9beefa6ed1ce014bb35fa005e67dc1828cd7cb7a
[muen/linux.git] / drivers / vhost / vhost.c
1 /* Copyright (C) 2009 Red Hat, Inc.
2  * Copyright (C) 2006 Rusty Russell IBM Corporation
3  *
4  * Author: Michael S. Tsirkin <mst@redhat.com>
5  *
6  * Inspiration, some code, and most witty comments come from
7  * Documentation/virtual/lguest/lguest.c, by Rusty Russell
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.
10  *
11  * Generic code for virtio server in host kernel.
12  */
13
14 #include <linux/eventfd.h>
15 #include <linux/vhost.h>
16 #include <linux/uio.h>
17 #include <linux/mm.h>
18 #include <linux/mmu_context.h>
19 #include <linux/miscdevice.h>
20 #include <linux/mutex.h>
21 #include <linux/poll.h>
22 #include <linux/file.h>
23 #include <linux/highmem.h>
24 #include <linux/slab.h>
25 #include <linux/vmalloc.h>
26 #include <linux/kthread.h>
27 #include <linux/cgroup.h>
28 #include <linux/module.h>
29 #include <linux/sort.h>
30 #include <linux/sched/mm.h>
31 #include <linux/sched/signal.h>
32 #include <linux/interval_tree_generic.h>
33
34 #include "vhost.h"
35
36 static ushort max_mem_regions = 64;
37 module_param(max_mem_regions, ushort, 0444);
38 MODULE_PARM_DESC(max_mem_regions,
39         "Maximum number of memory regions in memory map. (default: 64)");
40 static int max_iotlb_entries = 2048;
41 module_param(max_iotlb_entries, int, 0444);
42 MODULE_PARM_DESC(max_iotlb_entries,
43         "Maximum number of iotlb entries. (default: 2048)");
44
45 enum {
46         VHOST_MEMORY_F_LOG = 0x1,
47 };
48
49 #define vhost_used_event(vq) ((__virtio16 __user *)&vq->avail->ring[vq->num])
50 #define vhost_avail_event(vq) ((__virtio16 __user *)&vq->used->ring[vq->num])
51
52 INTERVAL_TREE_DEFINE(struct vhost_umem_node,
53                      rb, __u64, __subtree_last,
54                      START, LAST, static inline, vhost_umem_interval_tree);
55
56 #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
57 static void vhost_disable_cross_endian(struct vhost_virtqueue *vq)
58 {
59         vq->user_be = !virtio_legacy_is_little_endian();
60 }
61
62 static void vhost_enable_cross_endian_big(struct vhost_virtqueue *vq)
63 {
64         vq->user_be = true;
65 }
66
67 static void vhost_enable_cross_endian_little(struct vhost_virtqueue *vq)
68 {
69         vq->user_be = false;
70 }
71
72 static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp)
73 {
74         struct vhost_vring_state s;
75
76         if (vq->private_data)
77                 return -EBUSY;
78
79         if (copy_from_user(&s, argp, sizeof(s)))
80                 return -EFAULT;
81
82         if (s.num != VHOST_VRING_LITTLE_ENDIAN &&
83             s.num != VHOST_VRING_BIG_ENDIAN)
84                 return -EINVAL;
85
86         if (s.num == VHOST_VRING_BIG_ENDIAN)
87                 vhost_enable_cross_endian_big(vq);
88         else
89                 vhost_enable_cross_endian_little(vq);
90
91         return 0;
92 }
93
94 static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx,
95                                    int __user *argp)
96 {
97         struct vhost_vring_state s = {
98                 .index = idx,
99                 .num = vq->user_be
100         };
101
102         if (copy_to_user(argp, &s, sizeof(s)))
103                 return -EFAULT;
104
105         return 0;
106 }
107
108 static void vhost_init_is_le(struct vhost_virtqueue *vq)
109 {
110         /* Note for legacy virtio: user_be is initialized at reset time
111          * according to the host endianness. If userspace does not set an
112          * explicit endianness, the default behavior is native endian, as
113          * expected by legacy virtio.
114          */
115         vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) || !vq->user_be;
116 }
117 #else
118 static void vhost_disable_cross_endian(struct vhost_virtqueue *vq)
119 {
120 }
121
122 static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp)
123 {
124         return -ENOIOCTLCMD;
125 }
126
127 static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx,
128                                    int __user *argp)
129 {
130         return -ENOIOCTLCMD;
131 }
132
133 static void vhost_init_is_le(struct vhost_virtqueue *vq)
134 {
135         vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1)
136                 || virtio_legacy_is_little_endian();
137 }
138 #endif /* CONFIG_VHOST_CROSS_ENDIAN_LEGACY */
139
140 static void vhost_reset_is_le(struct vhost_virtqueue *vq)
141 {
142         vhost_init_is_le(vq);
143 }
144
145 struct vhost_flush_struct {
146         struct vhost_work work;
147         struct completion wait_event;
148 };
149
150 static void vhost_flush_work(struct vhost_work *work)
151 {
152         struct vhost_flush_struct *s;
153
154         s = container_of(work, struct vhost_flush_struct, work);
155         complete(&s->wait_event);
156 }
157
158 static void vhost_poll_func(struct file *file, wait_queue_head_t *wqh,
159                             poll_table *pt)
160 {
161         struct vhost_poll *poll;
162
163         poll = container_of(pt, struct vhost_poll, table);
164         poll->wqh = wqh;
165         add_wait_queue(wqh, &poll->wait);
166 }
167
168 static int vhost_poll_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync,
169                              void *key)
170 {
171         struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait);
172
173         if (!(key_to_poll(key) & poll->mask))
174                 return 0;
175
176         vhost_poll_queue(poll);
177         return 0;
178 }
179
180 void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn)
181 {
182         clear_bit(VHOST_WORK_QUEUED, &work->flags);
183         work->fn = fn;
184 }
185 EXPORT_SYMBOL_GPL(vhost_work_init);
186
187 /* Init poll structure */
188 void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
189                      __poll_t mask, struct vhost_dev *dev)
190 {
191         init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup);
192         init_poll_funcptr(&poll->table, vhost_poll_func);
193         poll->mask = mask;
194         poll->dev = dev;
195         poll->wqh = NULL;
196
197         vhost_work_init(&poll->work, fn);
198 }
199 EXPORT_SYMBOL_GPL(vhost_poll_init);
200
201 /* Start polling a file. We add ourselves to file's wait queue. The caller must
202  * keep a reference to a file until after vhost_poll_stop is called. */
203 int vhost_poll_start(struct vhost_poll *poll, struct file *file)
204 {
205         __poll_t mask;
206         int ret = 0;
207
208         if (poll->wqh)
209                 return 0;
210
211         mask = file->f_op->poll(file, &poll->table);
212         if (mask)
213                 vhost_poll_wakeup(&poll->wait, 0, 0, poll_to_key(mask));
214         if (mask & EPOLLERR) {
215                 vhost_poll_stop(poll);
216                 ret = -EINVAL;
217         }
218
219         return ret;
220 }
221 EXPORT_SYMBOL_GPL(vhost_poll_start);
222
223 /* Stop polling a file. After this function returns, it becomes safe to drop the
224  * file reference. You must also flush afterwards. */
225 void vhost_poll_stop(struct vhost_poll *poll)
226 {
227         if (poll->wqh) {
228                 remove_wait_queue(poll->wqh, &poll->wait);
229                 poll->wqh = NULL;
230         }
231 }
232 EXPORT_SYMBOL_GPL(vhost_poll_stop);
233
234 void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work)
235 {
236         struct vhost_flush_struct flush;
237
238         if (dev->worker) {
239                 init_completion(&flush.wait_event);
240                 vhost_work_init(&flush.work, vhost_flush_work);
241
242                 vhost_work_queue(dev, &flush.work);
243                 wait_for_completion(&flush.wait_event);
244         }
245 }
246 EXPORT_SYMBOL_GPL(vhost_work_flush);
247
248 /* Flush any work that has been scheduled. When calling this, don't hold any
249  * locks that are also used by the callback. */
250 void vhost_poll_flush(struct vhost_poll *poll)
251 {
252         vhost_work_flush(poll->dev, &poll->work);
253 }
254 EXPORT_SYMBOL_GPL(vhost_poll_flush);
255
256 void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work)
257 {
258         if (!dev->worker)
259                 return;
260
261         if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) {
262                 /* We can only add the work to the list after we're
263                  * sure it was not in the list.
264                  * test_and_set_bit() implies a memory barrier.
