poll_stop(&dev->vqs[i]->poll);
1001 	}
1002 
1003 	vhost_dev_flush(dev);
1004 }
1005 EXPORT_SYMBOL_GPL(vhost_dev_stop);
1006 
vhost_clear_msg(struct vhost_dev * dev)1007 void vhost_clear_msg(struct vhost_dev *dev)
1008 {
1009 	struct vhost_msg_node *node, *n;
1010 
1011 	spin_lock(&dev->iotlb_lock);
1012 
1013 	list_for_each_entry_safe(node, n, &dev->read_list, node) {
1014 		list_del(&node->node);
1015 		kfree(node);
1016 	}
1017 
1018 	list_for_each_entry_safe(node, n, &dev->pending_list, node) {
1019 		list_del(&node->node);
1020 		kfree(node);
1021 	}
1022 
1023 	spin_unlock(&dev->iotlb_lock);
1024 }
1025 EXPORT_SYMBOL_GPL(vhost_clear_msg);
1026 
vhost_dev_cleanup(struct vhost_dev * dev)1027 void vhost_dev_cleanup(struct vhost_dev *dev)
1028 {
1029 	int i;
1030 
1031 	for (i = 0; i < dev->nvqs; ++i) {
1032 		if (dev->vqs[i]->error_ctx)
1033 			eventfd_ctx_put(dev->vqs[i]->error_ctx);
1034 		if (dev->vqs[i]->kick)
1035 			fput(dev->vqs[i]->kick);
1036 		if (dev->vqs[i]->call_ctx.ctx)
1037 			eventfd_ctx_put(dev->vqs[i]->call_ctx.ctx);
1038 		vhost_vq_reset(dev, dev->vqs[i]);
1039 	}
1040 	vhost_dev_free_iovecs(dev);
1041 	if (dev->log_ctx)
1042 		eventfd_ctx_put(dev->log_ctx);
1043 	dev->log_ctx = NULL;
1044 	/* No one will access memory at this point */
1045 	vhost_iotlb_free(dev->umem);
1046 	dev->umem = NULL;
1047 	vhost_iotlb_free(dev->iotlb);
1048 	dev->iotlb = NULL;
1049 	vhost_clear_msg(dev);
1050 	wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
1051 	vhost_workers_free(dev);
1052 	vhost_detach_mm(dev);
1053 }
1054 EXPORT_SYMBOL_GPL(vhost_dev_cleanup);
1055 
log_access_ok(void __user * log_base,u64 addr,unsigned long sz)1056 static bool log_access_ok(void __user *log_base, u64 addr, unsigned long sz)
1057 {
1058 	u64 a = addr / VHOST_PAGE_SIZE / 8;
1059 
1060 	/* Make sure 64 bit math will not overflow. */
1061 	if (a > ULONG_MAX - (unsigned long)log_base ||
1062 	    a + (unsigned long)log_base > ULONG_MAX)
1063 		return false;
1064 
1065 	return access_ok(log_base + a,
1066 			 (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8);
1067 }
1068 
1069 /* Make sure 64 bit math will not overflow. */
vhost_overflow(u64 uaddr,u64 size)1070 static bool vhost_overflow(u64 uaddr, u64 size)
1071 {
1072 	if (uaddr > ULONG_MAX || size > ULONG_MAX)
1073 		return true;
1074 
1075 	if (!size)
1076 		return false;
1077 
1078 	return uaddr > ULONG_MAX - size + 1;
1079 }
1080 
1081 /* Caller should have vq mutex and device mutex. */
vq_memory_access_ok(void __user * log_base,struct vhost_iotlb * umem,int log_all)1082 static bool vq_memory_access_ok(void __user *log_base, struct vhost_iotlb *umem,
1083 				int log_all)
1084 {
1085 	struct vhost_iotlb_map *map;
1086 
1087 	if (!umem)
1088 		return false;
1089 
1090 	list_for_each_entry(map, &umem->list, link) {
1091 		unsigned long a = map->addr;
1092 
1093 		if (vhost_overflow(map->addr, map->size))
1094 			return false;
1095 
1096 
1097 		if (!access_ok((void __user *)a, map->size))
1098 			return false;
1099 		else if (log_all && !log_access_ok(log_base,
1100 						   map->start,
1101 						   map->size))
1102 			return false;
1103 	}
1104 	return true;
1105 }
1106 
vhost_vq_meta_fetch(struct vhost_virtqueue * vq,u64 addr,unsigned int size,int type)1107 static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq,
1108 					       u64 addr, unsigned int size,
1109 					       int type)
1110 {
1111 	const struct vhost_iotlb_map *map = vq->meta_iotlb[type];
1112 
1113 	if (!map)
1114 		return NULL;
1115 
1116 	return (void __user *)(uintptr_t)(map->addr + addr - map->start);
1117 }
1118 
1119 /* Can we switch to this memory table? */
1120 /* Caller should have device mutex but not vq mutex */
memory_access_ok(struct vhost_dev * d,struct vhost_iotlb * umem,int log_all)1121 static bool memory_access_ok(struct vhost_dev *d, struct vhost_iotlb *umem,
1122 			     int log_all)
1123 {
1124 	int i;
1125 
1126 	for (i = 0; i < d->nvqs; ++i) {
1127 		bool ok;
1128 		bool log;
1129 
1130 		mutex_lock(&d->vqs[i]->mutex);
1131 		log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL);
1132 		/* If ring is inactive, will check when it's enabled. */
1133 		if (d->vqs[i]->private_data)
1134 			ok = vq_memory_access_ok(d->vqs[i]->log_base,
1135 						 umem, log);
1136 		else
1137 			ok = true;
1138 		mutex_unlock(&d->vqs[i]->mutex);
1139 		if (!ok)
1140 			return false;
1141 	}
1142 	return true;
1143 }
1144 
1145 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
1146 			  struct iovec iov[], int iov_size, int access);
1147 
vhost_copy_to_user(struct vhost_virtqueue * vq,void __user * to,const void * from,unsigned size)1148 static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to,
1149 			      const void *from, unsigned size)
1150 {
1151 	int ret;
1152 
1153 	if (!vq->iotlb)
1154 		return __copy_to_user(to, from, size);
1155 	else {
1156 		/* This function should be called after iotlb
1157 		 * prefetch, which means we're sure that all vq
1158 		 * could be access through iotlb. So -EAGAIN should
1159 		 * not happen in this case.
1160 		 */
1161 		struct iov_iter t;
1162 		void __user *uaddr = vhost_vq_meta_fetch(vq,
1163 				     (u64)(uintptr_t)to, size,
1164 				     VHOST_ADDR_USED);
1165 
1166 		if (uaddr)
1167 			return __copy_to_user(uaddr, from, size);
1168 
1169 		ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov,
1170 				     ARRAY_SIZE(vq->iotlb_iov),
1171 				     VHOST_ACCESS_WO);
1172 		if (ret < 0)
1173 			goto out;
1174 		iov_iter_init(&t, ITER_DEST, vq->iotlb_iov, ret, size);
1175 		ret = copy_to_iter(from, size, &t);
1176 		if (ret == size)
1177 			ret = 0;
1178 	}
1179 out:
1180 	return ret;
1181 }
1182 
vhost_copy_from_user(struct vhost_virtqueue * vq,void * to,void __user * from,unsigned size)1183 static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to,
1184 				void __user *from, unsigned size)
1185 {
1186 	int ret;
1187 
1188 	if (!vq->iotlb)
1189 		return __copy_from_user(to, from, size);
1190 	else {
1191 		/* This function should be called after iotlb
1192 		 * prefetch, which means we're sure that vq
1193 		 * could be access through iotlb. So -EAGAIN should
1194 		 * not happen in this case.
1195 		 */
1196 		void __user *uaddr = vhost_vq_meta_fetch(vq,
1197 				     (u64)(uintptr_t)from, size,
1198 				     VHOST_ADDR_DESC);
1199 		struct iov_iter f;
1200 
1201 		if (uaddr)
1202 			return __copy_from_user(to, uaddr, size);
1203 
1204 		ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov,
1205 				     ARRAY_SIZE(vq->iotlb_iov),
1206 				     VHOST_ACCESS_RO);
1207 		if (ret < 0) {
1208 			vq_err(vq, "IOTLB translation failure: uaddr "
1209 			       "%p size 0x%llx\n", from,
1210 			       (unsigned long long) size);
1211 			goto out;
1212 		}
1213 		iov_iter_init(&f, ITER_SOURCE, vq->iotlb_iov, ret, size);
1214 		ret = copy_from_iter(to, size, &f);
1215 		if (ret == size)
1216 			ret = 0;
1217 	}
1218 
1219 out:
1220 	return ret;
1221 }
1222 
__vhost_get_user_slow(struct vhost_virtqueue * vq,void __user * addr,unsigned int size,int type)1223 static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq,
1224 					  void __user *addr, unsigned int size,
1225 					  int type)
1226 {
1227 	int ret;
1228 
1229 	ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov,
1230 			     ARRAY_SIZE(vq->iotlb_iov),
1231 			     VHOST_ACCESS_RO);
1232 	if (ret < 0) {
1233 		vq_err(vq, "IOTLB translation failure: uaddr "
1234 			"%p size 0x%llx\n", addr,
1235 			(unsigned long long) size);
1236 		return NULL;
1237 	}
1238 
1239 	if (ret != 1 || vq->iotlb_iov[0].iov_len != size) {
1240 		vq_err(vq, "Non atomic userspace memory access: uaddr "
1241 			"%p size 0x%llx\n", addr,
1242 			(unsigned long long) size);
1243 		return NULL;
1244 	}
1245 
1246 	return vq->iotlb_iov[0].iov_base;
1247 }
1248 
1249 /* This function should be called after iotlb
1250  * prefetch, which means we're sure that vq
1251  * could be access through iotlb. So -EAGAIN should
1252  * not happen in this case.
