xref: /linux/drivers/xen/gntdev.c (revision 5e8c0fb6a95728b852d56c0a9244425d474670c0)
1 /******************************************************************************
2  * gntdev.c
3  *
4  * Device for accessing (in user-space) pages that have been granted by other
5  * domains.
6  *
7  * Copyright (c) 2006-2007, D G Murray.
8  *           (c) 2009 Gerd Hoffmann <kraxel@redhat.com>
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write to the Free Software
17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18  */
19 
20 #undef DEBUG
21 
22 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
23 
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/init.h>
27 #include <linux/miscdevice.h>
28 #include <linux/fs.h>
29 #include <linux/mm.h>
30 #include <linux/mman.h>
31 #include <linux/mmu_notifier.h>
32 #include <linux/types.h>
33 #include <linux/uaccess.h>
34 #include <linux/sched.h>
35 #include <linux/spinlock.h>
36 #include <linux/slab.h>
37 #include <linux/highmem.h>
38 
39 #include <xen/xen.h>
40 #include <xen/grant_table.h>
41 #include <xen/balloon.h>
42 #include <xen/gntdev.h>
43 #include <xen/events.h>
44 #include <asm/xen/hypervisor.h>
45 #include <asm/xen/hypercall.h>
46 #include <asm/xen/page.h>
47 
48 MODULE_LICENSE("GPL");
49 MODULE_AUTHOR("Derek G. Murray <Derek.Murray@cl.cam.ac.uk>, "
50 	      "Gerd Hoffmann <kraxel@redhat.com>");
51 MODULE_DESCRIPTION("User-space granted page access driver");
52 
53 static int limit = 1024*1024;
54 module_param(limit, int, 0644);
55 MODULE_PARM_DESC(limit, "Maximum number of grants that may be mapped by "
56 		"the gntdev device");
57 
58 static atomic_t pages_mapped = ATOMIC_INIT(0);
59 
60 static int use_ptemod;
61 #define populate_freeable_maps use_ptemod
62 
63 struct gntdev_priv {
64 	/* maps with visible offsets in the file descriptor */
65 	struct list_head maps;
66 	/* maps that are not visible; will be freed on munmap.
67 	 * Only populated if populate_freeable_maps == 1 */
68 	struct list_head freeable_maps;
69 	/* lock protects maps and freeable_maps */
70 	spinlock_t lock;
71 	struct mm_struct *mm;
72 	struct mmu_notifier mn;
73 };
74 
75 struct unmap_notify {
76 	int flags;
77 	/* Address relative to the start of the grant_map */
78 	int addr;
79 	int event;
80 };
81 
82 struct grant_map {
83 	struct list_head next;
84 	struct vm_area_struct *vma;
85 	int index;
86 	int count;
87 	int flags;
88 	atomic_t users;
89 	struct unmap_notify notify;
90 	struct ioctl_gntdev_grant_ref *grants;
91 	struct gnttab_map_grant_ref   *map_ops;
92 	struct gnttab_unmap_grant_ref *unmap_ops;
93 	struct gnttab_map_grant_ref   *kmap_ops;
94 	struct page **pages;
95 };
96 
97 static int unmap_grant_pages(struct grant_map *map, int offset, int pages);
98 
99 /* ------------------------------------------------------------------ */
100 
101 static void gntdev_print_maps(struct gntdev_priv *priv,
102 			      char *text, int text_index)
103 {
104 #ifdef DEBUG
105 	struct grant_map *map;
106 
107 	pr_debug("%s: maps list (priv %p)\n", __func__, priv);
108 	list_for_each_entry(map, &priv->maps, next)
109 		pr_debug("  index %2d, count %2d %s\n",
110 		       map->index, map->count,
111 		       map->index == text_index && text ? text : "");
112 #endif
113 }
114 
115 static void gntdev_free_map(struct grant_map *map)
116 {
117 	if (map == NULL)
118 		return;
119 
120 	if (map->pages)
121 		free_xenballooned_pages(map->count, map->pages);
122 	kfree(map->pages);
123 	kfree(map->grants);
124 	kfree(map->map_ops);
125 	kfree(map->unmap_ops);
126 	kfree(map->kmap_ops);
127 	kfree(map);
128 }
129 
130 static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count)
131 {
132 	struct grant_map *add;
133 	int i;
134 
135 	add = kzalloc(sizeof(struct grant_map), GFP_KERNEL);
136 	if (NULL == add)
137 		return NULL;
138 