265                  */
266                 llist_add(&work->node, &dev->work_list);
267                 wake_up_process(dev->worker);
268         }
269 }
270 EXPORT_SYMBOL_GPL(vhost_work_queue);
271
272 /* A lockless hint for busy polling code to exit the loop */
273 bool vhost_has_work(struct vhost_dev *dev)
274 {
275         return !llist_empty(&dev->work_list);
276 }
277 EXPORT_SYMBOL_GPL(vhost_has_work);
278
279 void vhost_poll_queue(struct vhost_poll *poll)
280 {
281         vhost_work_queue(poll->dev, &poll->work);
282 }
283 EXPORT_SYMBOL_GPL(vhost_poll_queue);
284
285 static void __vhost_vq_meta_reset(struct vhost_virtqueue *vq)
286 {
287         int j;
288
289         for (j = 0; j < VHOST_NUM_ADDRS; j++)
290                 vq->meta_iotlb[j] = NULL;
291 }
292
293 static void vhost_vq_meta_reset(struct vhost_dev *d)
294 {
295         int i;
296
297         for (i = 0; i < d->nvqs; ++i)
298                 __vhost_vq_meta_reset(d->vqs[i]);
299 }
300
301 static void vhost_vq_reset(struct vhost_dev *dev,
302                            struct vhost_virtqueue *vq)
303 {
304         vq->num = 1;
305         vq->desc = NULL;
306         vq->avail = NULL;
307         vq->used = NULL;
308         vq->last_avail_idx = 0;
309         vq->avail_idx = 0;
310         vq->last_used_idx = 0;
311         vq->signalled_used = 0;
312         vq->signalled_used_valid = false;
313         vq->used_flags = 0;
314         vq->log_used = false;
315         vq->log_addr = -1ull;
316         vq->private_data = NULL;
317         vq->acked_features = 0;
318         vq->log_base = NULL;
319         vq->error_ctx = NULL;
320         vq->kick = NULL;
321         vq->call_ctx = NULL;
322         vq->log_ctx = NULL;
323         vhost_reset_is_le(vq);
324         vhost_disable_cross_endian(vq);
325         vq->busyloop_timeout = 0;
326         vq->umem = NULL;
327         vq->iotlb = NULL;
328         __vhost_vq_meta_reset(vq);
329 }
330
331 static int vhost_worker(void *data)
332 {
333         struct vhost_dev *dev = data;
334         struct vhost_work *work, *work_next;
335         struct llist_node *node;
336         mm_segment_t oldfs = get_fs();
337
338         set_fs(USER_DS);
339         use_mm(dev->mm);
340
341         for (;;) {
342                 /* mb paired w/ kthread_stop */
343                 set_current_state(TASK_INTERRUPTIBLE);
344
345                 if (kthread_should_stop()) {
346                         __set_current_state(TASK_RUNNING);
347                         break;
348                 }
349
350                 node = llist_del_all(&dev->work_list);
351                 if (!node)
352                         schedule();
353
354                 node = llist_reverse_order(node);
355                 /* make sure flag is seen after deletion */
356                 smp_wmb();
357                 llist_for_each_entry_safe(work, work_next, node, node) {
358                         clear_bit(VHOST_WORK_QUEUED, &work->flags);
359                         __set_current_state(TASK_RUNNING);
360                         work->fn(work);
361                         if (need_resched())
362                                 schedule();
363                 }
364         }
365         unuse_mm(dev->mm);
366         set_fs(oldfs);
367         return 0;
368 }
369
370 static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq)
371 {
372         kfree(vq->indirect);
373         vq->indirect = NULL;
374         kfree(vq->log);
375         vq->log = NULL;
376         kfree(vq->heads);
377         vq->heads = NULL;
378 }
379
380 /* Helper to allocate iovec buffers for all vqs. */
381 static long vhost_dev_alloc_iovecs(struct vhost_dev *dev)
382 {
383         struct vhost_virtqueue *vq;
384         int i;
385
386         for (i = 0; i < dev->nvqs; ++i) {
387                 vq = dev->vqs[i];
388                 vq->indirect = kmalloc(sizeof *vq->indirect * UIO_MAXIOV,
389                                        GFP_KERNEL);
390                 vq->log = kmalloc(sizeof *vq->log * UIO_MAXIOV, GFP_KERNEL);
391                 vq->heads = kmalloc(sizeof *vq->heads * UIO_MAXIOV, GFP_KERNEL);
392                 if (!vq->indirect || !vq->log || !vq->heads)
393                         goto err_nomem;
394         }
395         return 0;
396
397 err_nomem:
398         for (; i >= 0; --i)
399                 vhost_vq_free_iovecs(dev->vqs[i]);
400         return -ENOMEM;
401 }
402
403 static void vhost_dev_free_iovecs(struct vhost_dev *dev)
404 {
405         int i;
406
407         for (i = 0; i < dev->nvqs; ++i)
408                 vhost_vq_free_iovecs(dev->vqs[i]);
409 }
410
411 void vhost_dev_init(struct vhost_dev *dev,
412                     struct vhost_virtqueue **vqs, int nvqs)
413 {
414         struct vhost_virtqueue *vq;
415         int i;
416
417         dev->vqs = vqs;
418         dev->nvqs = nvqs;
419         mutex_init(&dev->mutex);
420         dev->log_ctx = NULL;
421         dev->umem = NULL;
422         dev->iotlb = NULL;
423         dev->mm = NULL;
424         dev->worker = NULL;
425         init_llist_head(&dev->work_list);
426         init_waitqueue_head(&dev->wait);
427         INIT_LIST_HEAD(&dev->read_list);
428         INIT_LIST_HEAD(&dev->pending_list);
429         spin_lock_init(&dev->iotlb_lock);
430
431
432         for (i = 0; i < dev->nvqs; ++i) {
433                 vq = dev->vqs[i];
434                 vq->log = NULL;
435                 vq->indirect = NULL;
436                 vq->heads = NULL;
437                 vq->dev = dev;
438                 mutex_init(&vq->mutex);
439                 vhost_vq_reset(dev, vq);
440                 if (vq->handle_kick)
441                         vhost_poll_init(&vq->poll, vq->handle_kick,
442                                         EPOLLIN, dev);
443         }
444 }
445 EXPORT_SYMBOL_GPL(vhost_dev_init);
446
447 /* Caller should have device mutex */
448 long vhost_dev_check_owner(struct vhost_dev *dev)
449 {
450         /* Are you the owner? If not, I don't think you mean to do that */
451         return dev->mm == current->mm ? 0 : -EPERM;
452 }
453 EXPORT_SYMBOL_GPL(vhost_dev_check_owner);
454
455 struct vhost_attach_cgroups_struct {
456         struct vhost_work work;
457         struct task_struct *owner;
458         int ret;
459 };
460
461 static void vhost_attach_cgroups_work(struct vhost_work *work)
462 {
463         struct vhost_attach_cgroups_struct *s;
464
465         s = container_of(work, struct vhost_attach_cgroups_struct, work);
466         s->ret = cgroup_attach_task_all(s->owner, current);
467 }
468
469 static int vhost_attach_cgroups(struct vhost_dev *dev)
470 {
471         struct vhost_attach_cgroups_struct attach;
472
473         attach.owner = current;
474         vhost_work_init(&attach.work, vhost_attach_cgroups_work);
475         vhost_work_queue(dev, &attach.work);
476         vhost_work_flush(dev, &attach.work);
477         return attach.ret;
478 }
479
480 /* Caller should have device mutex */
481 bool vhost_dev_has_owner(struct vhost_dev *dev)
482 {
483         return dev->mm;
484 }
485 EXPORT_SYMBOL_GPL(vhost_dev_has_owner);
486
487 /* Caller should have device mutex */
488 long vhost_dev_set_owner(struct vhost_dev *dev)
489 {
490         struct task_struct *worker;
491         int err;
492
493         /* Is there an owner already? */
494         if (vhost_dev_has_owner(dev)) {
495                 err = -EBUSY;
496                 goto err_mm;
497         }
498
499         /* No owner, become one */
500         dev->mm = get_task_mm(current);
501         worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid);
502         if (IS_ERR(worker)) {
503                 err = PTR_ERR(worker);
504                 goto err_worker;
505         }
506
507         dev->worker = worker;
508         wake_up_process(worker);        /* avoid contributing to loadavg */
509
510         err = vhost_attach_cgroups(dev);
511         if (err)
512                 goto err_cgroup;
513
514         err = vhost_dev_alloc_iovecs(dev);
515         if (err)
516                 goto err_cgroup;
517
518         return 0;
519 err_cgroup:
520         kthread_stop(worker);
521         dev->worker = NULL;
522 err_worker:
523         if (dev->mm)
524                 mmput(dev->mm);
525         dev->mm = NULL;
526 err_mm:
527         return err;
528 }
529 EXPORT_SYMBOL_GPL(vhost_dev_set_owner);
530
531 struct vhost_umem *vhost_dev_reset_owner_prepare(void)
532 {
533         return kvzalloc(sizeof(struct vhost_umem), GFP_KERNEL);
534 }
535 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare);
536
537 /* Caller should have device mutex */
538 void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_umem *umem)
539 {
540         int i;
541
542         vhost_dev_cleanup(dev);
543
544         /* Restore memory to default empty mapping. */
545         INIT_LIST_HEAD(&umem->umem_list);
546         dev->umem = umem;
547         /* We don't need VQ locks below since vhost_dev_cleanup makes sure
548          * VQs aren't running.