1253  */
__vhost_get_user(struct vhost_virtqueue * vq,void __user * addr,unsigned int size,int type)1254 static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq,
1255 					    void __user *addr, unsigned int size,
1256 					    int type)
1257 {
1258 	void __user *uaddr = vhost_vq_meta_fetch(vq,
1259 			     (u64)(uintptr_t)addr, size, type);
1260 	if (uaddr)
1261 		return uaddr;
1262 
1263 	return __vhost_get_user_slow(vq, addr, size, type);
1264 }
1265 
1266 #define vhost_put_user(vq, x, ptr)		\
1267 ({ \
1268 	int ret; \
1269 	if (!vq->iotlb) { \
1270 		ret = __put_user(x, ptr); \
1271 	} else { \
1272 		__typeof__(ptr) to = \
1273 			(__typeof__(ptr)) __vhost_get_user(vq, ptr,	\
1274 					  sizeof(*ptr), VHOST_ADDR_USED); \
1275 		if (to != NULL) \
1276 			ret = __put_user(x, to); \
1277 		else \
1278 			ret = -EFAULT;	\
1279 	} \
1280 	ret; \
1281 })
1282 
vhost_put_avail_event(struct vhost_virtqueue * vq)1283 static inline int vhost_put_avail_event(struct vhost_virtqueue *vq)
1284 {
1285 	return vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx),
1286 			      vhost_avail_event(vq));
1287 }
1288 
vhost_put_used(struct vhost_virtqueue * vq,struct vring_used_elem * head,int idx,int count)1289 static inline int vhost_put_used(struct vhost_virtqueue *vq,
1290 				 struct vring_used_elem *head, int idx,
1291 				 int count)
1292 {
1293 	return vhost_copy_to_user(vq, vq->used->ring + idx, head,
1294 				  count * sizeof(*head));
1295 }
1296 
vhost_put_used_flags(struct vhost_virtqueue * vq)1297 static inline int vhost_put_used_flags(struct vhost_virtqueue *vq)
1298 
1299 {
1300 	return vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags),
1301 			      &vq->used->flags);
1302 }
1303 
vhost_put_used_idx(struct vhost_virtqueue * vq)1304 static inline int vhost_put_used_idx(struct vhost_virtqueue *vq)
1305 
1306 {
1307 	return vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx),
1308 			      &vq->used->idx);
1309 }
1310 
1311 #define vhost_get_user(vq, x, ptr, type)		\
1312 ({ \
1313 	int ret; \
1314 	if (!vq->iotlb) { \
1315 		ret = __get_user(x, ptr); \
1316 	} else { \
1317 		__typeof__(ptr) from = \
1318 			(__typeof__(ptr)) __vhost_get_user(vq, ptr, \
1319 							   sizeof(*ptr), \
1320 							   type); \
1321 		if (from != NULL) \
1322 			ret = __get_user(x, from); \
1323 		else \
1324 			ret = -EFAULT; \
1325 	} \
1326 	ret; \
1327 })
1328 
1329 #define vhost_get_avail(vq, x, ptr) \
1330 	vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL)
1331 
1332 #define vhost_get_used(vq, x, ptr) \
1333 	vhost_get_user(vq, x, ptr, VHOST_ADDR_USED)
1334 
vhost_dev_lock_vqs(struct vhost_dev * d)1335 static void vhost_dev_lock_vqs(struct vhost_dev *d)
1336 {
1337 	int i = 0;
1338 	for (i = 0; i < d->nvqs; ++i)
1339 		mutex_lock_nested(&d->vqs[i]->mutex, i);
1340 }
1341 
vhost_dev_unlock_vqs(struct vhost_dev * d)1342 static void vhost_dev_unlock_vqs(struct vhost_dev *d)
1343 {
1344 	int i = 0;
1345 	for (i = 0; i < d->nvqs; ++i)
1346 		mutex_unlock(&d->vqs[i]->mutex);
1347 }
1348 
vhost_get_avail_idx(struct vhost_virtqueue * vq)1349 static inline int vhost_get_avail_idx(struct vhost_virtqueue *vq)
1350 {
1351 	__virtio16 idx;
1352 	int r;
1353 
1354 	r = vhost_get_avail(vq, idx, &vq->avail->idx);
1355 	if (unlikely(r < 0)) {
1356 		vq_err(vq, "Failed to access available index at %p (%d)\n",
1357 		       &vq->avail->idx, r);
1358 		return r;
1359 	}
1360 
1361 	/* Check it isn't doing very strange thing with available indexes */
1362 	vq->avail_idx = vhost16_to_cpu(vq, idx);
1363 	if (unlikely((u16)(vq->avail_idx - vq->last_avail_idx) > vq->num)) {
1364 		vq_err(vq, "Invalid available index change from %u to %u",
1365 		       vq->last_avail_idx, vq->avail_idx);
1366 		return -EINVAL;
1367 	}
1368 
1369 	/* We're done if there is nothing new */
1370 	if (vq->avail_idx == vq->last_avail_idx)
1371 		return 0;
1372 
1373 	/*
1374 	 * We updated vq->avail_idx so we need a memory barrier between
1375 	 * the index read above and the caller reading avail ring entries.
1376 	 */
1377 	smp_rmb();
1378 	return 1;
1379 }
1380 
vhost_get_avail_head(struct vhost_virtqueue * vq,__virtio16 * head,int idx)1381 static inline int vhost_get_avail_head(struct vhost_virtqueue *vq,
1382 				       __virtio16 *head, int idx)
1383 {
1384 	return vhost_get_avail(vq, *head,
1385 			       &vq->avail->ring[idx & (vq->num - 1)]);
1386 }
1387 
vhost_get_avail_flags(struct vhost_virtqueue * vq,__virtio16 * flags)1388 static inline int vhost_get_avail_flags(struct vhost_virtqueue *vq,
1389 					__virtio16 *flags)
1390 {
1391 	return vhost_get_avail(vq, *flags, &vq->avail->flags);
1392 }
1393 
vhost_get_used_event(struct vhost_virtqueue * vq,__virtio16 * event)1394 static inline int vhost_get_used_event(struct vhost_virtqueue *vq,
1395 				       __virtio16 *event)
1396 {
1397 	return vhost_get_avail(vq, *event, vhost_used_event(vq));
1398 }
1399 
vhost_get_used_idx(struct vhost_virtqueue * vq,__virtio16 * idx)1400 static inline int vhost_get_used_idx(struct vhost_virtqueue *vq,
1401 				     __virtio16 *idx)
1402 {
1403 	return vhost_get_used(vq, *idx, &vq->used->idx);
1404 }
1405 
vhost_get_desc(struct vhost_virtqueue * vq,struct vring_desc * desc,int idx)1406 static inline int vhost_get_desc(struct vhost_virtqueue *vq,
1407 				 struct vring_desc *desc, int idx)
1408 {
1409 	return vhost_copy_from_user(vq, desc, vq->desc + idx, sizeof(*desc));
1410 }
1411 
vhost_iotlb_notify_vq(struct vhost_dev * d,struct vhost_iotlb_msg * msg)1412 static void vhost_iotlb_notify_vq(struct vhost_dev *d,
1413 				  struct vhost_iotlb_msg *msg)
1414 {
1415 	struct vhost_msg_node *node, *n;
1416 
1417 	spin_lock(&d->iotlb_lock);
1418 
1419 	list_for_each_entry_safe(node, n, &d->pending_list, node) {
1420 		struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb;
1421 		if (msg->iova <= vq_msg->iova &&
1422 		    msg->iova + msg->size - 1 >= vq_msg->iova &&
1423 		    vq_msg->type == VHOST_IOTLB_MISS) {
1424 			vhost_poll_queue(&node->vq->poll);
1425 			list_del(&node->node);
1426 			kfree(node);
1427 		}
1428 	}
1429 
1430 	spin_unlock(&d->iotlb_lock);
1431 }
1432 
umem_access_ok(u64 uaddr,u64 size,int access)1433 static bool umem_access_ok(u64 uaddr, u64 size, int access)
1434 {
1435 	unsigned long a = uaddr;
1436 
1437 	/* Make sure 64 bit math will not overflow. */
1438 	if (vhost_overflow(uaddr, size))
1439 		return false;
1440 
1441 	if ((access & VHOST_ACCESS_RO) &&
1442 	    !access_ok((void __user *)a, size))
1443 		return false;
1444 	if ((access & VHOST_ACCESS_WO) &&
1445 	    !access_ok((void __user *)a, size))
1446 		return false;
1447 	return true;
1448 }
1449 
vhost_process_iotlb_msg(struct vhost_dev * dev,u32 asid,struct vhost_iotlb_msg * msg)1450 static int vhost_process_iotlb_msg(struct vhost_dev *dev, u32 asid,
1451 				   struct vhost_iotlb_msg *msg)
1452 {
1453 	int ret = 0;
1454 
1455 	if (asid != 0)
1456 		return -EINVAL;
1457 
1458 	mutex_lock(&dev->mutex);
1459 	vhost_dev_lock_vqs(dev);
1460 	switch (msg->type) {
1461 	case VHOST_IOTLB_UPDATE:
1462 		if (!dev->iotlb) {
1463 			ret = -EFAULT;
1464 			break;
1465 		}
1466 		if (!umem_access_ok(msg->uaddr, msg->size, msg->perm)) {
1467 			ret = -EFAULT;
1468 			break;
1469 		}
1470 		vhost_vq_meta_reset(dev);
1471 		if (vhost_iotlb_add_range(dev->iotlb, msg->iova,
1472 					  msg->iova + msg->size - 1,
1473 					  msg->uaddr, msg->perm)) {
1474 			ret = -ENOMEM;
1475 			break;
1476 		}
1477 		vhost_iotlb_notify_vq(dev, msg);
1478 		break;
1479 	case VHOST_IOTLB_INVALIDATE:
1480 		if (!dev->iotlb) {
1481 			ret = -EFAULT;
1482 			break;
1483 		}
1484 		vhost_vq_meta_reset(dev);
1485 		vhost_iotlb_del_range(dev->iotlb, msg->iova,
1486 				      msg->iova + msg->size - 1);
1487 		break;
1488 	default:
1489 		ret = -EINVAL;
1490 		break;
1491 	}
1492 
1493 	vhost_dev_unlock_vqs(dev);
1494 	mutex_unlock(&dev->mutex);
1495 
1496 	return ret;
1497 }
vhost_chr_write_iter(struct vhost_dev * dev,struct iov_iter * from)1498 ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
1499 			     struct iov_iter *from)
1500 {
1501 	struct vhost_iotlb_msg msg;
1502 	size_t offset;
1503 	int type, ret;
1504 	u32 asid = 0;
1505 
1506 	ret = copy_from_iter(&type, sizeof(type), from);
1507 	if (ret != sizeof(type)) {
1508 		ret = -EINVAL;
1509 		goto done;
1510 	}
1511 
1512 	switch (type) {
1513 	case VHOST_IOTLB_MSG:
1514 		/* There maybe a hole after type for V1 message type,
1515 		 * so skip it here.
