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 
gntdev_print_maps(struct gntdev_priv * priv,char * text,int text_index)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 
gntdev_free_map(struct grant_map * map)108 static void gntdev_free_map(struct grant_map *map)
109 {
110 	if (map == NULL)
111 		return;
112 
113 	if (map->pages)
114 		free_xenballooned_pages(map->count, map->pages);
115 	kfree(map->pages);
116 	kfree(map->grants);
117 	kfree(map->map_ops);
118 	kfree(map->unmap_ops);
119 	kfree(map->kmap_ops);
120 	kfree(map);
121 }
122 
gntdev_alloc_map(struct gntdev_priv * priv,int count)123 static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count)
124 {
125 	struct grant_map *add;
126 	int i;
127 
128 	add = kzalloc(sizeof(struct grant_map), GFP_KERNEL);
129 	if (NULL == add)
130 		return NULL;
131 
132 	add->grants    = kcalloc(count, sizeof(add->grants[0]), GFP_KERNEL);
133 	add->map_ops   = kcalloc(count, sizeof(add->map_ops[0]), GFP_KERNEL);
134 	add->unmap_ops = kcalloc(count, sizeof(add->unmap_ops[0]), GFP_KERNEL);
135 	add->kmap_ops  = kcalloc(count, sizeof(add->kmap_ops[0]), GFP_KERNEL);
136 	add->pages     = kcalloc(count, sizeof(add->pages[0]), GFP_KERNEL);
137 	if (NULL == add->grants    ||
138 	    NULL == add->map_ops   ||
139 	    NULL == add->unmap_ops ||
140 	    NULL == add->kmap_ops  ||
141 	    NULL == add->pages)
142 		goto err;
143 
144 	if (alloc_xenballooned_pages(count, add->pages, false /* lowmem */))
145 		goto err;
146 
147 	for (i = 0; i < count; i++) {
148 		add->map_ops[i].handle = -1;
149 		add->unmap_ops[i].handle = -1;
150 		add->kmap_ops[i].handle = -1;
151 	}
152 
153 	add->index = 0;
154 	add->count = count;
155 	atomic_set(&add->users, 1);
156 
157 	return add;
158 
159 err:
160 	gntdev_free_map(add);
161 	return NULL;
162 }
163 
gntdev_add_map(struct gntdev_priv * priv,struct grant_map * add)164 static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add)
165 {
166 	struct grant_map *map;
167 
168 	list_for_each_entry(map, &priv->maps, next) {
169 		if (add->index + add->count < map->index) {
170 			list_add_tail(&add->next, &map->next);
171 			goto done;
172 		}
173 		add->index = map->index + map->count;
174 	}
175 	list_add_tail(&add->next, &priv->maps);
176 
177 done:
178 	gntdev_print_maps(priv, "[new]", add->index);
179 }
180 
gntdev_find_map_index(struct gntdev_priv * priv,int index,int count)181 static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
182 		int index, int count)
183 {
184 	struct grant_map *map;
185 
186 	list_for_each_entry(map, &priv->maps, next) {
187 		if (map->index != index)
188 			continue;
189 		if (count && map->count != count)
190 			continue;
191 		return map;
192 	}
193 	return NULL;
194 }
195 
gntdev_put_map(struct grant_map * map)196 static void gntdev_put_map(struct grant_map *map)
197 {
198 	if (!map)
199 		return;
200 
201 	if (!atomic_dec_and_test(&map->users))
202 		return;
203 
204 	atomic_sub(map->count, &pages_mapped);
205 
206 	if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
207 		notify_remote_via_evtchn(map->notify.event);
208 		evtchn_put(map->notify.event);
209 	}
210 
211 	if (map->pages && !use_ptemod)
212 		unmap_grant_pages(map, 0, map->count);
213 	gntdev_free_map(map);
214 }
215 
216 /* ------------------------------------------------------------------ */
217 
find_grant_ptes(pte_t * pte,pgtable_t token,unsigned long addr,void * data)218 static int find_grant_ptes(pte_t *pte, pgtable_t token,
219 		unsigned long addr, void *data)
220 {
221 	struct grant_map *map = data;
222 	unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT;
223 	int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte;
224 	u64 pte_maddr;
225 
226 	BUG_ON(pgnr >= map->count);
227 	pte_maddr = arbitrary_virt_to_machine(pte).maddr;
228 
229 	gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags,