549          */
550         for (i = 0; i < dev->nvqs; ++i)
551                 dev->vqs[i]->umem = umem;
552 }
553 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner);
554
555 void vhost_dev_stop(struct vhost_dev *dev)
556 {
557         int i;
558
559         for (i = 0; i < dev->nvqs; ++i) {
560                 if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) {
561                         vhost_poll_stop(&dev->vqs[i]->poll);
562                         vhost_poll_flush(&dev->vqs[i]->poll);
563                 }
564         }
565 }
566 EXPORT_SYMBOL_GPL(vhost_dev_stop);
567
568 static void vhost_umem_free(struct vhost_umem *umem,
569                             struct vhost_umem_node *node)
570 {
571         vhost_umem_interval_tree_remove(node, &umem->umem_tree);
572         list_del(&node->link);
573         kfree(node);
574         umem->numem--;
575 }
576
577 static void vhost_umem_clean(struct vhost_umem *umem)
578 {
579         struct vhost_umem_node *node, *tmp;
580
581         if (!umem)
582                 return;
583
584         list_for_each_entry_safe(node, tmp, &umem->umem_list, link)
585                 vhost_umem_free(umem, node);
586
587         kvfree(umem);
588 }
589
590 static void vhost_clear_msg(struct vhost_dev *dev)
591 {
592         struct vhost_msg_node *node, *n;
593
594         spin_lock(&dev->iotlb_lock);
595
596         list_for_each_entry_safe(node, n, &dev->read_list, node) {
597                 list_del(&node->node);
598                 kfree(node);
599         }
600
601         list_for_each_entry_safe(node, n, &dev->pending_list, node) {
602                 list_del(&node->node);
603                 kfree(node);
604         }
605
606         spin_unlock(&dev->iotlb_lock);
607 }
608
609 void vhost_dev_cleanup(struct vhost_dev *dev)
610 {
611         int i;
612
613         for (i = 0; i < dev->nvqs; ++i) {
614                 if (dev->vqs[i]->error_ctx)
615                         eventfd_ctx_put(dev->vqs[i]->error_ctx);
616                 if (dev->vqs[i]->kick)
617                         fput(dev->vqs[i]->kick);
618                 if (dev->vqs[i]->call_ctx)
619                         eventfd_ctx_put(dev->vqs[i]->call_ctx);
620                 vhost_vq_reset(dev, dev->vqs[i]);
621         }
622         vhost_dev_free_iovecs(dev);
623         if (dev->log_ctx)
624                 eventfd_ctx_put(dev->log_ctx);
625         dev->log_ctx = NULL;
626         /* No one will access memory at this point */
627         vhost_umem_clean(dev->umem);
628         dev->umem = NULL;
629         vhost_umem_clean(dev->iotlb);
630         dev->iotlb = NULL;
631         vhost_clear_msg(dev);
632         wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
633         WARN_ON(!llist_empty(&dev->work_list));
634         if (dev->worker) {
635                 kthread_stop(dev->worker);
636                 dev->worker = NULL;
637         }
638         if (dev->mm)
639                 mmput(dev->mm);
640         dev->mm = NULL;
641 }
642 EXPORT_SYMBOL_GPL(vhost_dev_cleanup);
643
644 static bool log_access_ok(void __user *log_base, u64 addr, unsigned long sz)
645 {
646         u64 a = addr / VHOST_PAGE_SIZE / 8;
647
648         /* Make sure 64 bit math will not overflow. */
649         if (a > ULONG_MAX - (unsigned long)log_base ||
650             a + (unsigned long)log_base > ULONG_MAX)
651                 return false;
652
653         return access_ok(VERIFY_WRITE, log_base + a,
654                          (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8);
655 }
656
657 static bool vhost_overflow(u64 uaddr, u64 size)
658 {
659         /* Make sure 64 bit math will not overflow. */
660         return uaddr > ULONG_MAX || size > ULONG_MAX || uaddr > ULONG_MAX - size;
661 }
662
663 /* Caller should have vq mutex and device mutex. */
664 static bool vq_memory_access_ok(void __user *log_base, struct vhost_umem *umem,
665                                 int log_all)
666 {
667         struct vhost_umem_node *node;
668
669         if (!umem)
670                 return false;
671
672         list_for_each_entry(node, &umem->umem_list, link) {
673                 unsigned long a = node->userspace_addr;
674
675                 if (vhost_overflow(node->userspace_addr, node->size))
676                         return false;
677
678
679                 if (!access_ok(VERIFY_WRITE, (void __user *)a,
680                                     node->size))
681                         return false;
682                 else if (log_all && !log_access_ok(log_base,
683                                                    node->start,
684                                                    node->size))
685                         return false;
686         }
687         return true;
688 }
689
690 static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq,
691                                                u64 addr, unsigned int size,
692                                                int type)
693 {
694         const struct vhost_umem_node *node = vq->meta_iotlb[type];
695
696         if (!node)
697                 return NULL;
698
699         return (void *)(uintptr_t)(node->userspace_addr + addr - node->start);
700 }
701
702 /* Can we switch to this memory table? */
703 /* Caller should have device mutex but not vq mutex */
704 static bool memory_access_ok(struct vhost_dev *d, struct vhost_umem *umem,
705                              int log_all)
706 {
707         int i;
708
709         for (i = 0; i < d->nvqs; ++i) {
710                 bool ok;
711                 bool log;
712
713                 mutex_lock(&d->vqs[i]->mutex);
714                 log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL);
715                 /* If ring is inactive, will check when it's enabled. */
716                 if (d->vqs[i]->private_data)
717                         ok = vq_memory_access_ok(d->vqs[i]->log_base,
718                                                  umem, log);
719                 else
720                         ok = true;
721                 mutex_unlock(&d->vqs[i]->mutex);
722                 if (!ok)
723                         return false;
724         }
725         return true;
726 }
727
728 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
729                           struct iovec iov[], int iov_size, int access);
730
731 static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to,
732                               const void *from, unsigned size)
733 {
734         int ret;
735
736         if (!vq->iotlb)
737                 return __copy_to_user(to, from, size);
738         else {
739                 /* This function should be called after iotlb
740                  * prefetch, which means we're sure that all vq
741                  * could be access through iotlb. So -EAGAIN should
742                  * not happen in this case.
743                  */
744                 struct iov_iter t;
745                 void __user *uaddr = vhost_vq_meta_fetch(vq,
746                                      (u64)(uintptr_t)to, size,
747                                      VHOST_ADDR_USED);
748
749                 if (uaddr)
750                         return __copy_to_user(uaddr, from, size);
751
752                 ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov,
753                                      ARRAY_SIZE(vq->iotlb_iov),
754                                      VHOST_ACCESS_WO);
755                 if (ret < 0)
756                         goto out;
757                 iov_iter_init(&t, WRITE, vq->iotlb_iov, ret, size);
758                 ret = copy_to_iter(from, size, &t);
759                 if (ret == size)
760                         ret = 0;
761         }
762 out:
763         return ret;
764 }
765
766 static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to,
767                                 void __user *from, unsigned size)
768 {
769         int ret;
770
771         if (!vq->iotlb)
772                 return __copy_from_user(to, from, size);
773         else {
774                 /* This function should be called after iotlb
775                  * prefetch, which means we're sure that vq
776                  * could be access through iotlb. So -EAGAIN should
777                  * not happen in this case.
778                  */
779                 void __user *uaddr = vhost_vq_meta_fetch(vq,
780                                      (u64)(uintptr_t)from, size,
781                                      VHOST_ADDR_DESC);
782                 struct iov_iter f;
783
784                 if (uaddr)
785                         return __copy_from_user(to, uaddr, size);
786
787                 ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov,
788                                      ARRAY_SIZE(vq->iotlb_iov),
789                                      VHOST_ACCESS_RO);
790                 if (ret < 0) {
791                         vq_err(vq, "IOTLB translation failure: uaddr "
792                                "%p size 0x%llx\n", from,
793                                (unsigned long long) size);
794                         goto out;
795                 }
796                 iov_iter_init(&f, READ, vq->iotlb_iov, ret, size);
797                 ret = copy_from_iter(to, size, &f);
798                 if (ret == size)
799                         ret = 0;
800         }
801
802 out:
803         return ret;
804 }
805
806 static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq,
807                                           void __user *addr, unsigned int size,
808                                           int type)
809 {
810         int ret;
811
812         ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov,
813                              ARRAY_SIZE(vq->iotlb_iov),
814                              VHOST_ACCESS_RO);
815         if (ret < 0) {
816                 vq_err(vq, "IOTLB translation failure: uaddr "
817                         "%p size 0x%llx\n", addr,
818                         (unsigned long long) size);
819                 return NULL;
820         }
821
822         if (ret != 1 || vq->iotlb_iov[0].iov_len != size) {
823                 vq_err(vq, "Non atomic userspace memory access: uaddr "
824                         "%p size 0x%llx\n", addr,
825                         (unsigned long long) size);
826                 return NULL;
827         }
828
829         return vq->iotlb_iov[0].iov_base;
830 }
831
832 /* This function should be called after iotlb
833  * prefetch, which means we're sure that vq
834  * could be access through iotlb. So -EAGAIN should
835  * not happen in this case.
836  */
837 static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq,
838                                             void *addr, unsigned int size,
839                                             int type)
840 {
841         void __user *uaddr = vhost_vq_meta_fetch(vq,
842                              (u64)(uintptr_t)addr, size, type);
843         if (uaddr)
844                 return uaddr;
845
846         return __vhost_get_user_slow(vq, addr, size, type);
847 }
848
849 #define vhost_put_user(vq, x, ptr)              \
850 ({ \
851         int ret = -EFAULT; \
852         if (!vq->iotlb) { \
853                 ret = __put_user(x, ptr); \
854         } else { \
855                 __typeof__(ptr) to = \
856                         (__typeof__(ptr)) __vhost_get_user(vq, ptr,     \
857                                           sizeof(*ptr), VHOST_ADDR_USED); \
858                 if (to != NULL) \
859                         ret = __put_user(x, to); \