139 	add->grants    = kcalloc(count, sizeof(add->grants[0]), GFP_KERNEL);
140 	add->map_ops   = kcalloc(count, sizeof(add->map_ops[0]), GFP_KERNEL);
141 	add->unmap_ops = kcalloc(count, sizeof(add->unmap_ops[0]), GFP_KERNEL);
142 	add->kmap_ops  = kcalloc(count, sizeof(add->kmap_ops[0]), GFP_KERNEL);
143 	add->pages     = kcalloc(count, sizeof(add->pages[0]), GFP_KERNEL);
144 	if (NULL == add->grants    ||
145 	    NULL == add->map_ops   ||
146 	    NULL == add->unmap_ops ||
147 	    NULL == add->kmap_ops  ||
148 	    NULL == add->pages)
149 		goto err;
150 
151 	if (alloc_xenballooned_pages(count, add->pages, false /* lowmem */))
152 		goto err;
153 
154 	for (i = 0; i < count; i++) {
155 		add->map_ops[i].handle = -1;
156 		add->unmap_ops[i].handle = -1;
157 		add->kmap_ops[i].handle = -1;
158 	}
159 
160 	add->index = 0;
161 	add->count = count;
162 	atomic_set(&add->users, 1);
163 
164 	return add;
165 
166 err:
167 	gntdev_free_map(add);
168 	return NULL;
169 }
170 
171 static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add)
172 {
173 	struct grant_map *map;
174 
175 	list_for_each_entry(map, &priv->maps, next) {
176 		if (add->index + add->count < map->index) {
177 			list_add_tail(&add->next, &map->next);
178 			goto done;
179 		}
180 		add->index = map->index + map->count;
181 	}
182 	list_add_tail(&add->next, &priv->maps);
183 
184 done:
185 	gntdev_print_maps(priv, "[new]", add->index);
186 }
187 
188 static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
189 		int index, int count)
190 {
191 	struct grant_map *map;
192 
193 	list_for_each_entry(map, &priv->maps, next) {
194 		if (map->index != index)
195 			continue;
196 		if (count && map->count != count)
197 			continue;
198 		return map;
199 	}
200 	return NULL;
201 }
202 
203 static void gntdev_put_map(struct gntdev_priv *priv, struct grant_map *map)
204 {
205 	if (!map)
206 		return;
207 
208 	if (!atomic_dec_and_test(&map->users))
209 		return;
210 
211 	atomic_sub(map->count, &pages_mapped);
212 
213 	if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
214 		notify_remote_via_evtchn(map->notify.event);
215 		evtchn_put(map->notify.event);
216 	}
217 
218 	if (populate_freeable_maps && priv) {
219 		spin_lock(&priv->lock);
220 		list_del(&map->next);
221 		spin_unlock(&priv->lock);
222 	}
223 
224 	if (map->pages && !use_ptemod)
225 		unmap_grant_pages(map, 0, map->count);
226 	gntdev_free_map(map);
227 }
228 
229 /* ------------------------------------------------------------------ */
230 
231 static int find_grant_ptes(pte_t *pte, pgtable_t token,
232 		unsigned long addr, void *data)
233 {
234 	struct grant_map *map = data;
235 	unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT;
236 	int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte;
237 	u64 pte_maddr;
238 
239 	BUG_ON(pgnr >= map->count);
240 	pte_maddr = arbitrary_virt_to_machine(pte).maddr;
241 
242 	gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags,
243 			  map->grants[pgnr].ref,
244 			  map->grants[pgnr].domid);
245 	gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags,
246 			    -1 /* handle */);
247 	return 0;
248 }
249 
250 static int map_grant_pages(struct grant_map *map)
251 {
252 	int i, err = 0;
253 
254 	if (!use_ptemod) {
255 		/* Note: it could already be mapped */
256 		if (map->map_ops[0].handle != -1)
257 			return 0;
258 		for (i = 0; i < map->count; i++) {
259 			unsigned long addr = (unsigned long)
260 				pfn_to_kaddr(page_to_pfn(map->pages[i]));
261 			gnttab_set_map_op(&map->map_ops[i], addr, map->flags,
262 				map->grants[i].ref,
263 				map->grants[i].domid);
264 			gnttab_set_unmap_op(&map->unmap_ops[i], addr,
265 				map->flags, -1 /* handle */);
266 		}
267 	} else {
268 		/*
269 		 * Setup the map_ops corresponding to the pte entries pointing
270 		 * to the kernel linear addresses of the struct pages.