230 			  map->grants[pgnr].ref,
231 			  map->grants[pgnr].domid);
232 	gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags,
233 			    -1 /* handle */);
234 	return 0;
235 }
236 
map_grant_pages(struct grant_map * map)237 static int map_grant_pages(struct grant_map *map)
238 {
239 	int i, err = 0;
240 
241 	if (!use_ptemod) {
242 		/* Note: it could already be mapped */
243 		if (map->map_ops[0].handle != -1)
244 			return 0;
245 		for (i = 0; i < map->count; i++) {
246 			unsigned long addr = (unsigned long)
247 				pfn_to_kaddr(page_to_pfn(map->pages[i]));
248 			gnttab_set_map_op(&map->map_ops[i], addr, map->flags,
249 				map->grants[i].ref,
250 				map->grants[i].domid);
251 			gnttab_set_unmap_op(&map->unmap_ops[i], addr,
252 				map->flags, -1 /* handle */);
253 		}
254 	} else {
255 		/*
256 		 * Setup the map_ops corresponding to the pte entries pointing
257 		 * to the kernel linear addresses of the struct pages.
258 		 * These ptes are completely different from the user ptes dealt
259 		 * with find_grant_ptes.
260 		 */
261 		for (i = 0; i < map->count; i++) {
262 			unsigned level;
263 			unsigned long address = (unsigned long)
264 				pfn_to_kaddr(page_to_pfn(map->pages[i]));
265 			pte_t *ptep;
266 			u64 pte_maddr = 0;
267 			BUG_ON(PageHighMem(map->pages[i]));
268 
269 			ptep = lookup_address(address, &level);
270 			pte_maddr = arbitrary_virt_to_machine(ptep).maddr;
271 			gnttab_set_map_op(&map->kmap_ops[i], pte_maddr,
272 				map->flags |
273 				GNTMAP_host_map |
274 				GNTMAP_contains_pte,
275 				map->grants[i].ref,
276 				map->grants[i].domid);
277 		}
278 	}
279 
280 	pr_debug("map %d+%d\n", map->index, map->count);
281 	err = gnttab_map_refs(map->map_ops, use_ptemod ? map->kmap_ops : NULL,
282 			map->pages, map->count);
283 	if (err)
284 		return err;
285 
286 	for (i = 0; i < map->count; i++) {
287 		if (map->map_ops[i].status)
288 			err = -EINVAL;
289 		else {
290 			BUG_ON(map->map_ops[i].handle == -1);
291 			map->unmap_ops[i].handle = map->map_ops[i].handle;
292 			pr_debug("map handle=%d\n", map->map_ops[i].handle);
293 		}
294 	}
295 	return err;
296 }
297 
__unmap_grant_pages(struct grant_map * map,int offset,int pages)298 static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
299 {
300 	int i, err = 0;
301 
302 	if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) {
303 		int pgno = (map->notify.addr >> PAGE_SHIFT);
304 		if (pgno >= offset && pgno < offset + pages && use_ptemod) {
305 			void __user *tmp = (void __user *)
306 				map->vma->vm_start + map->notify.addr;
307 			err = copy_to_user(tmp, &err, 1);
308 			if (err)
309 				return -EFAULT;
310 			map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
311 		} else if (pgno >= offset && pgno < offset + pages) {
312 			uint8_t *tmp = kmap(map->pages[pgno]);
313 			tmp[map->notify.addr & (PAGE_SIZE-1)] = 0;
314 			kunmap(map->pages[pgno]);
315 			map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
316 		}
317 	}
318 
319 	err = gnttab_unmap_refs(map->unmap_ops + offset,
320 			use_ptemod ? map->kmap_ops + offset : NULL, map->pages + offset,
321 			pages);
322 	if (err)
323 		return err;
324 
325 	for (i = 0; i < pages; i++) {
326 		if (map->unmap_ops[offset+i].status)
327 			err = -EINVAL;
328 		pr_debug("unmap handle=%d st=%d\n",
329 			map->unmap_ops[offset+i].handle,
330 			map->unmap_ops[offset+i].status);
331 		map->unmap_ops[offset+i].handle = -1;
332 	}
333 	return err;
334 }
335 
unmap_grant_pages(struct grant_map * map,int offset,int pages)336 static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
337 {
338 	int range, err = 0;
339 
340 	pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
341 
342 	/* It is possible the requested range will have a "hole" where we
343 	 * already unmapped some of the grants. Only unmap valid ranges.