860                 else \
861                         ret = -EFAULT;  \
862         } \
863         ret; \
864 })
865
866 #define vhost_get_user(vq, x, ptr, type)                \
867 ({ \
868         int ret; \
869         if (!vq->iotlb) { \
870                 ret = __get_user(x, ptr); \
871         } else { \
872                 __typeof__(ptr) from = \
873                         (__typeof__(ptr)) __vhost_get_user(vq, ptr, \
874                                                            sizeof(*ptr), \
875                                                            type); \
876                 if (from != NULL) \
877                         ret = __get_user(x, from); \
878                 else \
879                         ret = -EFAULT; \
880         } \
881         ret; \
882 })
883
884 #define vhost_get_avail(vq, x, ptr) \
885         vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL)
886
887 #define vhost_get_used(vq, x, ptr) \
888         vhost_get_user(vq, x, ptr, VHOST_ADDR_USED)
889
890 static void vhost_dev_lock_vqs(struct vhost_dev *d)
891 {
892         int i = 0;
893         for (i = 0; i < d->nvqs; ++i)
894                 mutex_lock_nested(&d->vqs[i]->mutex, i);
895 }
896
897 static void vhost_dev_unlock_vqs(struct vhost_dev *d)
898 {
899         int i = 0;
900         for (i = 0; i < d->nvqs; ++i)
901                 mutex_unlock(&d->vqs[i]->mutex);
902 }
903
904 static int vhost_new_umem_range(struct vhost_umem *umem,
905                                 u64 start, u64 size, u64 end,
906                                 u64 userspace_addr, int perm)
907 {
908         struct vhost_umem_node *tmp, *node = kmalloc(sizeof(*node), GFP_ATOMIC);
909
910         if (!node)
911                 return -ENOMEM;
912
913         if (umem->numem == max_iotlb_entries) {
914                 tmp = list_first_entry(&umem->umem_list, typeof(*tmp), link);
915                 vhost_umem_free(umem, tmp);
916         }
917
918         node->start = start;
919         node->size = size;
920         node->last = end;
921         node->userspace_addr = userspace_addr;
922         node->perm = perm;
923         INIT_LIST_HEAD(&node->link);
924         list_add_tail(&node->link, &umem->umem_list);
925         vhost_umem_interval_tree_insert(node, &umem->umem_tree);
926         umem->numem++;
927
928         return 0;
929 }
930
931 static void vhost_del_umem_range(struct vhost_umem *umem,
932                                  u64 start, u64 end)
933 {
934         struct vhost_umem_node *node;
935
936         while ((node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
937                                                            start, end)))
938                 vhost_umem_free(umem, node);
939 }
940
941 static void vhost_iotlb_notify_vq(struct vhost_dev *d,
942                                   struct vhost_iotlb_msg *msg)
943 {
944         struct vhost_msg_node *node, *n;
945
946         spin_lock(&d->iotlb_lock);
947
948         list_for_each_entry_safe(node, n, &d->pending_list, node) {
949                 struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb;
950                 if (msg->iova <= vq_msg->iova &&
951                     msg->iova + msg->size - 1 > vq_msg->iova &&
952                     vq_msg->type == VHOST_IOTLB_MISS) {
953                         vhost_poll_queue(&node->vq->poll);
954                         list_del(&node->node);
955                         kfree(node);
956                 }
957         }
958
959         spin_unlock(&d->iotlb_lock);
960 }
961
962 static bool umem_access_ok(u64 uaddr, u64 size, int access)
963 {
964         unsigned long a = uaddr;
965
966         /* Make sure 64 bit math will not overflow. */
967         if (vhost_overflow(uaddr, size))
968                 return false;
969
970         if ((access & VHOST_ACCESS_RO) &&
971             !access_ok(VERIFY_READ, (void __user *)a, size))
972                 return false;
973         if ((access & VHOST_ACCESS_WO) &&
974             !access_ok(VERIFY_WRITE, (void __user *)a, size))
975                 return false;
976         return true;
977 }
978
979 static int vhost_process_iotlb_msg(struct vhost_dev *dev,
980                                    struct vhost_iotlb_msg *msg)
981 {
982         int ret = 0;
983
984         mutex_lock(&dev->mutex);
985         vhost_dev_lock_vqs(dev);
986         switch (msg->type) {
987         case VHOST_IOTLB_UPDATE:
988                 if (!dev->iotlb) {
989                         ret = -EFAULT;
990                         break;
991                 }
992                 if (!umem_access_ok(msg->uaddr, msg->size, msg->perm)) {
993                         ret = -EFAULT;
994                         break;
995                 }
996                 vhost_vq_meta_reset(dev);
997                 if (vhost_new_umem_range(dev->iotlb, msg->iova, msg->size,
998                                          msg->iova + msg->size - 1,
999                                          msg->uaddr, msg->perm)) {
1000                         ret = -ENOMEM;
1001                         break;
1002                 }
1003                 vhost_iotlb_notify_vq(dev, msg);
1004                 break;
1005         case VHOST_IOTLB_INVALIDATE:
1006                 if (!dev->iotlb) {
1007                         ret = -EFAULT;
1008                         break;
1009                 }
1010                 vhost_vq_meta_reset(dev);
1011                 vhost_del_umem_range(dev->iotlb, msg->iova,
1012                                      msg->iova + msg->size - 1);
1013                 break;
1014         default:
1015                 ret = -EINVAL;
1016                 break;
1017         }
1018
1019         vhost_dev_unlock_vqs(dev);
1020         mutex_unlock(&dev->mutex);
1021
1022         return ret;
1023 }
1024 ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
1025                              struct iov_iter *from)
1026 {
1027         struct vhost_msg_node node;
1028         unsigned size = sizeof(struct vhost_msg);
1029         size_t ret;
1030         int err;
1031
1032         if (iov_iter_count(from) < size)
1033                 return 0;
1034         ret = copy_from_iter(&node.msg, size, from);
1035         if (ret != size)
1036                 goto done;
1037
1038         switch (node.msg.type) {
1039         case VHOST_IOTLB_MSG:
1040                 err = vhost_process_iotlb_msg(dev, &node.msg.iotlb);
1041                 if (err)
1042                         ret = err;
1043                 break;
1044         default:
1045                 ret = -EINVAL;
1046                 break;
1047         }
1048
1049 done:
1050         return ret;
1051 }
1052 EXPORT_SYMBOL(vhost_chr_write_iter);
1053
1054 __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
1055                             poll_table *wait)
1056 {
1057         __poll_t mask = 0;
1058
1059         poll_wait(file, &dev->wait, wait);
1060
1061         if (!list_empty(&dev->read_list))
1062                 mask |= EPOLLIN | EPOLLRDNORM;
1063
1064         return mask;
1065 }
1066 EXPORT_SYMBOL(vhost_chr_poll);
1067
1068 ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
1069                             int noblock)
1070 {
1071         DEFINE_WAIT(wait);
1072         struct vhost_msg_node *node;
1073         ssize_t ret = 0;
1074         unsigned size = sizeof(struct vhost_msg);
1075
1076         if (iov_iter_count(to) < size)
1077                 return 0;
1078
1079         while (1) {
1080                 if (!noblock)
1081                         prepare_to_wait(&dev->wait, &wait,
1082                                         TASK_INTERRUPTIBLE);
1083
1084                 node = vhost_dequeue_msg(dev, &dev->read_list);
1085                 if (node)
1086                         break;
1087                 if (noblock) {
1088                         ret = -EAGAIN;
1089                         break;
1090                 }
1091                 if (signal_pending(current)) {
1092                         ret = -ERESTARTSYS;
1093                         break;
1094                 }
1095                 if (!dev->iotlb) {
1096                         ret = -EBADFD;
1097                         break;
1098                 }
1099
1100                 schedule();
1101         }
1102
1103         if (!noblock)
1104                 finish_wait(&dev->wait, &wait);
1105
1106         if (node) {
1107                 ret = copy_to_iter(&node->msg, size, to);
1108
1109                 if (ret != size || node->msg.type != VHOST_IOTLB_MISS) {
1110                         kfree(node);
1111                         return ret;
1112                 }
1113
1114                 vhost_enqueue_msg(dev, &dev->pending_list, node);
1115         }
1116
1117         return ret;
1118 }
1119 EXPORT_SYMBOL_GPL(vhost_chr_read_iter);
1120
1121 static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access)
1122 {
1123         struct vhost_dev *dev = vq->dev;
1124         struct vhost_msg_node *node;
1125         struct vhost_iotlb_msg *msg;
1126
1127         node = vhost_new_msg(vq, VHOST_IOTLB_MISS);
1128         if (!node)
1129                 return -ENOMEM;
1130
1131         msg = &node->msg.iotlb;
1132         msg->type = VHOST_IOTLB_MISS;
1133         msg->iova = iova;
1134         msg->perm = access;
1135
1136         vhost_enqueue_msg(dev, &dev->read_list, node);
1137
1138         return 0;
1139 }
1140
1141 static bool vq_access_ok(struct vhost_virtqueue *vq, unsigned int num,
1142                          struct vring_desc __user *desc,
1143                          struct vring_avail __user *avail,
1144                          struct vring_used __user *used)
1145
1146 {
1147         size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
1148
1149         return access_ok(VERIFY_READ, desc, num * sizeof *desc) &&
1150                access_ok(VERIFY_READ, avail,
1151                          sizeof *avail + num * sizeof *avail->ring + s) &&
1152                access_ok(VERIFY_WRITE, used,
1153                         sizeof *used + num * sizeof *used->ring + s);
1154 }
1155
1156 static void vhost_vq_meta_update(struct vhost_virtqueue *vq,
1157                                  const struct vhost_umem_node *node,
1158                                  int type)
1159 {
1160         int access = (type == VHOST_ADDR_USED) ?
1161                      VHOST_ACCESS_WO : VHOST_ACCESS_RO;
1162
1163         if (likely(node->perm & access))
1164                 vq->meta_iotlb[type] = node;
1165 }
1166
1167 static bool iotlb_access_ok(struct vhost_virtqueue *vq,
1168                             int access, u64 addr, u64 len, int type)
1169 {
1170         const struct vhost_umem_node *node;
1171         struct vhost_umem *umem = vq->iotlb;
1172         u64 s = 0, size, orig_addr = addr, last = addr + len - 1;
1173
1174         if (vhost_vq_meta_fetch(vq, addr, len, type))
1175                 return true;
1176
1177         while (len > s) {
1178                 node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1179                                                            addr,
1180                                                            last);
1181                 if (node == NULL || node->start > addr) {
1182                         vhost_iotlb_miss(vq, addr, access);
1183                         return false;
1184                 } else if (!(node->perm & access)) {
1185                         /* Report the possible access violation by
1186                          * request another translation from userspace.
