xref: /openbmc/linux/drivers/xen/gntdev.c (revision 95e9fd10)
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 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/init.h>
25 #include <linux/miscdevice.h>
26 #include <linux/fs.h>
27 #include <linux/mm.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/types.h>
31 #include <linux/uaccess.h>
32 #include <linux/sched.h>
33 #include <linux/spinlock.h>
34 #include <linux/slab.h>
35 #include <linux/highmem.h>
36 
37 #include <xen/xen.h>
38 #include <xen/grant_table.h>
39 #include <xen/balloon.h>
40 #include <xen/gntdev.h>
41 #include <xen/events.h>
42 #include <asm/xen/hypervisor.h>
43 #include <asm/xen/hypercall.h>
44 #include <asm/xen/page.h>
45 
46 MODULE_LICENSE("GPL");
47 MODULE_AUTHOR("Derek G. Murray <Derek.Murray@cl.cam.ac.uk>, "
48 	      "Gerd Hoffmann <kraxel@redhat.com>");
49 MODULE_DESCRIPTION("User-space granted page access driver");
50 
51 static int limit = 1024*1024;
52 module_param(limit, int, 0644);
53 MODULE_PARM_DESC(limit, "Maximum number of grants that may be mapped by "
54 		"the gntdev device");
55 
56 static atomic_t pages_mapped = ATOMIC_INIT(0);
57 
58 static int use_ptemod;
59 
60 struct gntdev_priv {
61 	struct list_head maps;
62 	/* lock protects maps from concurrent changes */
63 	spinlock_t lock;
64 	struct mm_struct *mm;
65 	struct mmu_notifier mn;
66 };
67 
68 struct unmap_notify {
69 	int flags;
70 	/* Address relative to the start of the grant_map */
71 	int addr;
72 	int event;
73 };
74 
75 struct grant_map {
76 	struct list_head next;
77 	struct vm_area_struct *vma;
78 	int index;
79 	int count;
80 	int flags;
81 	atomic_t users;
82 	struct unmap_notify notify;
83 	struct ioctl_gntdev_grant_ref *grants;
84 	struct gnttab_map_grant_ref   *map_ops;
85 	struct gnttab_unmap_grant_ref *unmap_ops;
86 	struct gnttab_map_grant_ref   *kmap_ops;
87 	struct page **pages;
88 };
89 
90 static int unmap_grant_pages(struct grant_map *map, int offset, int pages);
91 
92 /* ------------------------------------------------------------------ */
93 
94 static void gntdev_print_maps(struct gntdev_priv *priv,
95 			      char *text, int text_index)
96 {
97 #ifdef DEBUG
98 	struct grant_map *map;
99 
100 	pr_debug("%s: maps list (priv %p)\n", __func__, priv);
101 	list_for_each_entry(map, &priv->maps, next)
102 		pr_debug("  index %2d, count %2d %s\n",
103 		       map->index, map->count,
104 		       map->index == text_index && text ? text : "");
105 #endif
106 }
107 
108 static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count)
109 {
110 	struct grant_map *add;
111 	int i;
112 
113 	add = kzalloc(sizeof(struct grant_map), GFP_KERNEL);
114 	if (NULL == add)
115 		return NULL;
116 
117 	add->grants    = kcalloc(count, sizeof(add->grants[0]), GFP_KERNEL);
118 	add->map_ops   = kcalloc(count, sizeof(add->map_ops[0]), GFP_KERNEL);
119 	add->unmap_ops = kcalloc(count, sizeof(add->unmap_ops[0]), GFP_KERNEL);
120 	add->kmap_ops  = kcalloc(count, sizeof(add->kmap_ops[0]), GFP_KERNEL);
121 	add->pages     = kcalloc(count, sizeof(add->pages[0]), GFP_KERNEL);
122 	if (NULL == add->grants    ||
123 	    NULL == add->map_ops   ||
124 	    NULL == add->unmap_ops ||
125 	    NULL == add->kmap_ops  ||
126 	    NULL == add->pages)
127 		goto err;
128 
129 	if (alloc_xenballooned_pages(count, add->pages, false /* lowmem */))
130 		goto err;
131 
132 	for (i = 0; i < count; i++) {