344 	 */
345 	while (pages && !err) {
346 		while (pages && map->unmap_ops[offset].handle == -1) {
347 			offset++;
348 			pages--;
349 		}
350 		range = 0;
351 		while (range < pages) {
352 			if (map->unmap_ops[offset+range].handle == -1) {
353 				range--;
354 				break;
355 			}
356 			range++;
357 		}
358 		err = __unmap_grant_pages(map, offset, range);
359 		offset += range;
360 		pages -= range;
361 	}
362 
363 	return err;
364 }
365 
366 /* ------------------------------------------------------------------ */
367 
gntdev_vma_open(struct vm_area_struct * vma)368 static void gntdev_vma_open(struct vm_area_struct *vma)
369 {
370 	struct grant_map *map = vma->vm_private_data;
371 
372 	pr_debug("gntdev_vma_open %p\n", vma);
373 	atomic_inc(&map->users);
374 }
375 
gntdev_vma_close(struct vm_area_struct * vma)376 static void gntdev_vma_close(struct vm_area_struct *vma)
377 {
378 	struct grant_map *map = vma->vm_private_data;
379 
380 	pr_debug("gntdev_vma_close %p\n", vma);
381 	map->vma = NULL;
382 	vma->vm_private_data = NULL;
383 	gntdev_put_map(map);
384 }
385 
386 static struct vm_operations_struct gntdev_vmops = {
387 	.open = gntdev_vma_open,
388 	.close = gntdev_vma_close,
389 };
390 
391 /* ------------------------------------------------------------------ */
392 
mn_invl_range_start(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long start,unsigned long end)393 static void mn_invl_range_start(struct mmu_notifier *mn,
394 				struct mm_struct *mm,
395 				unsigned long start, unsigned long end)
396 {
397 	struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
398 	struct grant_map *map;
399 	unsigned long mstart, mend;
400 	int err;
401 
402 	spin_lock(&priv->lock);
403 	list_for_each_entry(map, &priv->maps, next) {
404 		if (!map->vma)
405 			continue;
406 		if (map->vma->vm_start >= end)
407 			continue;
408 		if (map->vma->vm_end <= start)
409 			continue;
410 		mstart = max(start, map->vma->vm_start);
411 		mend   = min(end,   map->vma->vm_end);
412 		pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
413 				map->index, map->count,
414 				map->vma->vm_start, map->vma->vm_end,
415 				start, end, mstart, mend);
416 		err = unmap_grant_pages(map,
417 					(mstart - map->vma->vm_start) >> PAGE_SHIFT,
418 					(mend - mstart) >> PAGE_SHIFT);
419 		WARN_ON(err);
420 	}
421 	spin_unlock(&priv->lock);
422 }
423 
mn_invl_page(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long address)424 static void mn_invl_page(struct mmu_notifier *mn,
425 			 struct mm_struct *mm,
426 			 unsigned long address)
427 {
428 	mn_invl_range_start(mn, mm, address, address + PAGE_SIZE);
429 }
430 
mn_release(struct mmu_notifier * mn,struct mm_struct * mm)431 static void mn_release(struct mmu_notifier *mn,
432 		       struct mm_struct *mm)
433 {
434 	struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
435 	struct grant_map *map;
436 	int err;
437 
438 	spin_lock(&priv->lock);
439 	list_for_each_entry(map, &priv->maps, next) {
440 		if (!map->vma)
441 			continue;
442 		pr_debug("map %d+%d (%lx %lx)\n",
443 				map->index, map->count,
444 				map->vma->vm_start, map->vma->vm_end);
445 		err = unmap_grant_pages(map, /* offset */ 0, map->count);
446 		WARN_ON(err);
447 	}
448 	spin_unlock(&priv->lock);
449 }
450 
451 struct mmu_notifier_ops gntdev_mmu_ops = {
452 	.release                = mn_release,
453 	.invalidate_page        = mn_invl_page,
454 	.invalidate_range_start = mn_invl_range_start,