1516 		 */
1517 		offset = offsetof(struct vhost_msg, iotlb) - sizeof(int);
1518 		break;
1519 	case VHOST_IOTLB_MSG_V2:
1520 		if (vhost_backend_has_feature(dev->vqs[0],
1521 					      VHOST_BACKEND_F_IOTLB_ASID)) {
1522 			ret = copy_from_iter(&asid, sizeof(asid), from);
1523 			if (ret != sizeof(asid)) {
1524 				ret = -EINVAL;
1525 				goto done;
1526 			}
1527 			offset = 0;
1528 		} else
1529 			offset = sizeof(__u32);
1530 		break;
1531 	default:
1532 		ret = -EINVAL;
1533 		goto done;
1534 	}
1535 
1536 	iov_iter_advance(from, offset);
1537 	ret = copy_from_iter(&msg, sizeof(msg), from);
1538 	if (ret != sizeof(msg)) {
1539 		ret = -EINVAL;
1540 		goto done;
1541 	}
1542 
1543 	if (msg.type == VHOST_IOTLB_UPDATE && msg.size == 0) {
1544 		ret = -EINVAL;
1545 		goto done;
1546 	}
1547 
1548 	if (dev->msg_handler)
1549 		ret = dev->msg_handler(dev, asid, &msg);
1550 	else
1551 		ret = vhost_process_iotlb_msg(dev, asid, &msg);
1552 	if (ret) {
1553 		ret = -EFAULT;
1554 		goto done;
1555 	}
1556 
1557 	ret = (type == VHOST_IOTLB_MSG) ? sizeof(struct vhost_msg) :
1558 	      sizeof(struct vhost_msg_v2);
1559 done:
1560 	return ret;
1561 }
1562 EXPORT_SYMBOL(vhost_chr_write_iter);
1563 
vhost_chr_poll(struct file * file,struct vhost_dev * dev,poll_table * wait)1564 __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
1565 			    poll_table *wait)
1566 {
1567 	__poll_t mask = 0;
1568 
1569 	poll_wait(file, &dev->wait, wait);
1570 
1571 	if (!list_empty(&dev->read_list))
1572 		mask |= EPOLLIN | EPOLLRDNORM;
1573 
1574 	return mask;
1575 }
1576 EXPORT_SYMBOL(vhost_chr_poll);
1577 
vhost_chr_read_iter(struct vhost_dev * dev,struct iov_iter * to,int noblock)1578 ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
1579 			    int noblock)
1580 {
1581 	DEFINE_WAIT(wait);
1582 	struct vhost_msg_node *node;
1583 	ssize_t ret = 0;
1584 	unsigned size = sizeof(struct vhost_msg);
1585 
1586 	if (iov_iter_count(to) < size)
1587 		return 0;
1588 
1589 	while (1) {
1590 		if (!noblock)
1591 			prepare_to_wait(&dev->wait, &wait,
1592 					TASK_INTERRUPTIBLE);
1593 
1594 		node = vhost_dequeue_msg(dev, &dev->read_list);
1595 		if (node)
1596 			break;
1597 		if (noblock) {
1598 			ret = -EAGAIN;
1599 			break;
1600 		}
1601 		if (signal_pending(current)) {
1602 			ret = -ERESTARTSYS;
1603 			break;
1604 		}
1605 		if (!dev->iotlb) {
1606 			ret = -EBADFD;
1607 			break;
1608 		}
1609 
1610 		schedule();
1611 	}
1612 
1613 	if (!noblock)
1614 		finish_wait(&dev->wait, &wait);
1615 
1616 	if (node) {
1617 		struct vhost_iotlb_msg *msg;
1618 		void *start = &node->msg;
1619 
1620 		switch (node->msg.type) {
1621 		case VHOST_IOTLB_MSG:
1622 			size = sizeof(node->msg);
1623 			msg = &node->msg.iotlb;
1624 			break;
1625 		case VHOST_IOTLB_MSG_V2:
1626 			size = sizeof(node->msg_v2);
1627 			msg = &node->msg_v2.iotlb;
1628 			break;
1629 		default:
1630 			BUG();
1631 			break;
1632 		}
1633 
1634 		ret = copy_to_iter(start, size, to);
1635 		if (ret != size || msg->type != VHOST_IOTLB_MISS) {
1636 			kfree(node);
1637 			return ret;
1638 		}
1639 		vhost_enqueue_msg(dev, &dev->pending_list, node);
1640 	}
1641 
1642 	return ret;
1643 }
1644 EXPORT_SYMBOL_GPL(vhost_chr_read_iter);
1645 
vhost_iotlb_miss(struct vhost_virtqueue * vq,u64 iova,int access)1646 static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access)
1647 {
1648 	struct vhost_dev *dev = vq->dev;
1649 	struct vhost_msg_node *node;
1650 	struct vhost_iotlb_msg *msg;
1651 	bool v2 = vhost_backend_has_feature(vq, VHOST_BACKEND_F_IOTLB_MSG_V2);
1652 
1653 	node = vhost_new_msg(vq, v2 ? VHOST_IOTLB_MSG_V2 : VHOST_IOTLB_MSG);
1654 	if (!node)
1655 		return -ENOMEM;
1656 
1657 	if (v2) {
1658 		node->msg_v2.type = VHOST_IOTLB_MSG_V2;
1659 		msg = &node->msg_v2.iotlb;
1660 	} else {
1661 		msg = &node->msg.iotlb;
1662 	}
1663 
1664 	msg->type = VHOST_IOTLB_MISS;
1665 	msg->iova = iova;
1666 	msg->perm = access;
1667 
1668 	vhost_enqueue_msg(dev, &dev->read_list, node);
1669 
1670 	return 0;
1671 }
1672 
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)1673 static bool vq_access_ok(struct vhost_virtqueue *vq, unsigned int num,
1674 			 vring_desc_t __user *desc,
1675 			 vring_avail_t __user *avail,
1676 			 vring_used_t __user *used)
1677 
1678 {
1679 	/* If an IOTLB device is present, the vring addresses are
1680 	 * GIOVAs. Access validation occurs at prefetch time. */
1681 	if (vq->iotlb)
1682 		return true;
1683 
1684 	return access_ok(desc, vhost_get_desc_size(vq, num)) &&
1685 	       access_ok(avail, vhost_get_avail_size(vq, num)) &&
1686 	       access_ok(used, vhost_get_used_size(vq, num));
1687 }
1688 
vhost_vq_meta_update(struct vhost_virtqueue * vq,const struct vhost_iotlb_map * map,int type)1689 static void vhost_vq_meta_update(struct vhost_virtqueue *vq,
1690 				 const struct vhost_iotlb_map *map,
1691 				 int type)
1692 {
1693 	int access = (type == VHOST_ADDR_USED) ?
1694 		     VHOST_ACCESS_WO : VHOST_ACCESS_RO;
1695 
1696 	if (likely(map->perm & access))
1697 		vq->meta_iotlb[type] = map;
1698 }
1699 
iotlb_access_ok(struct vhost_virtqueue * vq,int access,u64 addr,u64 len,int type)1700 static bool iotlb_access_ok(struct vhost_virtqueue *vq,
1701 			    int access, u64 addr, u64 len, int type)
1702 {
1703 	const struct vhost_iotlb_map *map;
1704 	struct vhost_iotlb *umem = vq->iotlb;
1705 	u64 s = 0, size, orig_addr = addr, last = addr + len - 1;
1706 
1707 	if (vhost_vq_meta_fetch(vq, addr, len, type))
1708 		return true;
1709 
1710 	while (len > s) {
1711 		map = vhost_iotlb_itree_first(umem, addr, last);
1712 		if (map == NULL || map->start > addr) {
1713 			vhost_iotlb_miss(vq, addr, access);
1714 			return false;
1715 		} else if (!(map->perm & access)) {
1716 			/* Report the possible access violation by
1717 			 * request another translation from userspace.