271 		 * These ptes are completely different from the user ptes dealt
272 		 * with find_grant_ptes.
273 		 */
274 		for (i = 0; i < map->count; i++) {
275 			unsigned long address = (unsigned long)
276 				pfn_to_kaddr(page_to_pfn(map->pages[i]));
277 			BUG_ON(PageHighMem(map->pages[i]));
278 
279 			gnttab_set_map_op(&map->kmap_ops[i], address,
280 				map->flags | GNTMAP_host_map,
281 				map->grants[i].ref,
282 				map->grants[i].domid);
283 		}
284 	}
285 
286 	pr_debug("map %d+%d\n", map->index, map->count);
287 	err = gnttab_map_refs(map->map_ops, use_ptemod ? map->kmap_ops : NULL,
288 			map->pages, map->count);
289 	if (err)
290 		return err;
291 
292 	for (i = 0; i < map->count; i++) {
293 		if (map->map_ops[i].status)
294 			err = -EINVAL;
295 		else {
296 			BUG_ON(map->map_ops[i].handle == -1);
297 			map->unmap_ops[i].handle = map->map_ops[i].handle;
298 			pr_debug("map handle=%d\n", map->map_ops[i].handle);
299 		}
300 	}
301 	return err;
302 }
303 
304 static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
305 {
306 	int i, err = 0;
307 
308 	if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) {
309 		int pgno = (map->notify.addr >> PAGE_SHIFT);
310 		if (pgno >= offset && pgno < offset + pages) {
311 			/* No need for kmap, pages are in lowmem */
312 			uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno]));
313 			tmp[map->notify.addr & (PAGE_SIZE-1)] = 0;
314 			map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
315 		}
316 	}
317 
318 	err = gnttab_unmap_refs(map->unmap_ops + offset,
319 			use_ptemod ? map->kmap_ops + offset : NULL, map->pages + offset,
320 			pages);
321 	if (err)
322 		return err;
323 
324 	for (i = 0; i < pages; i++) {
325 		if (map->unmap_ops[offset+i].status)
326 			err = -EINVAL;
327 		pr_debug("unmap handle=%d st=%d\n",
328 			map->unmap_ops[offset+i].handle,
329 			map->unmap_ops[offset+i].status);
330 		map->unmap_ops[offset+i].handle = -1;
331 	}
332 	return err;
333 }
334 
335 static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
336 {
337 	int range, err = 0;
338 
339 	pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
340 
341 	/* It is possible the requested range will have a "hole" where we
342 	 * already unmapped some of the grants. Only unmap valid ranges.