1187                          */
1188                         return false;
1189                 }
1190
1191                 size = node->size - addr + node->start;
1192
1193                 if (orig_addr == addr && size >= len)
1194                         vhost_vq_meta_update(vq, node, type);
1195
1196                 s += size;
1197                 addr += size;
1198         }
1199
1200         return true;
1201 }
1202
1203 int vq_iotlb_prefetch(struct vhost_virtqueue *vq)
1204 {
1205         size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
1206         unsigned int num = vq->num;
1207
1208         if (!vq->iotlb)
1209                 return 1;
1210
1211         return iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->desc,
1212                                num * sizeof(*vq->desc), VHOST_ADDR_DESC) &&
1213                iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->avail,
1214                                sizeof *vq->avail +
1215                                num * sizeof(*vq->avail->ring) + s,
1216                                VHOST_ADDR_AVAIL) &&
1217                iotlb_access_ok(vq, VHOST_ACCESS_WO, (u64)(uintptr_t)vq->used,
1218                                sizeof *vq->used +
1219                                num * sizeof(*vq->used->ring) + s,
1220                                VHOST_ADDR_USED);
1221 }
1222 EXPORT_SYMBOL_GPL(vq_iotlb_prefetch);
1223
1224 /* Can we log writes? */
1225 /* Caller should have device mutex but not vq mutex */
1226 bool vhost_log_access_ok(struct vhost_dev *dev)
1227 {
1228         return memory_access_ok(dev, dev->umem, 1);
1229 }
1230 EXPORT_SYMBOL_GPL(vhost_log_access_ok);
1231
1232 /* Verify access for write logging. */
1233 /* Caller should have vq mutex and device mutex */
1234 static bool vq_log_access_ok(struct vhost_virtqueue *vq,
1235                              void __user *log_base)
1236 {
1237         size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
1238
1239         return vq_memory_access_ok(log_base, vq->umem,
1240                                    vhost_has_feature(vq, VHOST_F_LOG_ALL)) &&
1241                 (!vq->log_used || log_access_ok(log_base, vq->log_addr,
1242                                         sizeof *vq->used +
1243                                         vq->num * sizeof *vq->used->ring + s));
1244 }
1245
1246 /* Can we start vq? */
1247 /* Caller should have vq mutex and device mutex */
1248 bool vhost_vq_access_ok(struct vhost_virtqueue *vq)
1249 {
1250         if (!vq_log_access_ok(vq, vq->log_base))
1251                 return false;
1252
1253         /* Access validation occurs at prefetch time with IOTLB */
1254         if (vq->iotlb)
1255                 return true;
1256
1257         return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used);
1258 }
1259 EXPORT_SYMBOL_GPL(vhost_vq_access_ok);
1260
1261 static struct vhost_umem *vhost_umem_alloc(void)
1262 {
1263         struct vhost_umem *umem = kvzalloc(sizeof(*umem), GFP_KERNEL);
1264
1265         if (!umem)
1266                 return NULL;
1267
1268         umem->umem_tree = RB_ROOT_CACHED;
1269         umem->numem = 0;
1270         INIT_LIST_HEAD(&umem->umem_list);
1271
1272         return umem;
1273 }
1274
1275 static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)
1276 {
1277         struct vhost_memory mem, *newmem;
1278         struct vhost_memory_region *region;
1279         struct vhost_umem *newumem, *oldumem;
1280         unsigned long size = offsetof(struct vhost_memory, regions);
1281         int i;
1282
1283         if (copy_from_user(&mem, m, size))
1284                 return -EFAULT;
1285         if (mem.padding)
1286                 return -EOPNOTSUPP;
1287         if (mem.nregions > max_mem_regions)
1288                 return -E2BIG;
1289         newmem = kvzalloc(size + mem.nregions * sizeof(*m->regions), GFP_KERNEL);
1290         if (!newmem)
1291                 return -ENOMEM;
1292
1293         memcpy(newmem, &mem, size);
1294         if (copy_from_user(newmem->regions, m->regions,
1295                            mem.nregions * sizeof *m->regions)) {
1296                 kvfree(newmem);
1297                 return -EFAULT;
1298         }
1299
1300         newumem = vhost_umem_alloc();
1301         if (!newumem) {
1302                 kvfree(newmem);
1303                 return -ENOMEM;
1304         }
1305
1306         for (region = newmem->regions;
1307              region < newmem->regions + mem.nregions;
1308              region++) {
1309                 if (vhost_new_umem_range(newumem,
1310                                          region->guest_phys_addr,
1311                                          region->memory_size,
1312                                          region->guest_phys_addr +
1313                                          region->memory_size - 1,
1314                                          region->userspace_addr,
1315                                          VHOST_ACCESS_RW))
1316                         goto err;
1317         }
1318
1319         if (!memory_access_ok(d, newumem, 0))
1320                 goto err;
1321
1322         oldumem = d->umem;
1323         d->umem = newumem;
1324
1325         /* All memory accesses are done under some VQ mutex. */
1326         for (i = 0; i < d->nvqs; ++i) {
1327                 mutex_lock(&d->vqs[i]->mutex);
1328                 d->vqs[i]->umem = newumem;
1329                 mutex_unlock(&d->vqs[i]->mutex);
1330         }
1331
1332         kvfree(newmem);
1333         vhost_umem_clean(oldumem);
1334         return 0;
1335
1336 err:
1337         vhost_umem_clean(newumem);
1338         kvfree(newmem);
1339         return -EFAULT;
1340 }
1341
1342 long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
1343 {
1344         struct file *eventfp, *filep = NULL;
1345         bool pollstart = false, pollstop = false;
1346         struct eventfd_ctx *ctx = NULL;
1347         u32 __user *idxp = argp;
1348         struct vhost_virtqueue *vq;
1349         struct vhost_vring_state s;
1350         struct vhost_vring_file f;
1351         struct vhost_vring_addr a;
1352         u32 idx;
1353         long r;
1354
1355         r = get_user(idx, idxp);
1356         if (r < 0)
1357                 return r;
1358         if (idx >= d->nvqs)
1359                 return -ENOBUFS;
1360
1361         vq = d->vqs[idx];
1362
1363         mutex_lock(&vq->mutex);
1364
1365         switch (ioctl) {
1366         case VHOST_SET_VRING_NUM:
1367                 /* Resizing ring with an active backend?
1368                  * You don't want to do that. */
1369                 if (vq->private_data) {
1370                         r = -EBUSY;
1371                         break;
1372                 }
1373                 if (copy_from_user(&s, argp, sizeof s)) {
1374                         r = -EFAULT;
1375                         break;
1376                 }
1377                 if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) {
1378                         r = -EINVAL;
1379                         break;
1380                 }
1381                 vq->num = s.num;
1382                 break;
1383         case VHOST_SET_VRING_BASE:
1384                 /* Moving base with an active backend?
1385                  * You don't want to do that. */
1386                 if (vq->private_data) {
1387                         r = -EBUSY;
1388                         break;
1389                 }
1390                 if (copy_from_user(&s, argp, sizeof s)) {
1391                         r = -EFAULT;
1392                         break;
1393                 }
1394                 if (s.num > 0xffff) {
1395                         r = -EINVAL;
1396                         break;
1397                 }
1398                 vq->last_avail_idx = s.num;
1399                 /* Forget the cached index value. */
1400                 vq->avail_idx = vq->last_avail_idx;
1401                 break;
1402         case VHOST_GET_VRING_BASE:
1403                 s.index = idx;
1404                 s.num = vq->last_avail_idx;
1405                 if (copy_to_user(argp, &s, sizeof s))
1406                         r = -EFAULT;
1407                 break;
1408         case VHOST_SET_VRING_ADDR:
1409                 if (copy_from_user(&a, argp, sizeof a)) {
1410                         r = -EFAULT;
1411                         break;
1412                 }
1413                 if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) {
1414                         r = -EOPNOTSUPP;
1415                         break;
1416                 }
1417                 /* For 32bit, verify that the top 32bits of the user
1418                    data are set to zero. */
1419                 if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr ||
1420                     (u64)(unsigned long)a.used_user_addr != a.used_user_addr ||
1421                     (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) {
1422                         r = -EFAULT;
1423                         break;
1424                 }
1425
1426                 /* Make sure it's safe to cast pointers to vring types. */
1427                 BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE);
1428                 BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE);
1429                 if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) ||
1430                     (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) ||
1431                     (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1))) {
1432                         r = -EINVAL;
1433                         break;
1434                 }
1435
1436                 /* We only verify access here if backend is configured.
1437                  * If it is not, we don't as size might not have been setup.
1438                  * We will verify when backend is configured. */
1439                 if (vq->private_data) {
1440                         if (!vq_access_ok(vq, vq->num,
1441                                 (void __user *)(unsigned long)a.desc_user_addr,
1442                                 (void __user *)(unsigned long)a.avail_user_addr,
1443                                 (void __user *)(unsigned long)a.used_user_addr)) {
1444                                 r = -EINVAL;
1445                                 break;
1446                         }
1447
1448                         /* Also validate log access for used ring if enabled. */
1449                         if ((a.flags & (0x1 << VHOST_VRING_F_LOG)) &&
1450                             !log_access_ok(vq->log_base, a.log_guest_addr,
1451                                            sizeof *vq->used +
1452                                            vq->num * sizeof *vq->used->ring)) {
1453                                 r = -EINVAL;
1454                                 break;
1455                         }
1456                 }
1457
1458                 vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG));
1459                 vq->desc = (void __user *)(unsigned long)a.desc_user_addr;
1460                 vq->avail = (void __user *)(unsigned long)a.avail_user_addr;
1461                 vq->log_addr = a.log_guest_addr;
1462                 vq->used = (void __user *)(unsigned long)a.used_user_addr;
1463                 break;
1464         case VHOST_SET_VRING_KICK:
1465                 if (copy_from_user(&f, argp, sizeof f)) {
1466                         r = -EFAULT;
1467                         break;
1468                 }
1469                 eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd);
1470                 if (IS_ERR(eventfp)) {
1471                         r = PTR_ERR(eventfp);
1472                         break;
1473                 }
1474                 if (eventfp != vq->kick) {
1475                         pollstop = (filep = vq->kick) != NULL;
1476                         pollstart = (vq->kick = eventfp) != NULL;
1477                 } else
1478                         filep = eventfp;
1479                 break;
1480         case VHOST_SET_VRING_CALL:
1481                 if (copy_from_user(&f, argp, sizeof f)) {
1482                         r = -EFAULT;
1483                         break;
1484                 }
1485                 ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd);
1486                 if (IS_ERR(ctx)) {
1487                         r = PTR_ERR(ctx);