1718 			 */
1719 			return false;
1720 		}
1721 
1722 		size = map->size - addr + map->start;
1723 
1724 		if (orig_addr == addr && size >= len)
1725 			vhost_vq_meta_update(vq, map, type);
1726 
1727 		s += size;
1728 		addr += size;
1729 	}
1730 
1731 	return true;
1732 }
1733 
vq_meta_prefetch(struct vhost_virtqueue * vq)1734 int vq_meta_prefetch(struct vhost_virtqueue *vq)
1735 {
1736 	unsigned int num = vq->num;
1737 
1738 	if (!vq->iotlb)
1739 		return 1;
1740 
1741 	return iotlb_access_ok(vq, VHOST_MAP_RO, (u64)(uintptr_t)vq->desc,
1742 			       vhost_get_desc_size(vq, num), VHOST_ADDR_DESC) &&
1743 	       iotlb_access_ok(vq, VHOST_MAP_RO, (u64)(uintptr_t)vq->avail,
1744 			       vhost_get_avail_size(vq, num),
1745 			       VHOST_ADDR_AVAIL) &&
1746 	       iotlb_access_ok(vq, VHOST_MAP_WO, (u64)(uintptr_t)vq->used,
1747 			       vhost_get_used_size(vq, num), VHOST_ADDR_USED);
1748 }
1749 EXPORT_SYMBOL_GPL(vq_meta_prefetch);
1750 
1751 /* Can we log writes? */
1752 /* Caller should have device mutex but not vq mutex */
vhost_log_access_ok(struct vhost_dev * dev)1753 bool vhost_log_access_ok(struct vhost_dev *dev)
1754 {
1755 	return memory_access_ok(dev, dev->umem, 1);
1756 }
1757 EXPORT_SYMBOL_GPL(vhost_log_access_ok);
1758 
vq_log_used_access_ok(struct vhost_virtqueue * vq,void __user * log_base,bool log_used,u64 log_addr)1759 static bool vq_log_used_access_ok(struct vhost_virtqueue *vq,
1760 				  void __user *log_base,
1761 				  bool log_used,
1762 				  u64 log_addr)
1763 {
1764 	/* If an IOTLB device is present, log_addr is a GIOVA that
1765 	 * will never be logged by log_used(). */
1766 	if (vq->iotlb)
1767 		return true;
1768 
1769 	return !log_used || log_access_ok(log_base, log_addr,
1770 					  vhost_get_used_size(vq, vq->num));
1771 }
1772 
1773 /* Verify access for write logging. */
1774 /* Caller should have vq mutex and device mutex */
vq_log_access_ok(struct vhost_virtqueue * vq,void __user * log_base)1775 static bool vq_log_access_ok(struct vhost_virtqueue *vq,
1776 			     void __user *log_base)
1777 {
1778 	return vq_memory_access_ok(log_base, vq->umem,
1779 				   vhost_has_feature(vq, VHOST_F_LOG_ALL)) &&
1780 		vq_log_used_access_ok(vq, log_base, vq->log_used, vq->log_addr);
1781 }
1782 
1783 /* Can we start vq? */
1784 /* Caller should have vq mutex and device mutex */
vhost_vq_access_ok(struct vhost_virtqueue * vq)1785 bool vhost_vq_access_ok(struct vhost_virtqueue *vq)
1786 {
1787 	if (!vq_log_access_ok(vq, vq->log_base))
1788 		return false;
1789 
1790 	return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used);
1791 }
1792 EXPORT_SYMBOL_GPL(vhost_vq_access_ok);
1793 
vhost_set_memory(struct vhost_dev * d,struct vhost_memory __user * m)1794 static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)
1795 {
1796 	struct vhost_memory mem, *newmem;
1797 	struct vhost_memory_region *region;
1798 	struct vhost_iotlb *newumem, *oldumem;
1799 	unsigned long size = offsetof(struct vhost_memory, regions);
1800 	int i;
1801 
1802 	if (copy_from_user(&mem, m, size))
1803 		return -EFAULT;
1804 	if (mem.padding)
1805 		return -EOPNOTSUPP;
1806 	if (mem.nregions > max_mem_regions)
1807 		return -E2BIG;
1808 	newmem = kvzalloc(struct_size(newmem, regions, mem.nregions),
1809 			GFP_KERNEL);
1810 	if (!newmem)
1811 		return -ENOMEM;
1812 
1813 	memcpy(newmem, &mem, size);
1814 	if (copy_from_user(newmem->regions, m->regions,
1815 			   flex_array_size(newmem, regions, mem.nregions))) {
1816 		kvfree(newmem);
1817 		return -EFAULT;
1818 	}
1819 
1820 	newumem = iotlb_alloc();
1821 	if (!newumem) {
1822 		kvfree(newmem);
1823 		return -ENOMEM;
1824 	}
1825 
1826 	for (region = newmem->regions;
1827 	     region < newmem->regions + mem.nregions;
1828 	     region++) {
1829 		if (vhost_iotlb_add_range(newumem,
1830 					  region->guest_phys_addr,
1831 					  region->guest_phys_addr +
1832 					  region->memory_size - 1,
1833 					  region->userspace_addr,
1834 					  VHOST_MAP_RW))
1835 			goto err;
1836 	}
1837 
1838 	if (!memory_access_ok(d, newumem, 0))
1839 		goto err;
1840 
1841 	oldumem = d->umem;
1842 	d->umem = newumem;
1843 
1844 	/* All memory accesses are done under some VQ mutex. */
1845 	for (i = 0; i < d->nvqs; ++i) {
1846 		mutex_lock(&d->vqs[i]->mutex);
1847 		d->vqs[i]->umem = newumem;
1848 		mutex_unlock(&d->vqs[i]->mutex);
1849 	}
1850 
1851 	kvfree(newmem);
1852 	vhost_iotlb_free(oldumem);
1853 	return 0;
1854 
1855 err:
1856 	vhost_iotlb_free(newumem);
1857 	kvfree(newmem);
1858 	return -EFAULT;
1859 }
1860 
vhost_vring_set_num(struct vhost_dev * d,struct vhost_virtqueue * vq,void __user * argp)1861 static long vhost_vring_set_num(struct vhost_dev *d,
1862 				struct vhost_virtqueue *vq,
1863 				void __user *argp)
1864 {
1865 	struct vhost_vring_state s;
1866 
1867 	/* Resizing ring with an active backend?
1868 	 * You don't want to do that. */
1869 	if (vq->private_data)
1870 		return -EBUSY;
1871 
1872 	if (copy_from_user(&s, argp, sizeof s))
1873 		return -EFAULT;
1874 
1875 	if (!s.num || s.num > 0xffff || (s.num & (s.num - 1)))
1876 		return -EINVAL;
1877 	vq->num = s.num;
1878 
1879 	return 0;
1880 }
1881 
vhost_vring_set_addr(struct vhost_dev * d,struct vhost_virtqueue * vq,void __user * argp)1882 static long vhost_vring_set_addr(struct vhost_dev *d,
1883 				 struct vhost_virtqueue *vq,
1884 				 void __user *argp)
1885 {
1886 	struct vhost_vring_addr a;
1887 
1888 	if (copy_from_user(&a, argp, sizeof a))
1889 		return -EFAULT;
1890 	if (a.flags & ~(0x1 << VHOST_VRING_F_LOG))
1891 		return -EOPNOTSUPP;
1892 
1893 	/* For 32bit, verify that the top 32bits of the user
1894 	   data are set to zero. */
1895 	if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr ||
1896 	    (u64)(unsigned long)a.used_user_addr != a.used_user_addr ||
1897 	    (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr)
1898 		return -EFAULT;
1899 
1900 	/* Make sure it's safe to cast pointers to vring types. */
1901 	BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE);
1902 	BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE);
1903 	if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) ||
1904 	    (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) ||
1905 	    (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1)))
1906 		return -EINVAL;
1907 
1908 	/* We only verify access here if backend is configured.
1909 	 * If it is not, we don't as size might not have been setup.
1910 	 * We will verify when backend is configured. */
1911 	if (vq->private_data) {
1912 		if (!vq_access_ok(vq, vq->num,
1913 			(void __user *)(unsigned long)a.desc_user_addr,
1914 			(void __user *)(unsigned long)a.avail_user_addr,
1915 			(void __user *)(unsigned long)a.used_user_addr))
1916 			return -EINVAL;
1917 
1918 		/* Also validate log access for used ring if enabled. */
1919 		if (!vq_log_used_access_ok(vq, vq->log_base,
1920 				a.flags & (0x1 << VHOST_VRING_F_LOG),
1921 				a.log_guest_addr))
1922 			return -EINVAL;
1923 	}
1924 
1925 	vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG));
1926 	vq->desc = (void __user *)(unsigned long)a.desc_user_addr;
1927 	vq->avail = (void __user *)(unsigned long)a.avail_user_addr;
1928 	vq->log_addr = a.log_guest_addr;
1929 	vq->used = (void __user *)(unsigned long)a.used_user_addr;
1930 
1931 	return 0;
1932 }
1933 
vhost_vring_set_num_addr(struct vhost_dev * d,struct vhost_virtqueue * vq,unsigned int ioctl,void __user * argp)1934 static long vhost_vring_set_num_addr(struct vhost_dev *d,
1935 				     struct vhost_virtqueue *vq,
1936 				     unsigned int ioctl,
1937 				     void __user *argp)
1938 {
1939 	long r;
1940 
1941 	mutex_lock(&vq->mutex);
1942 
1943 	switch (ioctl) {
1944 	case VHOST_SET_VRING_NUM:
1945 		r = vhost_vring_set_num(d, vq, argp);
1946 		break;
1947 	case VHOST_SET_VRING_ADDR:
1948 		r = vhost_vring_set_addr(d, vq, argp);
1949 		break;
1950 	default:
1951 		BUG();
1952 	}
1953 
1954 	mutex_unlock(&vq->mutex);
1955 
1956 	return r;
1957 }
vhost_vring_ioctl(struct vhost_dev * d,unsigned int ioctl,void __user * argp)1958 long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
1959 {
1960 	struct file *eventfp, *filep = NULL;
1961 	bool pollstart = false, pollstop = false;
1962 	struct eventfd_ctx *ctx = NULL;
1963 	struct vhost_virtqueue *vq;
1964 	struct vhost_vring_state s;
1965 	struct vhost_vring_file f;
1966 	u32 idx;
1967 	long r;
1968 
1969 	r = vhost_get_vq_from_user(d, argp, &vq, &idx);
1970 	if (r < 0)
1971 		return r;
1972 
1973 	if (ioctl == VHOST_SET_VRING_NUM ||
1974 	    ioctl == VHOST_SET_VRING_ADDR) {
1975 		return vhost_vring_set_num_addr(d, vq, ioctl, argp);
1976 	}
1977 
1978 	mutex_lock(&vq->mutex);
1979 
1980 	switch (ioctl) {
1981 	case VHOST_SET_VRING_BASE:
1982 		/* Moving base with an active backend?