455 };
456 
457 /* ------------------------------------------------------------------ */
458 
gntdev_open(struct inode * inode,struct file * flip)459 static int gntdev_open(struct inode *inode, struct file *flip)
460 {
461 	struct gntdev_priv *priv;
462 	int ret = 0;
463 
464 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
465 	if (!priv)
466 		return -ENOMEM;
467 
468 	INIT_LIST_HEAD(&priv->maps);
469 	spin_lock_init(&priv->lock);
470 
471 	if (use_ptemod) {
472 		priv->mm = get_task_mm(current);
473 		if (!priv->mm) {
474 			kfree(priv);
475 			return -ENOMEM;
476 		}
477 		priv->mn.ops = &gntdev_mmu_ops;
478 		ret = mmu_notifier_register(&priv->mn, priv->mm);
479 		mmput(priv->mm);
480 	}
481 
482 	if (ret) {
483 		kfree(priv);
484 		return ret;
485 	}
486 
487 	flip->private_data = priv;
488 	pr_debug("priv %p\n", priv);
489 
490 	return 0;
491 }
492 
gntdev_release(struct inode * inode,struct file * flip)493 static int gntdev_release(struct inode *inode, struct file *flip)
494 {
495 	struct gntdev_priv *priv = flip->private_data;
496 	struct grant_map *map;
497 
498 	pr_debug("priv %p\n", priv);
499 
500 	while (!list_empty(&priv->maps)) {
501 		map = list_entry(priv->maps.next, struct grant_map, next);
502 		list_del(&map->next);
503 		gntdev_put_map(map);
504 	}
505 
506 	if (use_ptemod)
507 		mmu_notifier_unregister(&priv->mn, priv->mm);
508 	kfree(priv);
509 	return 0;
510 }
511 
gntdev_ioctl_map_grant_ref(struct gntdev_priv * priv,struct ioctl_gntdev_map_grant_ref __user * u)512 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
513 				       struct ioctl_gntdev_map_grant_ref __user *u)
514 {
515 	struct ioctl_gntdev_map_grant_ref op;
516 	struct grant_map *map;
517 	int err;
518 
519 	if (copy_from_user(&op, u, sizeof(op)) != 0)
520 		return -EFAULT;
521 	pr_debug("priv %p, add %d\n", priv, op.count);
522 	if (unlikely(op.count <= 0))
523 		return -EINVAL;
524 
525 	err = -ENOMEM;
526 	map = gntdev_alloc_map(priv, op.count);
527 	if (!map)
528 		return err;
529 
530 	if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) {
531 		pr_debug("can't map: over limit\n");
532 		gntdev_put_map(map);
533 		return err;
534 	}
535 
536 	if (copy_from_user(map->grants, &u->refs,
537 			   sizeof(map->grants[0]) * op.count) != 0) {
538 		gntdev_put_map(map);
539 		return err;
540 	}
541 
542 	spin_lock(&priv->lock);
543 	gntdev_add_map(priv, map);
544 	op.index = map->index << PAGE_SHIFT;
545 	spin_unlock(&priv->lock);
546 
547 	if (copy_to_user(u, &op, sizeof(op)) != 0)
548 		return -EFAULT;
549 
550 	return 0;
551 }
552 
gntdev_ioctl_unmap_grant_ref(struct gntdev_priv * priv,struct ioctl_gntdev_unmap_grant_ref __user * u)553 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
554 					 struct ioctl_gntdev_unmap_grant_ref __user *u)
555 {
556 	struct ioctl_gntdev_unmap_grant_ref op;
557 	struct grant_map *map;
558 	int err = -ENOENT;
559 
560 	if (copy_from_user(&op, u, sizeof(op)) != 0)
561 		return -EFAULT;
562 	pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
563 
564 	spin_lock(&priv->lock);
565 	map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
566 	if (map) {
567 		list_del(&map->next);
568 		err = 0;
569 	}
570 	spin_unlock(&priv->lock);
571 	if (map)
572 		gntdev_put_map(map);
573 	return err;
574 }
575 
gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv * priv,struct ioctl_gntdev_get_offset_for_vaddr __user * u)576 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