343 	 */
344 	while (pages && !err) {
345 		while (pages && map->unmap_ops[offset].handle == -1) {
346 			offset++;
347 			pages--;
348 		}
349 		range = 0;
350 		while (range < pages) {
351 			if (map->unmap_ops[offset+range].handle == -1) {
352 				range--;
353 				break;
354 			}
355 			range++;
356 		}
357 		err = __unmap_grant_pages(map, offset, range);
358 		offset += range;
359 		pages -= range;
360 	}
361 
362 	return err;
363 }
364 
365 /* ------------------------------------------------------------------ */
366 
367 static void gntdev_vma_open(struct vm_area_struct *vma)
368 {
369 	struct grant_map *map = vma->vm_private_data;
370 
371 	pr_debug("gntdev_vma_open %p\n", vma);
372 	atomic_inc(&map->users);
373 }
374 
375 static void gntdev_vma_close(struct vm_area_struct *vma)
376 {
377 	struct grant_map *map = vma->vm_private_data;
378 	struct file *file = vma->vm_file;
379 	struct gntdev_priv *priv = file->private_data;
380 
381 	pr_debug("gntdev_vma_close %p\n", vma);
382 	if (use_ptemod) {
383 		/* It is possible that an mmu notifier could be running
384 		 * concurrently, so take priv->lock to ensure that the vma won't
385 		 * vanishing during the unmap_grant_pages call, since we will
386 		 * spin here until that completes. Such a concurrent call will
387 		 * not do any unmapping, since that has been done prior to
388 		 * closing the vma, but it may still iterate the unmap_ops list.
389 		 */
390 		spin_lock(&priv->lock);
391 		map->vma = NULL;
392 		spin_unlock(&priv->lock);
393 	}
394 	vma->vm_private_data = NULL;
395 	gntdev_put_map(priv, map);
396 }
397 
398 static struct vm_operations_struct gntdev_vmops = {
399 	.open = gntdev_vma_open,
400 	.close = gntdev_vma_close,
401 };
402 
403 /* ------------------------------------------------------------------ */
404 
405 static void unmap_if_in_range(struct grant_map *map,
406 			      unsigned long start, unsigned long end)
407 {
408 	unsigned long mstart, mend;
409 	int err;
410 
411 	if (!map->vma)
412 		return;
413 	if (map->vma->vm_start >= end)
414 		return;
415 	if (map->vma->vm_end <= start)
416 		return;
417 	mstart = max(start, map->vma->vm_start);
418 	mend   = min(end,   map->vma->vm_end);
419 	pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
420 			map->index, map->count,
421 			map->vma->vm_start, map->vma->vm_end,
422 			start, end, mstart, mend);
423 	err = unmap_grant_pages(map,
424 				(mstart - map->vma->vm_start) >> PAGE_SHIFT,
425 				(mend - mstart) >> PAGE_SHIFT);
426 	WARN_ON(err);
427 }
428 
429 static void mn_invl_range_start(struct mmu_notifier *mn,
430 				struct mm_struct *mm,
431 				unsigned long start, unsigned long end)
432 {
433 	struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
434 	struct grant_map *map;
435 
436 	spin_lock(&priv->lock);
437 	list_for_each_entry(map, &priv->maps, next) {
438 		unmap_if_in_range(map, start, end);
439 	}
440 	list_for_each_entry(map, &priv->freeable_maps, next) {
441 		unmap_if_in_range(map, start, end);
442 	}
443 	spin_unlock(&priv->lock);
444 }
445 
446 static void mn_invl_page(struct mmu_notifier *mn,
447 			 struct mm_struct *mm,
448 			 unsigned long address)
449 {
450 	mn_invl_range_start(mn, mm, address, address + PAGE_SIZE);
451 }
452 
453 static void mn_release(struct mmu_notifier *mn,
454 		       struct mm_struct *mm)
455 {
456 	struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
457 	struct grant_map *map;
458 	int err;
459 
460 	spin_lock(&priv->lock);
461 	list_for_each_entry(map, &priv->maps, next) {
462 		if (!map->vma)
463 			continue;
464 		pr_debug("map %d+%d (%lx %lx)\n",