1983 		 * You don't want to do that. */
1984 		if (vq->private_data) {
1985 			r = -EBUSY;
1986 			break;
1987 		}
1988 		if (copy_from_user(&s, argp, sizeof s)) {
1989 			r = -EFAULT;
1990 			break;
1991 		}
1992 		if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) {
1993 			vq->last_avail_idx = s.num & 0xffff;
1994 			vq->last_used_idx = (s.num >> 16) & 0xffff;
1995 		} else {
1996 			if (s.num > 0xffff) {
1997 				r = -EINVAL;
1998 				break;
1999 			}
2000 			vq->last_avail_idx = s.num;
2001 		}
2002 		/* Forget the cached index value. */
2003 		vq->avail_idx = vq->last_avail_idx;
2004 		break;
2005 	case VHOST_GET_VRING_BASE:
2006 		s.index = idx;
2007 		if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED))
2008 			s.num = (u32)vq->last_avail_idx | ((u32)vq->last_used_idx << 16);
2009 		else
2010 			s.num = vq->last_avail_idx;
2011 		if (copy_to_user(argp, &s, sizeof s))
2012 			r = -EFAULT;
2013 		break;
2014 	case VHOST_SET_VRING_KICK:
2015 		if (copy_from_user(&f, argp, sizeof f)) {
2016 			r = -EFAULT;
2017 			break;
2018 		}
2019 		eventfp = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_fget(f.fd);
2020 		if (IS_ERR(eventfp)) {
2021 			r = PTR_ERR(eventfp);
2022 			break;
2023 		}
2024 		if (eventfp != vq->kick) {
2025 			pollstop = (filep = vq->kick) != NULL;
2026 			pollstart = (vq->kick = eventfp) != NULL;
2027 		} else
2028 			filep = eventfp;
2029 		break;
2030 	case VHOST_SET_VRING_CALL:
2031 		if (copy_from_user(&f, argp, sizeof f)) {
2032 			r = -EFAULT;
2033 			break;
2034 		}
2035 		ctx = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(f.fd);
2036 		if (IS_ERR(ctx)) {
2037 			r = PTR_ERR(ctx);
2038 			break;
2039 		}
2040 
2041 		swap(ctx, vq->call_ctx.ctx);
2042 		break;
2043 	case VHOST_SET_VRING_ERR:
2044 		if (copy_from_user(&f, argp, sizeof f)) {
2045 			r = -EFAULT;
2046 			break;
2047 		}
2048 		ctx = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(f.fd);
2049 		if (IS_ERR(ctx)) {
2050 			r = PTR_ERR(ctx);
2051 			break;
2052 		}
2053 		swap(ctx, vq->error_ctx);
2054 		break;
2055 	case VHOST_SET_VRING_ENDIAN:
2056 		r = vhost_set_vring_endian(vq, argp);
2057 		break;
2058 	case VHOST_GET_VRING_ENDIAN:
2059 		r = vhost_get_vring_endian(vq, idx, argp);
2060 		break;
2061 	case VHOST_SET_VRING_BUSYLOOP_TIMEOUT:
2062 		if (copy_from_user(&s, argp, sizeof(s))) {
2063 			r = -EFAULT;
2064 			break;
2065 		}
2066 		vq->busyloop_timeout = s.num;
2067 		break;
2068 	case VHOST_GET_VRING_BUSYLOOP_TIMEOUT:
2069 		s.index = idx;
2070 		s.num = vq->busyloop_timeout;
2071 		if (copy_to_user(argp, &s, sizeof(s)))
2072 			r = -EFAULT;
2073 		break;
2074 	default:
2075 		r = -ENOIOCTLCMD;
2076 	}
2077 
2078 	if (pollstop && vq->handle_kick)
2079 		vhost_poll_stop(&vq->poll);
2080 
2081 	if (!IS_ERR_OR_NULL(ctx))
2082 		eventfd_ctx_put(ctx);
2083 	if (filep)
2084 		fput(filep);
2085 
2086 	if (pollstart && vq->handle_kick)
2087 		r = vhost_poll_start(&vq->poll, vq->kick);
2088 
2089 	mutex_unlock(&vq->mutex);
2090 
2091 	if (pollstop && vq->handle_kick)
2092 		vhost_dev_flush(vq->poll.dev);
2093 	return r;
2094 }
2095 EXPORT_SYMBOL_GPL(vhost_vring_ioctl);
2096 
vhost_init_device_iotlb(struct vhost_dev * d)2097 int vhost_init_device_iotlb(struct vhost_dev *d)
2098 {
2099 	struct vhost_iotlb *niotlb, *oiotlb;
2100 	int i;
2101 
2102 	niotlb = iotlb_alloc();
2103 	if (!niotlb)
2104 		return -ENOMEM;
2105 
2106 	oiotlb = d->iotlb;
2107 	d->iotlb = niotlb;
2108 
2109 	for (i = 0; i < d->nvqs; ++i) {
2110 		struct vhost_virtqueue *vq = d->vqs[i];
2111 
2112 		mutex_lock(&vq->mutex);
2113 		vq->iotlb = niotlb;
2114 		__vhost_vq_meta_reset(vq);
2115 		mutex_unlock(&vq->mutex);
2116 	}
2117 
2118 	vhost_iotlb_free(oiotlb);
2119 
2120 	return 0;
2121 }
2122 EXPORT_SYMBOL_GPL(vhost_init_device_iotlb);
2123 
2124 /* Caller must have device mutex */
vhost_dev_ioctl(struct vhost_dev * d,unsigned int ioctl,void __user * argp)2125 long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
2126 {
2127 	struct eventfd_ctx *ctx;
2128 	u64 p;
2129 	long r;
2130 	int i, fd;
2131 
2132 	/* If you are not the owner, you can become one */
2133 	if (ioctl == VHOST_SET_OWNER) {
2134 		r = vhost_dev_set_owner(d);
2135 		goto done;
2136 	}
2137 
2138 	/* You must be the owner to do anything else */
2139 	r = vhost_dev_check_owner(d);
2140 	if (r)
2141 		goto done;
2142 
2143 	switch (ioctl) {
2144 	case VHOST_SET_MEM_TABLE:
2145 		r = vhost_set_memory(d, argp);
2146 		break;
2147 	case VHOST_SET_LOG_BASE:
2148 		if (copy_from_user(&p, argp, sizeof p)) {
2149 			r = -EFAULT;
2150 			break;
2151 		}
2152 		if ((u64)(unsigned long)p != p) {
2153 			r = -EFAULT;
2154 			break;
2155 		}
2156 		for (i = 0; i < d->nvqs; ++i) {
2157 			struct vhost_virtqueue *vq;
2158 			void __user *base = (void __user *)(unsigned long)p;
2159 			vq = d->vqs[i];
2160 			mutex_lock(&vq->mutex);
2161 			/* If ring is inactive, will check when it's enabled. */
2162 			if (vq->private_data && !vq_log_access_ok(vq, base))
2163 				r = -EFAULT;
2164 			else
2165 				vq->log_base = base;
2166 			mutex_unlock(&vq->mutex);
2167 		}
2168 		break;
2169 	case VHOST_SET_LOG_FD:
2170 		r = get_user(fd, (int __user *)argp);
2171 		if (r < 0)
2172 			break;
2173 		ctx = fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(fd);
2174 		if (IS_ERR(ctx)) {
2175 			r = PTR_ERR(ctx);
2176 			break;
2177 		}
2178 		swap(ctx, d->log_ctx);
2179 		for (i = 0; i < d->nvqs; ++i) {
2180 			mutex_lock(&d->vqs[i]->mutex);
2181 			d->vqs[i]->log_ctx = d->log_ctx;
2182 			mutex_unlock(&d->vqs[i]->mutex);
2183 		}
2184 		if (ctx)
2185 			eventfd_ctx_put(ctx);
2186 		break;
2187 	default:
2188 		r = -ENOIOCTLCMD;
2189 		break;
2190 	}
2191 done:
2192 	return r;
2193 }
2194 EXPORT_SYMBOL_GPL(vhost_dev_ioctl);
2195 
2196 /* TODO: This is really inefficient.  We need something like get_user()
2197  * (instruction directly accesses the data, with an exception table entry
2198  * returning -EFAULT). See Documentation/arch/x86/exception-tables.rst.