133 		add->map_ops[i].handle = -1;
134 		add->unmap_ops[i].handle = -1;
135 		add->kmap_ops[i].handle = -1;
136 	}
137 
138 	add->index = 0;
139 	add->count = count;
140 	atomic_set(&add->users, 1);
141 
142 	return add;
143 
144 err:
145 	kfree(add->pages);
146 	kfree(add->grants);
147 	kfree(add->map_ops);
148 	kfree(add->unmap_ops);
149 	kfree(add->kmap_ops);
150 	kfree(add);
151 	return NULL;
152 }
153 
154 static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add)
155 {
156 	struct grant_map *map;
157 
158 	list_for_each_entry(map, &priv->maps, next) {
159 		if (add->index + add->count < map->index) {
160 			list_add_tail(&add->next, &map->next);
161 			goto done;
162 		}
163 		add->index = map->index + map->count;
164 	}
165 	list_add_tail(&add->next, &priv->maps);
166 
167 done:
168 	gntdev_print_maps(priv, "[new]", add->index);
169 }
170 
171 static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
172 		int index, int count)
173 {
174 	struct grant_map *map;
175 
176 	list_for_each_entry(map, &priv->maps, next) {
177 		if (map->index != index)
178 			continue;
179 		if (count && map->count != count)
180 			continue;
181 		return map;
182 	}
183 	return NULL;
184 }
185 
186 static void gntdev_put_map(struct grant_map *map)
187 {
188 	if (!map)
189 		return;
190 
191 	if (!atomic_dec_and_test(&map->users))
192 		return;
193 
194 	atomic_sub(map->count, &pages_mapped);
195 
196 	if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
197 		notify_remote_via_evtchn(map->notify.event);
198 		evtchn_put(map->notify.event);
199 	}
200 
201 	if (map->pages) {
202 		if (!use_ptemod)
203 			unmap_grant_pages(map, 0, map->count);
204 
205 		free_xenballooned_pages(map->count, map->pages);
206 	}
207 	kfree(map->pages);
208 	kfree(map->grants);
209 	kfree(map->map_ops);
210 	kfree(map->unmap_ops);
211 	kfree(map);
212 }
213 
214 /* ------------------------------------------------------------------ */
215 
216 static int find_grant_ptes(pte_t *pte, pgtable_t token,
217 		unsigned long addr, void *data)
218 {
219 	struct grant_map *map = data;
220 	unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT;
221 	int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte;
222 	u64 pte_maddr;
223 
224 	BUG_ON(pgnr >= map->count);
225 	pte_maddr = arbitrary_virt_to_machine(pte).maddr;
226 
227 	gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags,
228 			  map->grants[pgnr].ref,
229 			  map->grants[pgnr].domid);
230 	gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags,
231 			    -1 /* handle */);
232 	return 0;
233 }
234 
235 static int map_grant_pages(struct grant_map *map)
236 {
237 	int i, err = 0;
238 
239 	if (!use_ptemod) {
240 		/* Note: it could already be mapped */
241 		if (map->map_ops[0].handle != -1)
242 			return 0;
243 		for (i = 0; i < map->count; i++) {
244 			unsigned long addr = (unsigned long)
245 				pfn_to_kaddr(page_to_pfn(map->pages[i]));
246 			gnttab_set_map_op(&map->map_ops[i], addr, map->flags,
247 				map->grants[i].ref,
248 				map->grants[i].domid);
249 			gnttab_set_unmap_op(&map->unmap_ops[i], addr,
250 				map->flags, -1 /* handle */);
251 		}
252 	} else {
253 		/*
254 		 * Setup the map_ops corresponding to the pte entries pointing
255 		 * to the kernel linear addresses of the struct pages.
256 		 * These ptes are completely different from the user ptes dealt
257 		 * with find_grant_ptes.