577 					      struct ioctl_gntdev_get_offset_for_vaddr __user *u)
578 {
579 	struct ioctl_gntdev_get_offset_for_vaddr op;
580 	struct vm_area_struct *vma;
581 	struct grant_map *map;
582 
583 	if (copy_from_user(&op, u, sizeof(op)) != 0)
584 		return -EFAULT;
585 	pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
586 
587 	vma = find_vma(current->mm, op.vaddr);
588 	if (!vma || vma->vm_ops != &gntdev_vmops)
589 		return -EINVAL;
590 
591 	map = vma->vm_private_data;
592 	if (!map)
593 		return -EINVAL;
594 
595 	op.offset = map->index << PAGE_SHIFT;
596 	op.count = map->count;
597 
598 	if (copy_to_user(u, &op, sizeof(op)) != 0)
599 		return -EFAULT;
600 	return 0;
601 }
602 
gntdev_ioctl_notify(struct gntdev_priv * priv,void __user * u)603 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
604 {
605 	struct ioctl_gntdev_unmap_notify op;
606 	struct grant_map *map;
607 	int rc;
608 	int out_flags;
609 	unsigned int out_event;
610 
611 	if (copy_from_user(&op, u, sizeof(op)))
612 		return -EFAULT;
613 
614 	if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
615 		return -EINVAL;
616 
617 	/* We need to grab a reference to the event channel we are going to use
618 	 * to send the notify before releasing the reference we may already have
619 	 * (if someone has called this ioctl twice). This is required so that
620 	 * it is possible to change the clear_byte part of the notification
621 	 * without disturbing the event channel part, which may now be the last
622 	 * reference to that event channel.
623 	 */
624 	if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
625 		if (evtchn_get(op.event_channel_port))
626 			return -EINVAL;
627 	}
628 
629 	out_flags = op.action;
630 	out_event = op.event_channel_port;
631 
632 	spin_lock(&priv->lock);
633 
634 	list_for_each_entry(map, &priv->maps, next) {
635 		uint64_t begin = map->index << PAGE_SHIFT;
636 		uint64_t end = (map->index + map->count) << PAGE_SHIFT;
637 		if (op.index >= begin && op.index < end)
638 			goto found;
639 	}
640 	rc = -ENOENT;
641 	goto unlock_out;
642 
643  found:
644 	if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
645 			(map->flags & GNTMAP_readonly)) {
646 		rc = -EINVAL;
647 		goto unlock_out;
648 	}
649 
650 	out_flags = map->notify.flags;
651 	out_event = map->notify.event;
652 
653 	map->notify.flags = op.action;
654 	map->notify.addr = op.index - (map->index << PAGE_SHIFT);
655 	map->notify.event = op.event_channel_port;
656 
657 	rc = 0;
658 
659  unlock_out:
660 	spin_unlock(&priv->lock);
661 
662 	/* Drop the reference to the event channel we did not save in the map */
663 	if (out_flags & UNMAP_NOTIFY_SEND_EVENT)
664 		evtchn_put(out_event);
665 
666 	return rc;
667 }
668 
gntdev_ioctl(struct file * flip,unsigned int cmd,unsigned long arg)669 static long gntdev_ioctl(struct file *flip,
670 			 unsigned int cmd, unsigned long arg)
671 {
672 	struct gntdev_priv *priv = flip->private_data;
673 	void __user *ptr = (void __user *)arg;
674 
675 	switch (cmd) {
676 	case IOCTL_GNTDEV_MAP_GRANT_REF:
677 		return gntdev_ioctl_map_grant_ref(priv, ptr);
678 
679 	case IOCTL_GNTDEV_UNMAP_GRANT_REF:
680 		return gntdev_ioctl_unmap_grant_ref(priv, ptr);
681 
682 	case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
683 		return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
684 
685 	case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
686 		return gntdev_ioctl_notify(priv, ptr);