2199  */
set_bit_to_user(int nr,void __user * addr)2200 static int set_bit_to_user(int nr, void __user *addr)
2201 {
2202 	unsigned long log = (unsigned long)addr;
2203 	struct page *page;
2204 	void *base;
2205 	int bit = nr + (log % PAGE_SIZE) * 8;
2206 	int r;
2207 
2208 	r = pin_user_pages_fast(log, 1, FOLL_WRITE, &page);
2209 	if (r < 0)
2210 		return r;
2211 	BUG_ON(r != 1);
2212 	base = kmap_atomic(page);
2213 	set_bit(bit, base);
2214 	kunmap_atomic(base);
2215 	unpin_user_pages_dirty_lock(&page, 1, true);
2216 	return 0;
2217 }
2218 
log_write(void __user * log_base,u64 write_address,u64 write_length)2219 static int log_write(void __user *log_base,
2220 		     u64 write_address, u64 write_length)
2221 {
2222 	u64 write_page = write_address / VHOST_PAGE_SIZE;
2223 	int r;
2224 
2225 	if (!write_length)
2226 		return 0;
2227 	write_length += write_address % VHOST_PAGE_SIZE;
2228 	for (;;) {
2229 		u64 base = (u64)(unsigned long)log_base;
2230 		u64 log = base + write_page / 8;
2231 		int bit = write_page % 8;
2232 		if ((u64)(unsigned long)log != log)
2233 			return -EFAULT;
2234 		r = set_bit_to_user(bit, (void __user *)(unsigned long)log);
2235 		if (r < 0)
2236 			return r;
2237 		if (write_length <= VHOST_PAGE_SIZE)
2238 			break;
2239 		write_length -= VHOST_PAGE_SIZE;
2240 		write_page += 1;
2241 	}
2242 	return r;
2243 }
2244 
log_write_hva(struct vhost_virtqueue * vq,u64 hva,u64 len)2245 static int log_write_hva(struct vhost_virtqueue *vq, u64 hva, u64 len)
2246 {
2247 	struct vhost_iotlb *umem = vq->umem;
2248 	struct vhost_iotlb_map *u;
2249 	u64 start, end, l, min;
2250 	int r;
2251 	bool hit = false;
2252 
2253 	while (len) {
2254 		min = len;
2255 		/* More than one GPAs can be mapped into a single HVA. So
2256 		 * iterate all possible umems here to be safe.
2257 		 */
2258 		list_for_each_entry(u, &umem->list, link) {
2259 			if (u->addr > hva - 1 + len ||
2260 			    u->addr - 1 + u->size < hva)
2261 				continue;
2262 			start = max(u->addr, hva);
2263 			end = min(u->addr - 1 + u->size, hva - 1 + len);
2264 			l = end - start + 1;
2265 			r = log_write(vq->log_base,
2266 				      u->start + start - u->addr,
2267 				      l);
2268 			if (r < 0)
2269 				return r;
2270 			hit = true;
2271 			min = min(l, min);
2272 		}
2273 
2274 		if (!hit)
2275 			return -EFAULT;
2276 
2277 		len -= min;
2278 		hva += min;
2279 	}
2280 
2281 	return 0;
2282 }
2283 
log_used(struct vhost_virtqueue * vq,u64 used_offset,u64 len)2284 static int log_used(struct vhost_virtqueue *vq, u64 used_offset, u64 len)
2285 {
2286 	struct iovec *iov = vq->log_iov;
2287 	int i, ret;
2288 
2289 	if (!vq->iotlb)
2290 		return log_write(vq->log_base, vq->log_addr + used_offset, len);
2291 
2292 	ret = translate_desc(vq, (uintptr_t)vq->used + used_offset,
2293 			     len, iov, 64, VHOST_ACCESS_WO);
2294 	if (ret < 0)
2295 		return ret;
2296 
2297 	for (i = 0; i < ret; i++) {
2298 		ret = log_write_hva(vq,	(uintptr_t)iov[i].iov_base,
2299 				    iov[i].iov_len);
2300 		if (ret)
2301 			return ret;
2302 	}
2303 
2304 	return 0;
2305 }
2306 
vhost_log_write(struct vhost_virtqueue * vq,struct vhost_log * log,unsigned int log_num,u64 len,struct iovec * iov,int count)2307 int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
2308 		    unsigned int log_num, u64 len, struct iovec *iov, int count)
2309 {
2310 	int i, r;
2311 
2312 	/* Make sure data written is seen before log. */
2313 	smp_wmb();
2314 
2315 	if (vq->iotlb) {
2316 		for (i = 0; i < count; i++) {
2317 			r = log_write_hva(vq, (uintptr_t)iov[i].iov_base,
2318 					  iov[i].iov_len);
2319 			if (r < 0)
2320 				return r;
2321 		}
2322 		return 0;
2323 	}
2324 
2325 	for (i = 0; i < log_num; ++i) {
2326 		u64 l = min(log[i].len, len);
2327 		r = log_write(vq->log_base, log[i].addr, l);
2328 		if (r < 0)
2329 			return r;
2330 		len -= l;
2331 		if (!len) {
2332 			if (vq->log_ctx)
2333 				eventfd_signal(vq->log_ctx);
2334 			return 0;
2335 		}
2336 	}
2337 	/* Length written exceeds what we have stored. This is a bug. */
2338 	BUG();
2339 	return 0;
2340 }
2341 EXPORT_SYMBOL_GPL(vhost_log_write);
2342 
vhost_update_used_flags(struct vhost_virtqueue * vq)2343 static int vhost_update_used_flags(struct vhost_virtqueue *vq)
2344 {
2345 	void __user *used;
2346 	if (vhost_put_used_flags(vq))
2347 		return -EFAULT;
2348 	if (unlikely(vq->log_used)) {
2349 		/* Make sure the flag is seen before log. */
2350 		smp_wmb();
2351 		/* Log used flag write. */
2352 		used = &vq->used->flags;
2353 		log_used(vq, (used - (void __user *)vq->used),
2354 			 sizeof vq->used->flags);
2355 		if (vq->log_ctx)
2356 			eventfd_signal(vq->log_ctx);
2357 	}
2358 	return 0;
2359 }
2360 
vhost_update_avail_event(struct vhost_virtqueue * vq)2361 static int vhost_update_avail_event(struct vhost_virtqueue *vq)
2362 {
2363 	if (vhost_put_avail_event(vq))
2364 		return -EFAULT;
2365 	if (unlikely(vq->log_used)) {
2366 		void __user *used;
2367 		/* Make sure the event is seen before log. */
2368 		smp_wmb();
2369 		/* Log avail event write */
2370 		used = vhost_avail_event(vq);
2371 		log_used(vq, (used - (void __user *)vq->used),
2372 			 sizeof *vhost_avail_event(vq));
2373 		if (vq->log_ctx)
2374 			eventfd_signal(vq->log_ctx);
2375 	}
2376 	return 0;
2377 }
2378 
vhost_vq_init_access(struct vhost_virtqueue * vq)2379 int vhost_vq_init_access(struct vhost_virtqueue *vq)
2380 {
2381 	__virtio16 last_used_idx;
2382 	int r;
2383 	bool is_le = vq->is_le;
2384 
2385 	if (!vq->private_data)
2386 		return 0;
2387 
2388 	vhost_init_is_le(vq);
2389 
2390 	r = vhost_update_used_flags(vq);
2391 	if (r)
2392 		goto err;
2393 	vq->signalled_used_valid = false;
2394 	if (!vq->iotlb &&
2395 	    !access_ok(&vq->used->idx, sizeof vq->used->idx)) {
2396 		r = -EFAULT;
2397 		goto err;
2398 	}
2399 	r = vhost_get_used_idx(vq, &last_used_idx);
2400 	if (r) {
2401 		vq_err(vq, "Can't access used idx at %p\n",
2402 		       &vq->used->idx);
2403 		goto err;
2404 	}
2405 	vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx);
2406 	return 0;
2407 
2408 err:
2409 	vq->is_le = is_le;
2410 	return r;
2411 }
2412 EXPORT_SYMBOL_GPL(vhost_vq_init_access);
2413 
translate_desc(struct vhost_virtqueue * vq,u64 addr,u32 len,struct iovec iov[],int iov_size,int access)2414 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
2415 			  struct iovec iov[], int iov_size, int access)
2416 {
2417 	const struct vhost_iotlb_map *map;
2418 	struct vhost_dev *dev = vq->dev;
2419 	struct vhost_iotlb *umem = dev->iotlb ? dev->iotlb : dev->umem;
2420 	struct iovec *_iov;
2421 	u64 s = 0, last = addr + len - 1;
2422 	int ret = 0;
2423 
2424 	while ((u64)len > s) {
2425 		u64 size;
2426 		if (unlikely(ret >= iov_size)) {
2427 			ret = -ENOBUFS;
2428 			break;
2429 		}
2430 
2431 		map = vhost_iotlb_itree_first(umem, addr, last);
2432 		if (map == NULL || map->start > addr) {
2433 			if (umem != dev->iotlb) {
2434 				ret = -EFAULT;
2435 				break;
2436 			}
2437 			ret = -EAGAIN;
2438 			break;
2439 		} else if (!(map->perm & access)) {
2440 			ret = -EPERM;
2441 			break;
2442 		}
2443 
2444 		_iov = iov + ret;
2445 		size = map->size - addr + map->start;
2446 		_iov->iov_len = min((u64)len - s, size);
2447 		_iov->iov_base = (void __user *)(unsigned long)
2448 				 (map->addr + addr - map->start);
2449 		s += size;
2450 		addr += size;
2451 		++ret;
2452 	}
2453 
2454 	if (ret == -EAGAIN)
2455 		vhost_iotlb_miss(vq, addr, access);
2456 	return ret;
2457 }
2458 
2459 /* Each buffer in the virtqueues is actually a chain of descriptors.  This
2460  * function returns the next descriptor in the chain,
2461  * or -1U if we're at the end. */
next_desc(struct vhost_virtqueue * vq,struct vring_desc * desc)2462 static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc)
2463 {
2464 	unsigned int next;
2465 
2466 	/* If this descriptor says it doesn't chain, we're done. */
2467 	if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT)))
2468 		return -1U;
2469 
2470 	/* Check they're not leading us off end of descriptors. */
2471 	next = vhost16_to_cpu(vq, READ_ONCE(desc->next));
2472 	return next;
2473 }
2474 
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)2475 static int get_indirect(struct vhost_virtqueue *vq,
2476 			struct iovec iov[], unsigned int iov_size,
2477 			unsigned int *out_num, unsigned int *in_num,
2478 			struct vhost_log *log, unsigned int *log_num,
2479 			struct vring_desc *indirect)
2480 {
2481 	struct vring_desc desc;
2482 	unsigned int i = 0, count, found = 0;
2483 	u32 len = vhost32_to_cpu(vq, indirect->len);
2484 	struct iov_iter from;
2485 	int ret, access;
2486 
2487 	/* Sanity check */
2488 	if (unlikely(len % sizeof desc)) {