258 		 */
259 		for (i = 0; i < map->count; i++) {
260 			unsigned level;
261 			unsigned long address = (unsigned long)
262 				pfn_to_kaddr(page_to_pfn(map->pages[i]));
263 			pte_t *ptep;
264 			u64 pte_maddr = 0;
265 			BUG_ON(PageHighMem(map->pages[i]));
266 
267 			ptep = lookup_address(address, &level);
268 			pte_maddr = arbitrary_virt_to_machine(ptep).maddr;
269 			gnttab_set_map_op(&map->kmap_ops[i], pte_maddr,
270 				map->flags |
271 				GNTMAP_host_map |
272 				GNTMAP_contains_pte,
273 				map->grants[i].ref,
274 				map->grants[i].domid);
275 		}
276 	}
277 
278 	pr_debug("map %d+%d\n", map->index, map->count);
279 	err = gnttab_map_refs(map->map_ops, use_ptemod ? map->kmap_ops : NULL,
280 			map->pages, map->count);
281 	if (err)
282 		return err;
283 
284 	for (i = 0; i < map->count; i++) {
285 		if (map->map_ops[i].status)
286 			err = -EINVAL;
287 		else {
288 			BUG_ON(map->map_ops[i].handle == -1);
289 			map->unmap_ops[i].handle = map->map_ops[i].handle;
290 			pr_debug("map handle=%d\n", map->map_ops[i].handle);
291 		}
292 	}
293 	return err;
294 }
295 
296 static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
297 {
298 	int i, err = 0;
299 
300 	if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) {
301 		int pgno = (map->notify.addr >> PAGE_SHIFT);
302 		if (pgno >= offset && pgno < offset + pages && use_ptemod) {
303 			void __user *tmp = (void __user *)
304 				map->vma->vm_start + map->notify.addr;
305 			err = copy_to_user(tmp, &err, 1);
306 			if (err)
307 				return -EFAULT;
308 			map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
309 		} else if (pgno >= offset && pgno < offset + pages) {
310 			uint8_t *tmp = kmap(map->pages[pgno]);
311 			tmp[map->notify.addr & (PAGE_SIZE-1)] = 0;
312 			kunmap(map->pages[pgno]);
313 			map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
314 		}
315 	}
316 
317 	err = gnttab_unmap_refs(map->unmap_ops + offset, map->pages + offset,
318 				pages, true);
319 	if (err)
320 		return err;
321 
322 	for (i = 0; i < pages; i++) {
323 		if (map->unmap_ops[offset+i].status)
324 			err = -EINVAL;
325 		pr_debug("unmap handle=%d st=%d\n",
326 			map->unmap_ops[offset+i].handle,
327 			map->unmap_ops[offset+i].status);
328 		map->unmap_ops[offset+i].handle = -1;
329 	}
330 	return err;
331 }
332 
333 static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
334 {
335 	int range, err = 0;
336 
337 	pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
338 
339 	/* It is possible the requested range will have a "hole" where we
340 	 * already unmapped some of the grants. Only unmap valid ranges.
341 	 */
342 	while (pages && !err) {
343 		while (pages && map->unmap_ops[offset].handle == -1) {
344 			offset++;
345 			pages--;
346 		}
347 		range = 0;
348 		while (range < pages) {
349 			if (map->unmap_ops[offset+range].handle == -1) {
350 				range--;
351 				break;
352 			}
353 			range++;
354 		}
355 		err = __unmap_grant_pages(map, offset, range);
356 		offset += range;
357 		pages -= range;
358 	}
359 
360 	return err;
361 }
362 
363 /* ------------------------------------------------------------------ */
364 
365 static void gntdev_vma_open(struct vm_area_struct *vma)
366 {
367 	struct grant_map *map = vma->vm_private_data;
368 
369 	pr_debug("gntdev_vma_open %p\n", vma);
370 	atomic_inc(&map->users);
371 }
372 
373 static void gntdev_vma_close(struct vm_area_struct *vma)
374 {
375 	struct grant_map *map = vma->vm_private_data;
376 
377 	pr_debug("gntdev_vma_close %p\n", vma);
378 	map->vma = NULL;
379 	vma->vm_private_data = NULL;
380 	gntdev_put_map(map);
381 }
382 
383 static struct vm_operations_struct gntdev_vmops = {
384 	.open = gntdev_vma_open,
385 	.close = gntdev_vma_close,
386 };
387 
388 /* ------------------------------------------------------------------ */
389 
390 static void mn_invl_range_start(struct mmu_notifier *mn,