1488                         break;
1489                 }
1490                 swap(ctx, vq->call_ctx);
1491                 break;
1492         case VHOST_SET_VRING_ERR:
1493                 if (copy_from_user(&f, argp, sizeof f)) {
1494                         r = -EFAULT;
1495                         break;
1496                 }
1497                 ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd);
1498                 if (IS_ERR(ctx)) {
1499                         r = PTR_ERR(ctx);
1500                         break;
1501                 }
1502                 swap(ctx, vq->error_ctx);
1503                 break;
1504         case VHOST_SET_VRING_ENDIAN:
1505                 r = vhost_set_vring_endian(vq, argp);
1506                 break;
1507         case VHOST_GET_VRING_ENDIAN:
1508                 r = vhost_get_vring_endian(vq, idx, argp);
1509                 break;
1510         case VHOST_SET_VRING_BUSYLOOP_TIMEOUT:
1511                 if (copy_from_user(&s, argp, sizeof(s))) {
1512                         r = -EFAULT;
1513                         break;
1514                 }
1515                 vq->busyloop_timeout = s.num;
1516                 break;
1517         case VHOST_GET_VRING_BUSYLOOP_TIMEOUT:
1518                 s.index = idx;
1519                 s.num = vq->busyloop_timeout;
1520                 if (copy_to_user(argp, &s, sizeof(s)))
1521                         r = -EFAULT;
1522                 break;
1523         default:
1524                 r = -ENOIOCTLCMD;
1525         }
1526
1527         if (pollstop && vq->handle_kick)
1528                 vhost_poll_stop(&vq->poll);
1529
1530         if (!IS_ERR_OR_NULL(ctx))
1531                 eventfd_ctx_put(ctx);
1532         if (filep)
1533                 fput(filep);
1534
1535         if (pollstart && vq->handle_kick)
1536                 r = vhost_poll_start(&vq->poll, vq->kick);
1537
1538         mutex_unlock(&vq->mutex);
1539
1540         if (pollstop && vq->handle_kick)
1541                 vhost_poll_flush(&vq->poll);
1542         return r;
1543 }
1544 EXPORT_SYMBOL_GPL(vhost_vring_ioctl);
1545
1546 int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled)
1547 {
1548         struct vhost_umem *niotlb, *oiotlb;
1549         int i;
1550
1551         niotlb = vhost_umem_alloc();
1552         if (!niotlb)
1553                 return -ENOMEM;
1554
1555         oiotlb = d->iotlb;
1556         d->iotlb = niotlb;
1557
1558         for (i = 0; i < d->nvqs; ++i) {
1559                 mutex_lock(&d->vqs[i]->mutex);
1560                 d->vqs[i]->iotlb = niotlb;
1561                 mutex_unlock(&d->vqs[i]->mutex);
1562         }
1563
1564         vhost_umem_clean(oiotlb);
1565
1566         return 0;
1567 }
1568 EXPORT_SYMBOL_GPL(vhost_init_device_iotlb);
1569
1570 /* Caller must have device mutex */
1571 long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
1572 {
1573         struct eventfd_ctx *ctx;
1574         u64 p;
1575         long r;
1576         int i, fd;
1577
1578         /* If you are not the owner, you can become one */
1579         if (ioctl == VHOST_SET_OWNER) {
1580                 r = vhost_dev_set_owner(d);
1581                 goto done;
1582         }
1583
1584         /* You must be the owner to do anything else */
1585         r = vhost_dev_check_owner(d);
1586         if (r)
1587                 goto done;
1588
1589         switch (ioctl) {
1590         case VHOST_SET_MEM_TABLE:
1591                 r = vhost_set_memory(d, argp);
1592                 break;
1593         case VHOST_SET_LOG_BASE:
1594                 if (copy_from_user(&p, argp, sizeof p)) {
1595                         r = -EFAULT;
1596                         break;
1597                 }
1598                 if ((u64)(unsigned long)p != p) {
1599                         r = -EFAULT;
1600                         break;
1601                 }
1602                 for (i = 0; i < d->nvqs; ++i) {
1603                         struct vhost_virtqueue *vq;
1604                         void __user *base = (void __user *)(unsigned long)p;
1605                         vq = d->vqs[i];
1606                         mutex_lock(&vq->mutex);
1607                         /* If ring is inactive, will check when it's enabled. */
1608                         if (vq->private_data && !vq_log_access_ok(vq, base))
1609                                 r = -EFAULT;
1610                         else
1611                                 vq->log_base = base;
1612                         mutex_unlock(&vq->mutex);
1613                 }
1614                 break;
1615         case VHOST_SET_LOG_FD:
1616                 r = get_user(fd, (int __user *)argp);
1617                 if (r < 0)
1618                         break;
1619                 ctx = fd == -1 ? NULL : eventfd_ctx_fdget(fd);
1620                 if (IS_ERR(ctx)) {
1621                         r = PTR_ERR(ctx);
1622                         break;
1623                 }
1624                 swap(ctx, d->log_ctx);
1625                 for (i = 0; i < d->nvqs; ++i) {
1626                         mutex_lock(&d->vqs[i]->mutex);
1627                         d->vqs[i]->log_ctx = d->log_ctx;
1628                         mutex_unlock(&d->vqs[i]->mutex);
1629                 }
1630                 if (ctx)
1631                         eventfd_ctx_put(ctx);
1632                 break;
1633         default:
1634                 r = -ENOIOCTLCMD;
1635                 break;
1636         }
1637 done:
1638         return r;
1639 }
1640 EXPORT_SYMBOL_GPL(vhost_dev_ioctl);
1641
1642 /* TODO: This is really inefficient.  We need something like get_user()
1643  * (instruction directly accesses the data, with an exception table entry
1644  * returning -EFAULT). See Documentation/x86/exception-tables.txt.
1645  */
1646 static int set_bit_to_user(int nr, void __user *addr)
1647 {
1648         unsigned long log = (unsigned long)addr;
1649         struct page *page;
1650         void *base;
1651         int bit = nr + (log % PAGE_SIZE) * 8;
1652         int r;
1653
1654         r = get_user_pages_fast(log, 1, 1, &page);
1655         if (r < 0)
1656                 return r;
1657         BUG_ON(r != 1);
1658         base = kmap_atomic(page);
1659         set_bit(bit, base);
1660         kunmap_atomic(base);
1661         set_page_dirty_lock(page);
1662         put_page(page);
1663         return 0;
1664 }
1665
1666 static int log_write(void __user *log_base,
1667                      u64 write_address, u64 write_length)
1668 {
1669         u64 write_page = write_address / VHOST_PAGE_SIZE;
1670         int r;
1671
1672         if (!write_length)
1673                 return 0;
1674         write_length += write_address % VHOST_PAGE_SIZE;
1675         for (;;) {
1676                 u64 base = (u64)(unsigned long)log_base;
1677                 u64 log = base + write_page / 8;
1678                 int bit = write_page % 8;
1679                 if ((u64)(unsigned long)log != log)
1680                         return -EFAULT;
1681                 r = set_bit_to_user(bit, (void __user *)(unsigned long)log);
1682                 if (r < 0)
1683                         return r;
1684                 if (write_length <= VHOST_PAGE_SIZE)
1685                         break;
1686                 write_length -= VHOST_PAGE_SIZE;
1687                 write_page += 1;
1688         }
1689         return r;
1690 }
1691
1692 int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
1693                     unsigned int log_num, u64 len)
1694 {
1695         int i, r;
1696
1697         /* Make sure data written is seen before log. */
1698         smp_wmb();
1699         for (i = 0; i < log_num; ++i) {
1700                 u64 l = min(log[i].len, len);
1701                 r = log_write(vq->log_base, log[i].addr, l);
1702                 if (r < 0)
1703                         return r;
1704                 len -= l;
1705                 if (!len) {
1706                         if (vq->log_ctx)
1707                                 eventfd_signal(vq->log_ctx, 1);
1708                         return 0;
1709                 }
1710         }
1711         /* Length written exceeds what we have stored. This is a bug. */
1712         BUG();
1713         return 0;
1714 }
1715 EXPORT_SYMBOL_GPL(vhost_log_write);
1716
1717 static int vhost_update_used_flags(struct vhost_virtqueue *vq)
1718 {
1719         void __user *used;
1720         if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags),
1721                            &vq->used->flags) < 0)
1722                 return -EFAULT;
1723         if (unlikely(vq->log_used)) {
1724                 /* Make sure the flag is seen before log. */
1725                 smp_wmb();
1726                 /* Log used flag write. */
1727                 used = &vq->used->flags;
1728                 log_write(vq->log_base, vq->log_addr +
1729                           (used - (void __user *)vq->used),
1730                           sizeof vq->used->flags);
1731                 if (vq->log_ctx)
1732                         eventfd_signal(vq->log_ctx, 1);
1733         }
1734         return 0;
1735 }
1736
1737 static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event)
1738 {
1739         if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx),
1740                            vhost_avail_event(vq)))
1741                 return -EFAULT;
1742         if (unlikely(vq->log_used)) {
1743                 void __user *used;
1744                 /* Make sure the event is seen before log. */
1745                 smp_wmb();
1746                 /* Log avail event write */
1747                 used = vhost_avail_event(vq);
1748                 log_write(vq->log_base, vq->log_addr +
1749                           (used - (void __user *)vq->used),
1750                           sizeof *vhost_avail_event(vq));
1751                 if (vq->log_ctx)
1752                         eventfd_signal(vq->log_ctx, 1);
1753         }
1754         return 0;
1755 }
1756
1757 int vhost_vq_init_access(struct vhost_virtqueue *vq)
1758 {
1759         __virtio16 last_used_idx;
1760         int r;
1761         bool is_le = vq->is_le;
1762
1763         if (!vq->private_data)
1764                 return 0;
1765
1766         vhost_init_is_le(vq);
1767
1768         r = vhost_update_used_flags(vq);
1769         if (r)
1770                 goto err;
1771         vq->signalled_used_valid = false;
1772         if (!vq->iotlb &&
1773             !access_ok(VERIFY_READ, &vq->used->idx, sizeof vq->used->idx)) {
1774                 r = -EFAULT;
1775                 goto err;
1776         }
1777         r = vhost_get_used(vq, last_used_idx, &vq->used->idx);
1778         if (r) {
1779                 vq_err(vq, "Can't access used idx at %p\n",
1780                        &vq->used->idx);
1781                 goto err;
1782         }
1783         vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx);
1784         return 0;
1785
1786 err:
1787         vq->is_le = is_le;
1788         return r;
1789 }
1790 EXPORT_SYMBOL_GPL(vhost_vq_init_access);
1791
1792 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
1793                           struct iovec iov[], int iov_size, int access)
1794 {
1795         const struct vhost_umem_node *node;
1796         struct vhost_dev *dev = vq->dev;
1797         struct vhost_umem *umem = dev->iotlb ? dev->iotlb : dev->umem;
1798         struct iovec *_iov;
1799         u64 s = 0;
1800         int ret = 0;
1801
1802         while ((u64)len > s) {
1803                 u64 size;
1804                 if (unlikely(ret >= iov_size)) {
1805                         ret = -ENOBUFS;
1806                         break;
1807                 }
1808
1809                 node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1810                                                         addr, addr + len - 1);
1811                 if (node == NULL || node->start > addr) {
1812                         if (umem != dev->iotlb) {
1813                                 ret = -EFAULT;
1814                                 break;
1815                         }
1816                         ret = -EAGAIN;
1817                         break;
1818                 } else if (!(node->perm & access)) {
1819                         ret = -EPERM;
1820                         break;
1821                 }
1822
1823                 _iov = iov + ret;
1824                 size = node->size - addr + node->start;