2489 		vq_err(vq, "Invalid length in indirect descriptor: "
2490 		       "len 0x%llx not multiple of 0x%zx\n",
2491 		       (unsigned long long)len,
2492 		       sizeof desc);
2493 		return -EINVAL;
2494 	}
2495 
2496 	ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect,
2497 			     UIO_MAXIOV, VHOST_ACCESS_RO);
2498 	if (unlikely(ret < 0)) {
2499 		if (ret != -EAGAIN)
2500 			vq_err(vq, "Translation failure %d in indirect.\n", ret);
2501 		return ret;
2502 	}
2503 	iov_iter_init(&from, ITER_SOURCE, vq->indirect, ret, len);
2504 	count = len / sizeof desc;
2505 	/* Buffers are chained via a 16 bit next field, so
2506 	 * we can have at most 2^16 of these. */
2507 	if (unlikely(count > USHRT_MAX + 1)) {
2508 		vq_err(vq, "Indirect buffer length too big: %d\n",
2509 		       indirect->len);
2510 		return -E2BIG;
2511 	}
2512 
2513 	do {
2514 		unsigned iov_count = *in_num + *out_num;
2515 		if (unlikely(++found > count)) {
2516 			vq_err(vq, "Loop detected: last one at %u "
2517 			       "indirect size %u\n",
2518 			       i, count);
2519 			return -EINVAL;
2520 		}
2521 		if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) {
2522 			vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n",
2523 			       i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
2524 			return -EINVAL;
2525 		}
2526 		if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) {
2527 			vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n",
2528 			       i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
2529 			return -EINVAL;
2530 		}
2531 
2532 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2533 			access = VHOST_ACCESS_WO;
2534 		else
2535 			access = VHOST_ACCESS_RO;
2536 
2537 		ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2538 				     vhost32_to_cpu(vq, desc.len), iov + iov_count,
2539 				     iov_size - iov_count, access);
2540 		if (unlikely(ret < 0)) {
2541 			if (ret != -EAGAIN)
2542 				vq_err(vq, "Translation failure %d indirect idx %d\n",
2543 					ret, i);
2544 			return ret;
2545 		}
2546 		/* If this is an input descriptor, increment that count. */
2547 		if (access == VHOST_ACCESS_WO) {
2548 			*in_num += ret;
2549 			if (unlikely(log && ret)) {
2550 				log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2551 				log[*log_num].len = vhost32_to_cpu(vq, desc.len);
2552 				++*log_num;
2553 			}
2554 		} else {
2555 			/* If it's an output descriptor, they're all supposed
2556 			 * to come before any input descriptors. */
2557 			if (unlikely(*in_num)) {
2558 				vq_err(vq, "Indirect descriptor "
2559 				       "has out after in: idx %d\n", i);
2560 				return -EINVAL;
2561 			}
2562 			*out_num += ret;
2563 		}
2564 	} while ((i = next_desc(vq, &desc)) != -1);
2565 	return 0;
2566 }
2567 
2568 /* This looks in the virtqueue and for the first available buffer, and converts
2569  * it to an iovec for convenient access.  Since descriptors consist of some
2570  * number of output then some number of input descriptors, it's actually two
2571  * iovecs, but we pack them into one and note how many of each there were.
2572  *
2573  * This function returns the descriptor number found, or vq->num (which is
2574  * never a valid descriptor number) if none was found.  A negative code is
2575  * 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)2576 int vhost_get_vq_desc(struct vhost_virtqueue *vq,
2577 		      struct iovec iov[], unsigned int iov_size,
2578 		      unsigned int *out_num, unsigned int *in_num,
2579 		      struct vhost_log *log, unsigned int *log_num)
2580 {
2581 	struct vring_desc desc;
2582 	unsigned int i, head, found = 0;
2583 	u16 last_avail_idx = vq->last_avail_idx;
2584 	__virtio16 ring_head;
2585 	int ret, access;
2586 
2587 	if (vq->avail_idx == vq->last_avail_idx) {
2588 		ret = vhost_get_avail_idx(vq);
2589 		if (unlikely(ret < 0))
2590 			return ret;
2591 
2592 		if (!ret)
2593 			return vq->num;
2594 	}
2595 
2596 	/* Grab the next descriptor number they're advertising, and increment
2597 	 * the index we've seen. */
2598 	if (unlikely(vhost_get_avail_head(vq, &ring_head, last_avail_idx))) {
2599 		vq_err(vq, "Failed to read head: idx %d address %p\n",
2600 		       last_avail_idx,
2601 		       &vq->avail->ring[last_avail_idx % vq->num]);
2602 		return -EFAULT;
2603 	}
2604 
2605 	head = vhost16_to_cpu(vq, ring_head);
2606 
2607 	/* If their number is silly, that's an error. */
2608 	if (unlikely(head >= vq->num)) {
2609 		vq_err(vq, "Guest says index %u > %u is available",
2610 		       head, vq->num);
2611 		return -EINVAL;
2612 	}
2613 
2614 	/* When we start there are none of either input nor output. */
2615 	*out_num = *in_num = 0;
2616 	if (unlikely(log))
2617 		*log_num = 0;
2618 
2619 	i = head;
2620 	do {
2621 		unsigned iov_count = *in_num + *out_num;
2622 		if (unlikely(i >= vq->num)) {
2623 			vq_err(vq, "Desc index is %u > %u, head = %u",
2624 			       i, vq->num, head);
2625 			return -EINVAL;
2626 		}
2627 		if (unlikely(++found > vq->num)) {
2628 			vq_err(vq, "Loop detected: last one at %u "
2629 			       "vq size %u head %u\n",
2630 			       i, vq->num, head);
2631 			return -EINVAL;
2632 		}
2633 		ret = vhost_get_desc(vq, &desc, i);
2634 		if (unlikely(ret)) {
2635 			vq_err(vq, "Failed to get descriptor: idx %d addr %p\n",
2636 			       i, vq->desc + i);
2637 			return -EFAULT;
2638 		}
2639 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) {
2640 			ret = get_indirect(vq, iov, iov_size,
2641 					   out_num, in_num,
2642 					   log, log_num, &desc);
2643 			if (unlikely(ret < 0)) {
2644 				if (ret != -EAGAIN)
2645 					vq_err(vq, "Failure detected "
2646 						"in indirect descriptor at idx %d\n", i);
2647 				return ret;
2648 			}
2649 			continue;
2650 		}
2651 
2652 		if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2653 			access = VHOST_ACCESS_WO;
2654 		else
2655 			access = VHOST_ACCESS_RO;
2656 		ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2657 				     vhost32_to_cpu(vq, desc.len), iov + iov_count,
2658 				     iov_size - iov_count, access);
2659 		if (unlikely(ret < 0)) {
2660 			if (ret != -EAGAIN)
2661 				vq_err(vq, "Translation failure %d descriptor idx %d\n",
2662 					ret, i);
2663 			return ret;
2664 		}
2665 		if (access == VHOST_ACCESS_WO) {
2666 			/* If this is an input descriptor,
2667 			 * increment that count. */
2668 			*in_num += ret;
2669 			if (unlikely(log && ret)) {
2670 				log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2671 				log[*log_num].len = vhost32_to_cpu(vq, desc.len);
2672 				++*log_num;
2673 			}
2674 		} else {
2675 			/* If it's an output descriptor, they're all supposed
2676 			 * to come before any input descriptors. */
2677 			if (unlikely(*in_num)) {
2678 				vq_err(vq, "Descriptor has out after in: "
2679 				       "idx %d\n", i);
2680 				return -EINVAL;
2681 			}
2682 			*out_num += ret;
2683 		}
2684 	} while ((i = next_desc(vq, &desc)) != -1);
2685 
2686 	/* On success, increment avail index. */
2687 	vq->last_avail_idx++;
2688 
2689 	/* Assume notifications from guest are disabled at this point,
2690 	 * if they aren't we would need to update avail_event index. */
2691 	BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY));
2692 	return head;
2693 }
2694 EXPORT_SYMBOL_GPL(vhost_get_vq_desc);
2695 
2696 /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */
vhost_discard_vq_desc(struct vhost_virtqueue * vq,int n)2697 void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n)
2698 {
2699 	vq->last_avail_idx -= n;
2700 }
2701 EXPORT_SYMBOL_GPL(vhost_discard_vq_desc);
2702 
2703 /* After we've used one of their buffers, we tell them about it.  We'll then
2704  * want to notify the guest, using eventfd. */
vhost_add_used(struct vhost_virtqueue * vq,unsigned int head,int len)2705 int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len)
2706 {
2707 	struct vring_used_elem heads = {
2708 		cpu_to_vhost32(vq, head),
2709 		cpu_to_vhost32(vq, len)
2710 	};
2711 
2712 	return vhost_add_used_n(vq, &heads, 1);
2713 }
2714 EXPORT_SYMBOL_GPL(vhost_add_used);
2715 
__vhost_add_used_n(struct vhost_virtqueue * vq,struct vring_used_elem * heads,unsigned count)2716 static int __vhost_add_used_n(struct vhost_virtqueue *vq,
2717 			    struct vring_used_elem *heads,
2718 			    unsigned count)
2719 {
2720 	vring_used_elem_t __user *used;
2721 	u16 old, new;
2722 	int start;
2723 
2724 	start = vq->last_used_idx & (vq->num - 1);
2725 	used = vq->used->ring + start;
2726 	if (vhost_put_used(vq, heads, start, count)) {
2727 		vq_err(vq, "Failed to write used");
2728 		return -EFAULT;
2729 	}
2730 	if (unlikely(vq->log_used)) {
2731 		/* Make sure data is seen before log. */
2732 		smp_wmb();
2733 		/* Log used ring entry write. */