465 				map->index, map->count,
466 				map->vma->vm_start, map->vma->vm_end);
467 		err = unmap_grant_pages(map, /* offset */ 0, map->count);
468 		WARN_ON(err);
469 	}
470 	list_for_each_entry(map, &priv->freeable_maps, next) {
471 		if (!map->vma)
472 			continue;
473 		pr_debug("map %d+%d (%lx %lx)\n",
474 				map->index, map->count,
475 				map->vma->vm_start, map->vma->vm_end);
476 		err = unmap_grant_pages(map, /* offset */ 0, map->count);
477 		WARN_ON(err);
478 	}
479 	spin_unlock(&priv->lock);
480 }
481 
482 static struct mmu_notifier_ops gntdev_mmu_ops = {
483 	.release                = mn_release,
484 	.invalidate_page        = mn_invl_page,
485 	.invalidate_range_start = mn_invl_range_start,
486 };
487 
488 /* ------------------------------------------------------------------ */
489 
490 static int gntdev_open(struct inode *inode, struct file *flip)
491 {
492 	struct gntdev_priv *priv;
493 	int ret = 0;
494 
495 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
496 	if (!priv)
497 		return -ENOMEM;
498 
499 	INIT_LIST_HEAD(&priv->maps);
500 	INIT_LIST_HEAD(&priv->freeable_maps);
501 	spin_lock_init(&priv->lock);
502 
503 	if (use_ptemod) {
504 		priv->mm = get_task_mm(current);
505 		if (!priv->mm) {
506 			kfree(priv);
507 			return -ENOMEM;
508 		}
509 		priv->mn.ops = &gntdev_mmu_ops;
510 		ret = mmu_notifier_register(&priv->mn, priv->mm);
511 		mmput(priv->mm);
512 	}
513 
514 	if (ret) {
515 		kfree(priv);
516 		return ret;
517 	}
518 
519 	flip->private_data = priv;
520 	pr_debug("priv %p\n", priv);
521 
522 	return 0;
523 }
524 
525 static int gntdev_release(struct inode *inode, struct file *flip)
526 {
527 	struct gntdev_priv *priv = flip->private_data;
528 	struct grant_map *map;
529 
530 	pr_debug("priv %p\n", priv);
531 
532 	while (!list_empty(&priv->maps)) {
533 		map = list_entry(priv->maps.next, struct grant_map, next);
534 		list_del(&map->next);
535 		gntdev_put_map(NULL /* already removed */, map);
536 	}
537 	WARN_ON(!list_empty(&priv->freeable_maps));
538 
539 	if (use_ptemod)
540 		mmu_notifier_unregister(&priv->mn, priv->mm);
541 	kfree(priv);
542 	return 0;
543 }
544 
545 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
546 				       struct ioctl_gntdev_map_grant_ref __user *u)
547 {
548 	struct ioctl_gntdev_map_grant_ref op;
549 	struct grant_map *map;
550 	int err;
551 
552 	if (copy_from_user(&op, u, sizeof(op)) != 0)
553 		return -EFAULT;
554 	pr_debug("priv %p, add %d\n", priv, op.count);
555 	if (unlikely(op.count <= 0))
556 		return -EINVAL;
557 
558 	err = -ENOMEM;
559 	map = gntdev_alloc_map(priv, op.count);
560 	if (!map)
561 		return err;
562 
563 	if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) {
564 		pr_debug("can't map: over limit\n");
565 		gntdev_put_map(NULL, map);
566 		return err;
567 	}
568 
569 	if (copy_from_user(map->grants, &u->refs,
570 			   sizeof(map->grants[0]) * op.count) != 0) {
571 		gntdev_put_map(NULL, map);
572 		return -EFAULT;
573 	}
574 
575 	spin_lock(&priv->lock);
576 	gntdev_add_map(priv, map);
577 	op.index = map->index << PAGE_SHIFT;
578 	spin_unlock(&priv->lock);
579 
580 	if (copy_to_user(u, &op, sizeof(op)) != 0)
581 		return -EFAULT;
582 
583 	return 0;
584 }
585 
586 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
587 					 struct ioctl_gntdev_unmap_grant_ref __user *u)
588 {
589 	struct ioctl_gntdev_unmap_grant_ref op;
590 	struct grant_map *map;
591 	int err = -ENOENT;
592 
593 	if (copy_from_user(&op, u, sizeof(op)) != 0)
594 		return -EFAULT;
595 	pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
596 
597 	spin_lock(&priv->lock);