391 				struct mm_struct *mm,
392 				unsigned long start, unsigned long end)
393 {
394 	struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
395 	struct grant_map *map;
396 	unsigned long mstart, mend;
397 	int err;
398 
399 	spin_lock(&priv->lock);
400 	list_for_each_entry(map, &priv->maps, next) {
401 		if (!map->vma)
402 			continue;
403 		if (map->vma->vm_start >= end)
404 			continue;
405 		if (map->vma->vm_end <= start)
406 			continue;
407 		mstart = max(start, map->vma->vm_start);
408 		mend   = min(end,   map->vma->vm_end);
409 		pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
410 				map->index, map->count,
411 				map->vma->vm_start, map->vma->vm_end,
412 				start, end, mstart, mend);
413 		err = unmap_grant_pages(map,
414 					(mstart - map->vma->vm_start) >> PAGE_SHIFT,
415 					(mend - mstart) >> PAGE_SHIFT);
416 		WARN_ON(err);
417 	}
418 	spin_unlock(&priv->lock);
419 }
420 
421 static void mn_invl_page(struct mmu_notifier *mn,
422 			 struct mm_struct *mm,
423 			 unsigned long address)
424 {
425 	mn_invl_range_start(mn, mm, address, address + PAGE_SIZE);
426 }
427 
428 static void mn_release(struct mmu_notifier *mn,
429 		       struct mm_struct *mm)
430 {
431 	struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
432 	struct grant_map *map;
433 	int err;
434 
435 	spin_lock(&priv->lock);
436 	list_for_each_entry(map, &priv->maps, next) {
437 		if (!map->vma)
438 			continue;
439 		pr_debug("map %d+%d (%lx %lx)\n",
440 				map->index, map->count,
441 				map->vma->vm_start, map->vma->vm_end);
442 		err = unmap_grant_pages(map, /* offset */ 0, map->count);
443 		WARN_ON(err);
444 	}
445 	spin_unlock(&priv->lock);
446 }
447 
448 struct mmu_notifier_ops gntdev_mmu_ops = {
449 	.release                = mn_release,
450 	.invalidate_page        = mn_invl_page,
451 	.invalidate_range_start = mn_invl_range_start,
452 };
453 
454 /* ------------------------------------------------------------------ */
455 
456 static int gntdev_open(struct inode *inode, struct file *flip)
457 {
458 	struct gntdev_priv *priv;
459 	int ret = 0;
460 
461 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
462 	if (!priv)
463 		return -ENOMEM;
464 
465 	INIT_LIST_HEAD(&priv->maps);
466 	spin_lock_init(&priv->lock);
467 
468 	if (use_ptemod) {
469 		priv->mm = get_task_mm(current);
470 		if (!priv->mm) {
471 			kfree(priv);
472 			return -ENOMEM;
473 		}
474 		priv->mn.ops = &gntdev_mmu_ops;
475 		ret = mmu_notifier_register(&priv->mn, priv->mm);
476 		mmput(priv->mm);
477 	}
478 
479 	if (ret) {
480 		kfree(priv);
481 		return ret;
482 	}
483 
484 	flip->private_data = priv;
485 	pr_debug("priv %p\n", priv);
486 
487 	return 0;
488 }
489 
490 static int gntdev_release(struct inode *inode, struct file *flip)
491 {
492 	struct gntdev_priv *priv = flip->private_data;
493 	struct grant_map *map;
494 
495 	pr_debug("priv %p\n", priv);
496 
497 	while (!list_empty(&priv->maps)) {
498 		map = list_entry(priv->maps.next, struct grant_map, next);
499 		list_del(&map->next);
500 		gntdev_put_map(map);
501 	}
502 
503 	if (use_ptemod)
504 		mmu_notifier_unregister(&priv->mn, priv->mm);
505 	kfree(priv);
506 	return 0;
507 }
508 
509 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
510 				       struct ioctl_gntdev_map_grant_ref __user *u)
511 {
512 	struct ioctl_gntdev_map_grant_ref op;
513 	struct grant_map *map;
514 	int err;
515 
516 	if (copy_from_user(&op, u, sizeof(op)) != 0)
517 		return -EFAULT;
518 	pr_debug("priv %p, add %d\n", priv, op.count);
519 	if (unlikely(op.count <= 0))
520 		return -EINVAL;
521 
522 	err = -ENOMEM;
523 	map = gntdev_alloc_map(priv, op.count);
524 	if (!map)
525 		return err;
526 
527 	if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) {
528 		pr_debug("can't map: over limit\n");
529 		gntdev_put_map(map);
530 		return err;
531 	}
532 