1825                 _iov->iov_len = min((u64)len - s, size);
1826                 _iov->iov_base = (void __user *)(unsigned long)
1827                         (node->userspace_addr + addr - node->start);
1828                 s += size;
1829                 addr += size;
1830                 ++ret;
1831         }
1832
1833         if (ret == -EAGAIN)
1834                 vhost_iotlb_miss(vq, addr, access);
1835         return ret;
1836 }
1837
1838 /* Each buffer in the virtqueues is actually a chain of descriptors.  This
1839  * function returns the next descriptor in the chain,
1840  * or -1U if we're at the end. */
1841 static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc)
1842 {
1843         unsigned int next;
1844
1845         /* If this descriptor says it doesn't chain, we're done. */
1846         if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT)))
1847                 return -1U;
1848
1849         /* Check they're not leading us off end of descriptors. */
1850         next = vhost16_to_cpu(vq, READ_ONCE(desc->next));
1851         return next;
1852 }
1853
1854 static int get_indirect(struct vhost_virtqueue *vq,
1855                         struct iovec iov[], unsigned int iov_size,
1856                         unsigned int *out_num, unsigned int *in_num,
1857                         struct vhost_log *log, unsigned int *log_num,
1858                         struct vring_desc *indirect)
1859 {
1860         struct vring_desc desc;
1861         unsigned int i = 0, count, found = 0;
1862         u32 len = vhost32_to_cpu(vq, indirect->len);
1863         struct iov_iter from;
1864         int ret, access;
1865
1866         /* Sanity check */
1867         if (unlikely(len % sizeof desc)) {
1868                 vq_err(vq, "Invalid length in indirect descriptor: "
1869                        "len 0x%llx not multiple of 0x%zx\n",
1870                        (unsigned long long)len,
1871                        sizeof desc);
1872                 return -EINVAL;
1873         }
1874
1875         ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect,
1876                              UIO_MAXIOV, VHOST_ACCESS_RO);
1877         if (unlikely(ret < 0)) {
1878                 if (ret != -EAGAIN)
1879                         vq_err(vq, "Translation failure %d in indirect.\n", ret);
1880                 return ret;
1881         }
1882         iov_iter_init(&from, READ, vq->indirect, ret, len);
1883
1884         /* We will use the result as an address to read from, so most
1885          * architectures only need a compiler barrier here. */
1886         read_barrier_depends();
1887
1888         count = len / sizeof desc;
1889         /* Buffers are chained via a 16 bit next field, so
1890          * we can have at most 2^16 of these. */
1891         if (unlikely(count > USHRT_MAX + 1)) {
1892                 vq_err(vq, "Indirect buffer length too big: %d\n",
1893                        indirect->len);
1894                 return -E2BIG;
1895         }
1896
1897         do {
1898                 unsigned iov_count = *in_num + *out_num;
1899                 if (unlikely(++found > count)) {
1900                         vq_err(vq, "Loop detected: last one at %u "
1901                                "indirect size %u\n",
1902                                i, count);
1903                         return -EINVAL;
1904                 }
1905                 if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) {
1906                         vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n",
1907                                i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
1908                         return -EINVAL;
1909                 }
1910                 if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) {
1911                         vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n",
1912                                i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
1913                         return -EINVAL;
1914                 }
1915
1916                 if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
1917                         access = VHOST_ACCESS_WO;
1918                 else
1919                         access = VHOST_ACCESS_RO;
1920
1921                 ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
1922                                      vhost32_to_cpu(vq, desc.len), iov + iov_count,
1923                                      iov_size - iov_count, access);
1924                 if (unlikely(ret < 0)) {
1925                         if (ret != -EAGAIN)
1926                                 vq_err(vq, "Translation failure %d indirect idx %d\n",
1927                                         ret, i);
1928                         return ret;
1929                 }
1930                 /* If this is an input descriptor, increment that count. */
1931                 if (access == VHOST_ACCESS_WO) {
1932                         *in_num += ret;
1933                         if (unlikely(log)) {
1934                                 log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
1935                                 log[*log_num].len = vhost32_to_cpu(vq, desc.len);
1936                                 ++*log_num;
1937                         }
1938                 } else {
1939                         /* If it's an output descriptor, they're all supposed
1940                          * to come before any input descriptors. */
1941                         if (unlikely(*in_num)) {
1942                                 vq_err(vq, "Indirect descriptor "
1943                                        "has out after in: idx %d\n", i);
1944                                 return -EINVAL;
1945                         }
1946                         *out_num += ret;
1947                 }
1948         } while ((i = next_desc(vq, &desc)) != -1);
1949         return 0;
1950 }
1951
1952 /* This looks in the virtqueue and for the first available buffer, and converts
1953  * it to an iovec for convenient access.  Since descriptors consist of some
1954  * number of output then some number of input descriptors, it's actually two
1955  * iovecs, but we pack them into one and note how many of each there were.
1956  *
1957  * This function returns the descriptor number found, or vq->num (which is
1958  * never a valid descriptor number) if none was found.  A negative code is
1959  * returned on error. */
1960 int vhost_get_vq_desc(struct vhost_virtqueue *vq,
1961                       struct iovec iov[], unsigned int iov_size,
1962                       unsigned int *out_num, unsigned int *in_num,
1963                       struct vhost_log *log, unsigned int *log_num)
1964 {
1965         struct vring_desc desc;
1966         unsigned int i, head, found = 0;
1967         u16 last_avail_idx;
1968         __virtio16 avail_idx;
1969         __virtio16 ring_head;
1970         int ret, access;
1971
1972         /* Check it isn't doing very strange things with descriptor numbers. */
1973         last_avail_idx = vq->last_avail_idx;
1974
1975         if (vq->avail_idx == vq->last_avail_idx) {
1976                 if (unlikely(vhost_get_avail(vq, avail_idx, &vq->avail->idx))) {
1977                         vq_err(vq, "Failed to access avail idx at %p\n",
1978                                 &vq->avail->idx);
1979                         return -EFAULT;
1980                 }
1981                 vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
1982
1983                 if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) {
1984                         vq_err(vq, "Guest moved used index from %u to %u",
1985                                 last_avail_idx, vq->avail_idx);
1986                         return -EFAULT;
1987                 }
1988
1989                 /* If there's nothing new since last we looked, return
1990                  * invalid.
1991                  */
1992                 if (vq->avail_idx == last_avail_idx)
1993                         return vq->num;
1994
1995                 /* Only get avail ring entries after they have been
1996                  * exposed by guest.
1997                  */
1998                 smp_rmb();
1999         }
2000
2001         /* Grab the next descriptor number they're advertising, and increment
2002          * the index we've seen. */
2003         if (unlikely(vhost_get_avail(vq, ring_head,
2004                      &vq->avail->ring[last_avail_idx & (vq->num - 1)]))) {
2005                 vq_err(vq, "Failed to read head: idx %d address %p\n",
2006                        last_avail_idx,
2007                        &vq->avail->ring[last_avail_idx % vq->num]);
2008                 return -EFAULT;
2009         }
2010
2011         head = vhost16_to_cpu(vq, ring_head);
2012
2013         /* If their number is silly, that's an error. */
2014         if (unlikely(head >= vq->num)) {
2015                 vq_err(vq, "Guest says index %u > %u is available",
2016                        head, vq->num);
2017                 return -EINVAL;
2018         }
2019
2020         /* When we start there are none of either input nor output. */
2021         *out_num = *in_num = 0;
2022         if (unlikely(log))
2023                 *log_num = 0;
2024
2025         i = head;
2026         do {
2027                 unsigned iov_count = *in_num + *out_num;
2028                 if (unlikely(i >= vq->num)) {
2029                         vq_err(vq, "Desc index is %u > %u, head = %u",
2030                                i, vq->num, head);
2031                         return -EINVAL;
2032                 }
2033                 if (unlikely(++found > vq->num)) {
2034                         vq_err(vq, "Loop detected: last one at %u "
2035                                "vq size %u head %u\n",
2036                                i, vq->num, head);
2037                         return -EINVAL;
2038                 }
2039                 ret = vhost_copy_from_user(vq, &desc, vq->desc + i,
2040                                            sizeof desc);
2041                 if (unlikely(ret)) {
2042                         vq_err(vq, "Failed to get descriptor: idx %d addr %p\n",
2043                                i, vq->desc + i);
2044                         return -EFAULT;
2045                 }
2046                 if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) {
2047                         ret = get_indirect(vq, iov, iov_size,
2048                                            out_num, in_num,
2049                                            log, log_num, &desc);
2050                         if (unlikely(ret < 0)) {
2051                                 if (ret != -EAGAIN)
2052                                         vq_err(vq, "Failure detected "
2053                                                 "in indirect descriptor at idx %d\n", i);
2054                                 return ret;
2055                         }
2056                         continue;
2057                 }
2058
2059                 if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2060                         access = VHOST_ACCESS_WO;
2061                 else
2062                         access = VHOST_ACCESS_RO;
2063                 ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2064                                      vhost32_to_cpu(vq, desc.len), iov + iov_count,
2065                                      iov_size - iov_count, access);
2066                 if (unlikely(ret < 0)) {
2067                         if (ret != -EAGAIN)
2068                                 vq_err(vq, "Translation failure %d descriptor idx %d\n",
2069                                         ret, i);
2070                         return ret;
2071                 }
2072                 if (access == VHOST_ACCESS_WO) {
2073                         /* If this is an input descriptor,
2074                          * increment that count. */
2075                         *in_num += ret;
2076                         if (unlikely(log)) {
2077                                 log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2078                                 log[*log_num].len = vhost32_to_cpu(vq, desc.len);
2079                                 ++*log_num;
2080                         }
2081                 } else {
2082                         /* If it's an output descriptor, they're all supposed