687 
688 	default:
689 		pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
690 		return -ENOIOCTLCMD;
691 	}
692 
693 	return 0;
694 }
695 
gntdev_mmap(struct file * flip,struct vm_area_struct * vma)696 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
697 {
698 	struct gntdev_priv *priv = flip->private_data;
699 	int index = vma->vm_pgoff;
700 	int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
701 	struct grant_map *map;
702 	int i, err = -EINVAL;
703 
704 	if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
705 		return -EINVAL;
706 
707 	pr_debug("map %d+%d at %lx (pgoff %lx)\n",
708 			index, count, vma->vm_start, vma->vm_pgoff);
709 
710 	spin_lock(&priv->lock);
711 	map = gntdev_find_map_index(priv, index, count);
712 	if (!map)
713 		goto unlock_out;
714 	if (use_ptemod && map->vma)
715 		goto unlock_out;
716 	if (use_ptemod && priv->mm != vma->vm_mm) {
717 		printk(KERN_WARNING "Huh? Other mm?\n");
718 		goto unlock_out;
719 	}
720 
721 	atomic_inc(&map->users);
722 
723 	vma->vm_ops = &gntdev_vmops;
724 
725 	vma->vm_flags |= VM_RESERVED|VM_DONTEXPAND;
726 
727 	if (use_ptemod)
728 		vma->vm_flags |= VM_DONTCOPY;
729 
730 	vma->vm_private_data = map;
731 
732 	if (use_ptemod)
733 		map->vma = vma;
734 
735 	if (map->flags) {
736 		if ((vma->vm_flags & VM_WRITE) &&
737 				(map->flags & GNTMAP_readonly))
738 			goto out_unlock_put;
739 	} else {
740 		map->flags = GNTMAP_host_map;
741 		if (!(vma->vm_flags & VM_WRITE))
742 			map->flags |= GNTMAP_readonly;
743 	}
744 
745 	spin_unlock(&priv->lock);
746 
747 	if (use_ptemod) {
748 		err = apply_to_page_range(vma->vm_mm, vma->vm_start,
749 					  vma->vm_end - vma->vm_start,
750 					  find_grant_ptes, map);
751 		if (err) {
752 			printk(KERN_WARNING "find_grant_ptes() failure.\n");
753 			goto out_put_map;
754 		}
755 	}
756 
757 	err = map_grant_pages(map);
758 	if (err)
759 		goto out_put_map;
760 
761 	if (!use_ptemod) {
762 		for (i = 0; i < count; i++) {
763 			err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
764 				map->pages[i]);
765 			if (err)
766 				goto out_put_map;
767 		}
768 	}
769 
770 	return 0;
771 
772 unlock_out:
773 	spin_unlock(&priv->lock);
774 	return err;
775 
776 out_unlock_put:
777 	spin_unlock(&priv->lock);
778 out_put_map:
779 	if (use_ptemod)
780 		map->vma = NULL;
781 	gntdev_put_map(map);
782 	return err;
783 }
784 
785 static const struct file_operations gntdev_fops = {
786 	.owner = THIS_MODULE,
787 	.open = gntdev_open,
788 	.release = gntdev_release,
789 	.mmap = gntdev_mmap,
790 	.unlocked_ioctl = gntdev_ioctl
791 };
792 
793 static struct miscdevice gntdev_miscdev = {
794 	.minor        = MISC_DYNAMIC_MINOR,
795 	.name         = "xen/gntdev",
796 	.fops         = &gntdev_fops,
797 };
798 
799 /* ------------------------------------------------------------------ */
800 
gntdev_init(void)801 static int __init gntdev_init(void)
802 {
803 	int err;
804 
805 	if (!xen_domain())
806 		return -ENODEV;
807 
808 	use_ptemod = xen_pv_domain();
809 
810 	err = misc_register(&gntdev_miscdev);
811 	if (err != 0) {
812 		printk(KERN_ERR "Could not register gntdev device\n");
813 		return err;
814 	}
815 	return 0;
816 }
817 
gntdev_exit(void)818 static void __exit gntdev_exit(void)
819 {
820 	misc_deregister(&gntdev_miscdev);
821 }
822 
823 module_init(gntdev_init);
824 module_exit(gntdev_exit);
825 
826 /* ------------------------------------------------------------------ */
827