533 	if (copy_from_user(map->grants, &u->refs,
534 			   sizeof(map->grants[0]) * op.count) != 0) {
535 		gntdev_put_map(map);
536 		return err;
537 	}
538 
539 	spin_lock(&priv->lock);
540 	gntdev_add_map(priv, map);
541 	op.index = map->index << PAGE_SHIFT;
542 	spin_unlock(&priv->lock);
543 
544 	if (copy_to_user(u, &op, sizeof(op)) != 0)
545 		return -EFAULT;
546 
547 	return 0;
548 }
549 
550 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
551 					 struct ioctl_gntdev_unmap_grant_ref __user *u)
552 {
553 	struct ioctl_gntdev_unmap_grant_ref op;
554 	struct grant_map *map;
555 	int err = -ENOENT;
556 
557 	if (copy_from_user(&op, u, sizeof(op)) != 0)
558 		return -EFAULT;
559 	pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
560 
561 	spin_lock(&priv->lock);
562 	map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
563 	if (map) {
564 		list_del(&map->next);
565 		err = 0;
566 	}
567 	spin_unlock(&priv->lock);
568 	if (map)
569 		gntdev_put_map(map);
570 	return err;
571 }
572 
573 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
574 					      struct ioctl_gntdev_get_offset_for_vaddr __user *u)
575 {
576 	struct ioctl_gntdev_get_offset_for_vaddr op;
577 	struct vm_area_struct *vma;
578 	struct grant_map *map;
579 
580 	if (copy_from_user(&op, u, sizeof(op)) != 0)
581 		return -EFAULT;
582 	pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
583 
584 	vma = find_vma(current->mm, op.vaddr);
585 	if (!vma || vma->vm_ops != &gntdev_vmops)
586 		return -EINVAL;
587 
588 	map = vma->vm_private_data;
589 	if (!map)
590 		return -EINVAL;
591 
592 	op.offset = map->index << PAGE_SHIFT;
593 	op.count = map->count;
594 
595 	if (copy_to_user(u, &op, sizeof(op)) != 0)
596 		return -EFAULT;
597 	return 0;
598 }
599 
600 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
601 {
602 	struct ioctl_gntdev_unmap_notify op;
603 	struct grant_map *map;
604 	int rc;
605 	int out_flags;
606 	unsigned int out_event;
607 
608 	if (copy_from_user(&op, u, sizeof(op)))
609 		return -EFAULT;
610 
611 	if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
612 		return -EINVAL;
613 
614 	/* We need to grab a reference to the event channel we are going to use
615 	 * to send the notify before releasing the reference we may already have
616 	 * (if someone has called this ioctl twice). This is required so that
617 	 * it is possible to change the clear_byte part of the notification
618 	 * without disturbing the event channel part, which may now be the last
619 	 * reference to that event channel.
620 	 */
621 	if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
622 		if (evtchn_get(op.event_channel_port))
623 			return -EINVAL;
624 	}
625 
626 	out_flags = op.action;
627 	out_event = op.event_channel_port;
628 
629 	spin_lock(&priv->lock);
630 
631 	list_for_each_entry(map, &priv->maps, next) {
632 		uint64_t begin = map->index << PAGE_SHIFT;
633 		uint64_t end = (map->index + map->count) << PAGE_SHIFT;
634 		if (op.index >= begin && op.index < end)
635 			goto found;
636 	}
637 	rc = -ENOENT;
638 	goto unlock_out;
639 
640  found:
641 	if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
642 			(map->flags & GNTMAP_readonly)) {
643 		rc = -EINVAL;
644 		goto unlock_out;
645 	}
646 
647 	out_flags = map->notify.flags;
648 	out_event = map->notify.event;
649 
650 	map->notify.flags = op.action;
651 	map->notify.addr = op.index - (map->index << PAGE_SHIFT);
652 	map->notify.event = op.event_channel_port;
653 
654 	rc = 0;
655 
656  unlock_out:
657 	spin_unlock(&priv->lock);
658 
659 	/* Drop the reference to the event channel we did not save in the map */
660 	if (out_flags & UNMAP_NOTIFY_SEND_EVENT)
661 		evtchn_put(out_event);
662 
663 	return rc;
664 }
665 
666 static long gntdev_ioctl(struct file *flip,
667 			 unsigned int cmd, unsigned long arg)
668 {
669 	struct gntdev_priv *priv = flip->private_data;