2083                          * to come before any input descriptors. */
2084                         if (unlikely(*in_num)) {
2085                                 vq_err(vq, "Descriptor has out after in: "
2086                                        "idx %d\n", i);
2087                                 return -EINVAL;
2088                         }
2089                         *out_num += ret;
2090                 }
2091         } while ((i = next_desc(vq, &desc)) != -1);
2092
2093         /* On success, increment avail index. */
2094         vq->last_avail_idx++;
2095
2096         /* Assume notifications from guest are disabled at this point,
2097          * if they aren't we would need to update avail_event index. */
2098         BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY));
2099         return head;
2100 }
2101 EXPORT_SYMBOL_GPL(vhost_get_vq_desc);
2102
2103 /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */
2104 void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n)
2105 {
2106         vq->last_avail_idx -= n;
2107 }
2108 EXPORT_SYMBOL_GPL(vhost_discard_vq_desc);
2109
2110 /* After we've used one of their buffers, we tell them about it.  We'll then
2111  * want to notify the guest, using eventfd. */
2112 int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len)
2113 {
2114         struct vring_used_elem heads = {
2115                 cpu_to_vhost32(vq, head),
2116                 cpu_to_vhost32(vq, len)
2117         };
2118
2119         return vhost_add_used_n(vq, &heads, 1);
2120 }
2121 EXPORT_SYMBOL_GPL(vhost_add_used);
2122
2123 static int __vhost_add_used_n(struct vhost_virtqueue *vq,
2124                             struct vring_used_elem *heads,
2125                             unsigned count)
2126 {
2127         struct vring_used_elem __user *used;
2128         u16 old, new;
2129         int start;
2130
2131         start = vq->last_used_idx & (vq->num - 1);
2132         used = vq->used->ring + start;
2133         if (count == 1) {
2134                 if (vhost_put_user(vq, heads[0].id, &used->id)) {
2135                         vq_err(vq, "Failed to write used id");
2136                         return -EFAULT;
2137                 }
2138                 if (vhost_put_user(vq, heads[0].len, &used->len)) {
2139                         vq_err(vq, "Failed to write used len");
2140                         return -EFAULT;
2141                 }
2142         } else if (vhost_copy_to_user(vq, used, heads, count * sizeof *used)) {
2143                 vq_err(vq, "Failed to write used");
2144                 return -EFAULT;
2145         }
2146         if (unlikely(vq->log_used)) {
2147                 /* Make sure data is seen before log. */
2148                 smp_wmb();
2149                 /* Log used ring entry write. */
2150                 log_write(vq->log_base,
2151                           vq->log_addr +
2152                            ((void __user *)used - (void __user *)vq->used),
2153                           count * sizeof *used);
2154         }
2155         old = vq->last_used_idx;
2156         new = (vq->last_used_idx += count);
2157         /* If the driver never bothers to signal in a very long while,
2158          * used index might wrap around. If that happens, invalidate
2159          * signalled_used index we stored. TODO: make sure driver
2160          * signals at least once in 2^16 and remove this. */
2161         if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old)))
2162                 vq->signalled_used_valid = false;
2163         return 0;
2164 }
2165
2166 /* After we've used one of their buffers, we tell them about it.  We'll then
2167  * want to notify the guest, using eventfd. */
2168 int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads,
2169                      unsigned count)
2170 {
2171         int start, n, r;
2172
2173         start = vq->last_used_idx & (vq->num - 1);
2174         n = vq->num - start;
2175         if (n < count) {
2176                 r = __vhost_add_used_n(vq, heads, n);
2177                 if (r < 0)
2178                         return r;
2179                 heads += n;
2180                 count -= n;
2181         }
2182         r = __vhost_add_used_n(vq, heads, count);
2183
2184         /* Make sure buffer is written before we update index. */
2185         smp_wmb();
2186         if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx),
2187                            &vq->used->idx)) {
2188                 vq_err(vq, "Failed to increment used idx");
2189                 return -EFAULT;
2190         }
2191         if (unlikely(vq->log_used)) {
2192                 /* Log used index update. */
2193                 log_write(vq->log_base,
2194                           vq->log_addr + offsetof(struct vring_used, idx),
2195                           sizeof vq->used->idx);
2196                 if (vq->log_ctx)
2197                         eventfd_signal(vq->log_ctx, 1);
2198         }
2199         return r;
2200 }
2201 EXPORT_SYMBOL_GPL(vhost_add_used_n);
2202
2203 static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2204 {
2205         __u16 old, new;
2206         __virtio16 event;
2207         bool v;
2208         /* Flush out used index updates. This is paired
2209          * with the barrier that the Guest executes when enabling
2210          * interrupts. */
2211         smp_mb();
2212
2213         if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2214             unlikely(vq->avail_idx == vq->last_avail_idx))
2215                 return true;
2216
2217         if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2218                 __virtio16 flags;
2219                 if (vhost_get_avail(vq, flags, &vq->avail->flags)) {
2220                         vq_err(vq, "Failed to get flags");
2221                         return true;
2222                 }
2223                 return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT));
2224         }
2225         old = vq->signalled_used;
2226         v = vq->signalled_used_valid;
2227         new = vq->signalled_used = vq->last_used_idx;
2228         vq->signalled_used_valid = true;
2229
2230         if (unlikely(!v))
2231                 return true;
2232
2233         if (vhost_get_avail(vq, event, vhost_used_event(vq))) {
2234                 vq_err(vq, "Failed to get used event idx");
2235                 return true;
2236         }
2237         return vring_need_event(vhost16_to_cpu(vq, event), new, old);
2238 }
2239
2240 /* This actually signals the guest, using eventfd. */
2241 void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2242 {
2243         /* Signal the Guest tell them we used something up. */
2244         if (vq->call_ctx && vhost_notify(dev, vq))
2245                 eventfd_signal(vq->call_ctx, 1);
2246 }
2247 EXPORT_SYMBOL_GPL(vhost_signal);
2248
2249 /* And here's the combo meal deal.  Supersize me! */
2250 void vhost_add_used_and_signal(struct vhost_dev *dev,
2251                                struct vhost_virtqueue *vq,
2252                                unsigned int head, int len)
2253 {
2254         vhost_add_used(vq, head, len);
2255         vhost_signal(dev, vq);
2256 }
2257 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal);
2258
2259 /* multi-buffer version of vhost_add_used_and_signal */
2260 void vhost_add_used_and_signal_n(struct vhost_dev *dev,
2261                                  struct vhost_virtqueue *vq,
2262                                  struct vring_used_elem *heads, unsigned count)
2263 {
2264         vhost_add_used_n(vq, heads, count);
2265         vhost_signal(dev, vq);
2266 }
2267 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n);
2268
2269 /* return true if we're sure that avaiable ring is empty */
2270 bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2271 {
2272         __virtio16 avail_idx;
2273         int r;
2274
2275         if (vq->avail_idx != vq->last_avail_idx)
2276                 return false;
2277
2278         r = vhost_get_avail(vq, avail_idx, &vq->avail->idx);
2279         if (unlikely(r))
2280                 return false;
2281         vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
2282
2283         return vq->avail_idx == vq->last_avail_idx;
2284 }
2285 EXPORT_SYMBOL_GPL(vhost_vq_avail_empty);
2286
2287 /* OK, now we need to know about added descriptors. */
2288 bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2289 {
2290         __virtio16 avail_idx;
2291         int r;
2292
2293         if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY))
2294                 return false;
2295         vq->used_flags &= ~VRING_USED_F_NO_NOTIFY;
2296         if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2297                 r = vhost_update_used_flags(vq);
2298                 if (r) {
2299                         vq_err(vq, "Failed to enable notification at %p: %d\n",
2300                                &vq->used->flags, r);
2301                         return false;
2302                 }
2303         } else {
2304                 r = vhost_update_avail_event(vq, vq->avail_idx);
2305                 if (r) {
2306                         vq_err(vq, "Failed to update avail event index at %p: %d\n",
2307                                vhost_avail_event(vq), r);
2308                         return false;
2309                 }
2310         }
2311         /* They could have slipped one in as we were doing that: make
2312          * sure it's written, then check again. */
2313         smp_mb();
2314         r = vhost_get_avail(vq, avail_idx, &vq->avail->idx);
2315         if (r) {
2316                 vq_err(vq, "Failed to check avail idx at %p: %d\n",
2317                        &vq->avail->idx, r);
2318                 return false;
2319         }
2320
2321         return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx;
2322 }
2323 EXPORT_SYMBOL_GPL(vhost_enable_notify);
2324
2325 /* We don't need to be notified again. */
2326 void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2327 {
2328         int r;
2329
2330         if (vq->used_flags & VRING_USED_F_NO_NOTIFY)
2331                 return;
2332         vq->used_flags |= VRING_USED_F_NO_NOTIFY;
2333         if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2334                 r = vhost_update_used_flags(vq);
2335                 if (r)
2336                         vq_err(vq, "Failed to enable notification at %p: %d\n",
2337                                &vq->used->flags, r);
2338         }
2339 }
2340 EXPORT_SYMBOL_GPL(vhost_disable_notify);
2341
2342 /* Create a new message. */
2343 struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type)
2344 {
2345         struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL);
2346         if (!node)
2347                 return NULL;
2348
2349         /* Make sure all padding within the structure is initialized. */
2350         memset(&node->msg, 0, sizeof node->msg);
2351         node->vq = vq;
2352         node->msg.type = type;
2353         return node;
2354 }
2355 EXPORT_SYMBOL_GPL(vhost_new_msg);
2356
2357 void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head,
2358                        struct vhost_msg_node *node)
2359 {
2360         spin_lock(&dev->iotlb_lock);
2361         list_add_tail(&node->node, head);
2362         spin_unlock(&dev->iotlb_lock);
2363
2364         wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
2365 }
2366 EXPORT_SYMBOL_GPL(vhost_enqueue_msg);
2367
2368 struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
2369                                          struct list_head *head)
2370 {
2371         struct vhost_msg_node *node = NULL;
2372
2373         spin_lock(&dev->iotlb_lock);
2374         if (!list_empty(head)) {
2375                 node = list_first_entry(head, struct vhost_msg_node,
2376                                         node);
2377                 list_del(&node->node);
2378         }
2379         spin_unlock(&dev->iotlb_lock);
2380
2381         return node;
2382 }
2383 EXPORT_SYMBOL_GPL(vhost_dequeue_msg);
2384
2385
2386 static int __init vhost_init(void)
2387 {
2388         return 0;
2389 }
2390
2391 static void __exit vhost_exit(void)
2392 {
2393 }
2394
2395 module_init(vhost_init);
2396 module_exit(vhost_exit);
2397
2398 MODULE_VERSION("0.0.1");
2399 MODULE_LICENSE("GPL v2");
2400 MODULE_AUTHOR("Michael S. Tsirkin");
2401 MODULE_DESCRIPTION("Host kernel accelerator for virtio");