2734 		log_used(vq, ((void __user *)used - (void __user *)vq->used),
2735 			 count * sizeof *used);
2736 	}
2737 	old = vq->last_used_idx;
2738 	new = (vq->last_used_idx += count);
2739 	/* If the driver never bothers to signal in a very long while,
2740 	 * used index might wrap around. If that happens, invalidate
2741 	 * signalled_used index we stored. TODO: make sure driver
2742 	 * signals at least once in 2^16 and remove this. */
2743 	if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old)))
2744 		vq->signalled_used_valid = false;
2745 	return 0;
2746 }
2747 
2748 /* After we've used one of their buffers, we tell them about it.  We'll then
2749  * want to notify the guest, using eventfd. */
vhost_add_used_n(struct vhost_virtqueue * vq,struct vring_used_elem * heads,unsigned count)2750 int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads,
2751 		     unsigned count)
2752 {
2753 	int start, n, r;
2754 
2755 	start = vq->last_used_idx & (vq->num - 1);
2756 	n = vq->num - start;
2757 	if (n < count) {
2758 		r = __vhost_add_used_n(vq, heads, n);
2759 		if (r < 0)
2760 			return r;
2761 		heads += n;
2762 		count -= n;
2763 	}
2764 	r = __vhost_add_used_n(vq, heads, count);
2765 
2766 	/* Make sure buffer is written before we update index. */
2767 	smp_wmb();
2768 	if (vhost_put_used_idx(vq)) {
2769 		vq_err(vq, "Failed to increment used idx");
2770 		return -EFAULT;
2771 	}
2772 	if (unlikely(vq->log_used)) {
2773 		/* Make sure used idx is seen before log. */
2774 		smp_wmb();
2775 		/* Log used index update. */
2776 		log_used(vq, offsetof(struct vring_used, idx),
2777 			 sizeof vq->used->idx);
2778 		if (vq->log_ctx)
2779 			eventfd_signal(vq->log_ctx);
2780 	}
2781 	return r;
2782 }
2783 EXPORT_SYMBOL_GPL(vhost_add_used_n);
2784 
vhost_notify(struct vhost_dev * dev,struct vhost_virtqueue * vq)2785 static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2786 {
2787 	__u16 old, new;
2788 	__virtio16 event;
2789 	bool v;
2790 	/* Flush out used index updates. This is paired
2791 	 * with the barrier that the Guest executes when enabling
2792 	 * interrupts. */
2793 	smp_mb();
2794 
2795 	if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2796 	    unlikely(vq->avail_idx == vq->last_avail_idx))
2797 		return true;
2798 
2799 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2800 		__virtio16 flags;
2801 		if (vhost_get_avail_flags(vq, &flags)) {
2802 			vq_err(vq, "Failed to get flags");
2803 			return true;
2804 		}
2805 		return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT));
2806 	}
2807 	old = vq->signalled_used;
2808 	v = vq->signalled_used_valid;
2809 	new = vq->signalled_used = vq->last_used_idx;
2810 	vq->signalled_used_valid = true;
2811 
2812 	if (unlikely(!v))
2813 		return true;
2814 
2815 	if (vhost_get_used_event(vq, &event)) {
2816 		vq_err(vq, "Failed to get used event idx");
2817 		return true;
2818 	}
2819 	return vring_need_event(vhost16_to_cpu(vq, event), new, old);
2820 }
2821 
2822 /* This actually signals the guest, using eventfd. */
vhost_signal(struct vhost_dev * dev,struct vhost_virtqueue * vq)2823 void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2824 {
2825 	/* Signal the Guest tell them we used something up. */
2826 	if (vq->call_ctx.ctx && vhost_notify(dev, vq))
2827 		eventfd_signal(vq->call_ctx.ctx);
2828 }
2829 EXPORT_SYMBOL_GPL(vhost_signal);
2830 
2831 /* 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)2832 void vhost_add_used_and_signal(struct vhost_dev *dev,
2833 			       struct vhost_virtqueue *vq,
2834 			       unsigned int head, int len)
2835 {
2836 	vhost_add_used(vq, head, len);
2837 	vhost_signal(dev, vq);
2838 }
2839 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal);
2840 
2841 /* 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)2842 void vhost_add_used_and_signal_n(struct vhost_dev *dev,
2843 				 struct vhost_virtqueue *vq,
2844 				 struct vring_used_elem *heads, unsigned count)
2845 {
2846 	vhost_add_used_n(vq, heads, count);
2847 	vhost_signal(dev, vq);
2848 }
2849 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n);
2850 
2851 /* return true if we're sure that avaiable ring is empty */
vhost_vq_avail_empty(struct vhost_dev * dev,struct vhost_virtqueue * vq)2852 bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2853 {
2854 	int r;
2855 
2856 	if (vq->avail_idx != vq->last_avail_idx)
2857 		return false;
2858 
2859 	r = vhost_get_avail_idx(vq);
2860 
2861 	/* Note: we treat error as non-empty here */
2862 	return r == 0;
2863 }
2864 EXPORT_SYMBOL_GPL(vhost_vq_avail_empty);
2865 
2866 /* OK, now we need to know about added descriptors. */
vhost_enable_notify(struct vhost_dev * dev,struct vhost_virtqueue * vq)2867 bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2868 {
2869 	int r;
2870 
2871 	if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY))
2872 		return false;
2873 	vq->used_flags &= ~VRING_USED_F_NO_NOTIFY;
2874 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2875 		r = vhost_update_used_flags(vq);
2876 		if (r) {
2877 			vq_err(vq, "Failed to enable notification at %p: %d\n",
2878 			       &vq->used->flags, r);
2879 			return false;
2880 		}
2881 	} else {
2882 		r = vhost_update_avail_event(vq);
2883 		if (r) {
2884 			vq_err(vq, "Failed to update avail event index at %p: %d\n",
2885 			       vhost_avail_event(vq), r);
2886 			return false;
2887 		}
2888 	}
2889 	/* They could have slipped one in as we were doing that: make
2890 	 * sure it's written, then check again. */
2891 	smp_mb();
2892 
2893 	r = vhost_get_avail_idx(vq);
2894 	/* Note: we treat error as empty here */
2895 	if (unlikely(r < 0))
2896 		return false;
2897 
2898 	return r;
2899 }
2900 EXPORT_SYMBOL_GPL(vhost_enable_notify);
2901 
2902 /* We don't need to be notified again. */
vhost_disable_notify(struct vhost_dev * dev,struct vhost_virtqueue * vq)2903 void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2904 {
2905 	int r;
2906 
2907 	if (vq->used_flags & VRING_USED_F_NO_NOTIFY)
2908 		return;
2909 	vq->used_flags |= VRING_USED_F_NO_NOTIFY;
2910 	if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2911 		r = vhost_update_used_flags(vq);
2912 		if (r)
2913 			vq_err(vq, "Failed to disable notification at %p: %d\n",
2914 			       &vq->used->flags, r);
2915 	}
2916 }
2917 EXPORT_SYMBOL_GPL(vhost_disable_notify);
2918 
2919 /* Create a new message. */
vhost_new_msg(struct vhost_virtqueue * vq,int type)2920 struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type)
2921 {
2922 	/* Make sure all padding within the structure is initialized. */
2923 	struct vhost_msg_node *node = kzalloc(sizeof(*node), GFP_KERNEL);
2924 	if (!node)
2925 		return NULL;
2926 
2927 	node->vq = vq;
2928 	node->msg.type = type;
2929 	return node;
2930 }
2931 EXPORT_SYMBOL_GPL(vhost_new_msg);
2932 
vhost_enqueue_msg(struct vhost_dev * dev,struct list_head * head,struct vhost_msg_node * node)2933 void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head,
2934 		       struct vhost_msg_node *node)
2935 {
2936 	spin_lock(&dev->iotlb_lock);
2937 	list_add_tail(&node->node, head);
2938 	spin_unlock(&dev->iotlb_lock);
2939 
2940 	wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
2941 }
2942 EXPORT_SYMBOL_GPL(vhost_enqueue_msg);
2943 
vhost_dequeue_msg(struct vhost_dev * dev,struct list_head * head)2944 struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
2945 					 struct list_head *head)
2946 {
2947 	struct vhost_msg_node *node = NULL;
2948 
2949 	spin_lock(&dev->iotlb_lock);
2950 	if (!list_empty(head)) {
2951 		node = list_first_entry(head, struct vhost_msg_node,
2952 					node);
2953 		list_del(&node->node);
2954 	}
2955 	spin_unlock(&dev->iotlb_lock);
2956 
2957 	return node;
2958 }
2959 EXPORT_SYMBOL_GPL(vhost_dequeue_msg);
2960 
vhost_set_backend_features(struct vhost_dev * dev,u64 features)2961 void vhost_set_backend_features(struct vhost_dev *dev, u64 features)
2962 {
2963 	struct vhost_virtqueue *vq;
2964 	int i;
2965 
2966 	mutex_lock(&dev->mutex);
2967 	for (i = 0; i < dev->nvqs; ++i) {
2968 		vq = dev->vqs[i];
2969 		mutex_lock(&vq->mutex);
2970 		vq->acked_backend_features = features;
2971 		mutex_unlock(&vq->mutex);
2972 	}
2973 	mutex_unlock(&dev->mutex);
2974 }
2975 EXPORT_SYMBOL_GPL(vhost_set_backend_features);
2976 
vhost_init(void)2977 static int __init vhost_init(void)
2978 {
2979 	return 0;
2980 }
2981 
vhost_exit(void)2982 static void __exit vhost_exit(void)
2983 {
2984 }
2985 
2986 module_init(vhost_init);
2987 module_exit(vhost_exit);
2988 
2989 MODULE_VERSION("0.0.1");
2990 MODULE_LICENSE("GPL v2");
2991 MODULE_AUTHOR("Michael S. Tsirkin");
2992 MODULE_DESCRIPTION("Host kernel accelerator for virtio");
2993