598 	map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
599 	if (map) {
600 		list_del(&map->next);
601 		if (populate_freeable_maps)
602 			list_add_tail(&map->next, &priv->freeable_maps);
603 		err = 0;
604 	}
605 	spin_unlock(&priv->lock);
606 	if (map)
607 		gntdev_put_map(priv, map);
608 	return err;
609 }
610 
611 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
612 					      struct ioctl_gntdev_get_offset_for_vaddr __user *u)
613 {
614 	struct ioctl_gntdev_get_offset_for_vaddr op;
615 	struct vm_area_struct *vma;
616 	struct grant_map *map;
617 	int rv = -EINVAL;
618 
619 	if (copy_from_user(&op, u, sizeof(op)) != 0)
620 		return -EFAULT;
621 	pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
622 
623 	down_read(&current->mm->mmap_sem);
624 	vma = find_vma(current->mm, op.vaddr);
625 	if (!vma || vma->vm_ops != &gntdev_vmops)
626 		goto out_unlock;
627 
628 	map = vma->vm_private_data;
629 	if (!map)
630 		goto out_unlock;
631 
632 	op.offset = map->index << PAGE_SHIFT;
633 	op.count = map->count;
634 	rv = 0;
635 
636  out_unlock:
637 	up_read(&current->mm->mmap_sem);
638 
639 	if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0)
640 		return -EFAULT;
641 	return rv;
642 }
643 
644 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
645 {
646 	struct ioctl_gntdev_unmap_notify op;
647 	struct grant_map *map;
648 	int rc;
649 	int out_flags;
650 	unsigned int out_event;
651 
652 	if (copy_from_user(&op, u, sizeof(op)))
653 		return -EFAULT;
654 
655 	if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
656 		return -EINVAL;
657 
658 	/* We need to grab a reference to the event channel we are going to use
659 	 * to send the notify before releasing the reference we may already have
660 	 * (if someone has called this ioctl twice). This is required so that
661 	 * it is possible to change the clear_byte part of the notification
662 	 * without disturbing the event channel part, which may now be the last
663 	 * reference to that event channel.
664 	 */
665 	if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
666 		if (evtchn_get(op.event_channel_port))
667 			return -EINVAL;
668 	}
669 
670 	out_flags = op.action;
671 	out_event = op.event_channel_port;
672 
673 	spin_lock(&priv->lock);
674 
675 	list_for_each_entry(map, &priv->maps, next) {
676 		uint64_t begin = map->index << PAGE_SHIFT;
677 		uint64_t end = (map->index + map->count) << PAGE_SHIFT;
678 		if (op.index >= begin && op.index < end)
679 			goto found;
680 	}
681 	rc = -ENOENT;
682 	goto unlock_out;
683 
684  found:
685 	if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
686 			(map->flags & GNTMAP_readonly)) {
687 		rc = -EINVAL;
688 		goto unlock_out;
689 	}
690 
691 	out_flags = map->notify.flags;
692 	out_event = map->notify.event;
693 
694 	map->notify.flags = op.action;
695 	map->notify.addr = op.index - (map->index << PAGE_SHIFT);
696 	map->notify.event = op.event_channel_port;
697 
698 	rc = 0;
699 
700  unlock_out:
701 	spin_unlock(&priv->lock);
702 
703 	/* Drop the reference to the event channel we did not save in the map */
704 	if (out_flags & UNMAP_NOTIFY_SEND_EVENT)
705 		evtchn_put(out_event);
706 
707 	return rc;
708 }
709 
710 static long gntdev_ioctl(struct file *flip,
711 			 unsigned int cmd, unsigned long arg)
712 {
713 	struct gntdev_priv *priv = flip->private_data;
714 	void __user *ptr = (void __user *)arg;
715 
716 	switch (cmd) {
717 	case IOCTL_GNTDEV_MAP_GRANT_REF:
718 		return gntdev_ioctl_map_grant_ref(priv, ptr);
719 
720 	case IOCTL_GNTDEV_UNMAP_GRANT_REF:
721 		return gntdev_ioctl_unmap_grant_ref(priv, ptr);
722 