670 	void __user *ptr = (void __user *)arg;
671 
672 	switch (cmd) {
673 	case IOCTL_GNTDEV_MAP_GRANT_REF:
674 		return gntdev_ioctl_map_grant_ref(priv, ptr);
675 
676 	case IOCTL_GNTDEV_UNMAP_GRANT_REF:
677 		return gntdev_ioctl_unmap_grant_ref(priv, ptr);
678 
679 	case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
680 		return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
681 
682 	case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
683 		return gntdev_ioctl_notify(priv, ptr);
684 
685 	default:
686 		pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
687 		return -ENOIOCTLCMD;
688 	}
689 
690 	return 0;
691 }
692 
693 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
694 {
695 	struct gntdev_priv *priv = flip->private_data;
696 	int index = vma->vm_pgoff;
697 	int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
698 	struct grant_map *map;
699 	int i, err = -EINVAL;
700 
701 	if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
702 		return -EINVAL;
703 
704 	pr_debug("map %d+%d at %lx (pgoff %lx)\n",
705 			index, count, vma->vm_start, vma->vm_pgoff);
706 
707 	spin_lock(&priv->lock);
708 	map = gntdev_find_map_index(priv, index, count);
709 	if (!map)
710 		goto unlock_out;
711 	if (use_ptemod && map->vma)
712 		goto unlock_out;
713 	if (use_ptemod && priv->mm != vma->vm_mm) {
714 		printk(KERN_WARNING "Huh? Other mm?\n");
715 		goto unlock_out;
716 	}
717 
718 	atomic_inc(&map->users);
719 
720 	vma->vm_ops = &gntdev_vmops;
721 
722 	vma->vm_flags |= VM_RESERVED|VM_DONTEXPAND;
723 
724 	if (use_ptemod)
725 		vma->vm_flags |= VM_DONTCOPY;
726 
727 	vma->vm_private_data = map;
728 
729 	if (use_ptemod)
730 		map->vma = vma;
731 
732 	if (map->flags) {
733 		if ((vma->vm_flags & VM_WRITE) &&
734 				(map->flags & GNTMAP_readonly))
735 			goto out_unlock_put;
736 	} else {
737 		map->flags = GNTMAP_host_map;
738 		if (!(vma->vm_flags & VM_WRITE))
739 			map->flags |= GNTMAP_readonly;
740 	}
741 
742 	spin_unlock(&priv->lock);
743 
744 	if (use_ptemod) {
745 		err = apply_to_page_range(vma->vm_mm, vma->vm_start,
746 					  vma->vm_end - vma->vm_start,
747 					  find_grant_ptes, map);
748 		if (err) {
749 			printk(KERN_WARNING "find_grant_ptes() failure.\n");
750 			goto out_put_map;
751 		}
752 	}
753 
754 	err = map_grant_pages(map);
755 	if (err)
756 		goto out_put_map;
757 
758 	if (!use_ptemod) {
759 		for (i = 0; i < count; i++) {
760 			err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
761 				map->pages[i]);
762 			if (err)
763 				goto out_put_map;
764 		}
765 	}
766 
767 	return 0;
768 
769 unlock_out:
770 	spin_unlock(&priv->lock);
771 	return err;
772 
773 out_unlock_put:
774 	spin_unlock(&priv->lock);
775 out_put_map:
776 	if (use_ptemod)
777 		map->vma = NULL;
778 	gntdev_put_map(map);
779 	return err;
780 }
781 
782 static const struct file_operations gntdev_fops = {
783 	.owner = THIS_MODULE,
784 	.open = gntdev_open,
785 	.release = gntdev_release,
786 	.mmap = gntdev_mmap,
787 	.unlocked_ioctl = gntdev_ioctl
788 };
789 
790 static struct miscdevice gntdev_miscdev = {
791 	.minor        = MISC_DYNAMIC_MINOR,
792 	.name         = "xen/gntdev",
793 	.fops         = &gntdev_fops,
794 };
795 
796 /* ------------------------------------------------------------------ */
797 
798 static int __init gntdev_init(void)
799 {
800 	int err;
801 
802 	if (!xen_domain())
803 		return -ENODEV;
804 
805 	use_ptemod = xen_pv_domain();
806 
807 	err = misc_register(&gntdev_miscdev);
808 	if (err != 0) {
809 		printk(KERN_ERR "Could not register gntdev device\n");
810 		return err;
811 	}
812 	return 0;
813 }
814 
815 static void __exit gntdev_exit(void)
816 {
817 	misc_deregister(&gntdev_miscdev);
818 }
819 
820 module_init(gntdev_init);
821 module_exit(gntdev_exit);
822 
823 /* ------------------------------------------------------------------ */
824