723 	case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
724 		return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
725 
726 	case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
727 		return gntdev_ioctl_notify(priv, ptr);
728 
729 	default:
730 		pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
731 		return -ENOIOCTLCMD;
732 	}
733 
734 	return 0;
735 }
736 
737 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
738 {
739 	struct gntdev_priv *priv = flip->private_data;
740 	int index = vma->vm_pgoff;
741 	int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
742 	struct grant_map *map;
743 	int i, err = -EINVAL;
744 
745 	if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
746 		return -EINVAL;
747 
748 	pr_debug("map %d+%d at %lx (pgoff %lx)\n",
749 			index, count, vma->vm_start, vma->vm_pgoff);
750 
751 	spin_lock(&priv->lock);
752 	map = gntdev_find_map_index(priv, index, count);
753 	if (!map)
754 		goto unlock_out;
755 	if (use_ptemod && map->vma)
756 		goto unlock_out;
757 	if (use_ptemod && priv->mm != vma->vm_mm) {
758 		pr_warn("Huh? Other mm?\n");
759 		goto unlock_out;
760 	}
761 
762 	atomic_inc(&map->users);
763 
764 	vma->vm_ops = &gntdev_vmops;
765 
766 	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
767 
768 	if (use_ptemod)
769 		vma->vm_flags |= VM_DONTCOPY;
770 
771 	vma->vm_private_data = map;
772 
773 	if (use_ptemod)
774 		map->vma = vma;
775 
776 	if (map->flags) {
777 		if ((vma->vm_flags & VM_WRITE) &&
778 				(map->flags & GNTMAP_readonly))
779 			goto out_unlock_put;
780 	} else {
781 		map->flags = GNTMAP_host_map;
782 		if (!(vma->vm_flags & VM_WRITE))
783 			map->flags |= GNTMAP_readonly;
784 	}
785 
786 	spin_unlock(&priv->lock);
787 
788 	if (use_ptemod) {
789 		err = apply_to_page_range(vma->vm_mm, vma->vm_start,
790 					  vma->vm_end - vma->vm_start,
791 					  find_grant_ptes, map);
792 		if (err) {
793 			pr_warn("find_grant_ptes() failure.\n");
794 			goto out_put_map;
795 		}
796 	}
797 
798 	err = map_grant_pages(map);
799 	if (err)
800 		goto out_put_map;
801 
802 	if (!use_ptemod) {
803 		for (i = 0; i < count; i++) {
804 			err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
805 				map->pages[i]);
806 			if (err)
807 				goto out_put_map;
808 		}
809 	}
810 
811 	return 0;
812 
813 unlock_out:
814 	spin_unlock(&priv->lock);
815 	return err;
816 
817 out_unlock_put:
818 	spin_unlock(&priv->lock);
819 out_put_map:
820 	if (use_ptemod)
821 		map->vma = NULL;
822 	gntdev_put_map(priv, map);
823 	return err;
824 }
825 
826 static const struct file_operations gntdev_fops = {
827 	.owner = THIS_MODULE,
828 	.open = gntdev_open,
829 	.release = gntdev_release,
830 	.mmap = gntdev_mmap,
831 	.unlocked_ioctl = gntdev_ioctl
832 };
833 
834 static struct miscdevice gntdev_miscdev = {
835 	.minor        = MISC_DYNAMIC_MINOR,
836 	.name         = "xen/gntdev",
837 	.fops         = &gntdev_fops,
838 };
839 
840 /* ------------------------------------------------------------------ */
841 
842 static int __init gntdev_init(void)
843 {
844 	int err;
845 
846 	if (!xen_domain())
847 		return -ENODEV;
848 
849 	use_ptemod = !xen_feature(XENFEAT_auto_translated_physmap);
850 
851 	err = misc_register(&gntdev_miscdev);
852 	if (err != 0) {
853 		pr_err("Could not register gntdev device\n");
854 		return err;
855 	}
856 	return 0;
857 }
858 
859 static void __exit gntdev_exit(void)
860 {
861 	misc_deregister(&gntdev_miscdev);
862 }
863 
864 module_init(gntdev_init);
865 module_exit(gntdev_exit);
866 
867 /* ------------------------------------------------------------------ */
868