1 /******************************************************************************
4 * Device for accessing (in user-space) pages that have been granted by other
7 * Copyright (c) 2006-2007, D G Murray.
8 * (c) 2009 Gerd Hoffmann <kraxel@redhat.com>
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.
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
22 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/init.h>
25 #include <linux/miscdevice.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>
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>
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");
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 "
56 static atomic_t pages_mapped
= ATOMIC_INIT(0);
58 static int use_ptemod
;
61 struct list_head maps
;
62 /* lock protects maps from concurrent changes */
65 struct mmu_notifier mn
;
70 /* Address relative to the start of the grant_map */
76 struct list_head next
;
77 struct vm_area_struct
*vma
;
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
;
90 static int unmap_grant_pages(struct grant_map
*map
, int offset
, int pages
);
92 /* ------------------------------------------------------------------ */
94 static void gntdev_print_maps(struct gntdev_priv
*priv
,
95 char *text
, int text_index
)
98 struct grant_map
*map
;
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
: "");
108 static struct grant_map
*gntdev_alloc_map(struct gntdev_priv
*priv
, int count
)
110 struct grant_map
*add
;
113 add
= kzalloc(sizeof(struct grant_map
), GFP_KERNEL
);
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
||
129 if (alloc_xenballooned_pages(count
, add
->pages
, false /* lowmem */))
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;
140 atomic_set(&add
->users
, 1);
148 kfree(add
->unmap_ops
);
149 kfree(add
->kmap_ops
);
154 static void gntdev_add_map(struct gntdev_priv
*priv
, struct grant_map
*add
)
156 struct grant_map
*map
;
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
);
163 add
->index
= map
->index
+ map
->count
;
165 list_add_tail(&add
->next
, &priv
->maps
);
168 gntdev_print_maps(priv
, "[new]", add
->index
);
171 static struct grant_map
*gntdev_find_map_index(struct gntdev_priv
*priv
,
172 int index
, int count
)
174 struct grant_map
*map
;
176 list_for_each_entry(map
, &priv
->maps
, next
) {
177 if (map
->index
!= index
)
179 if (count
&& map
->count
!= count
)
186 static void gntdev_put_map(struct grant_map
*map
)
191 if (!atomic_dec_and_test(&map
->users
))
194 atomic_sub(map
->count
, &pages_mapped
);
196 if (map
->notify
.flags
& UNMAP_NOTIFY_SEND_EVENT
)
197 notify_remote_via_evtchn(map
->notify
.event
);
201 unmap_grant_pages(map
, 0, map
->count
);
203 free_xenballooned_pages(map
->count
, map
->pages
);
208 kfree(map
->unmap_ops
);
212 /* ------------------------------------------------------------------ */
214 static int find_grant_ptes(pte_t
*pte
, pgtable_t token
,
215 unsigned long addr
, void *data
)
217 struct grant_map
*map
= data
;
218 unsigned int pgnr
= (addr
- map
->vma
->vm_start
) >> PAGE_SHIFT
;
219 int flags
= map
->flags
| GNTMAP_application_map
| GNTMAP_contains_pte
;
222 BUG_ON(pgnr
>= map
->count
);
223 pte_maddr
= arbitrary_virt_to_machine(pte
).maddr
;
225 gnttab_set_map_op(&map
->map_ops
[pgnr
], pte_maddr
, flags
,
226 map
->grants
[pgnr
].ref
,
227 map
->grants
[pgnr
].domid
);
228 gnttab_set_unmap_op(&map
->unmap_ops
[pgnr
], pte_maddr
, flags
,
233 static int map_grant_pages(struct grant_map
*map
)
238 /* Note: it could already be mapped */
239 if (map
->map_ops
[0].handle
!= -1)
241 for (i
= 0; i
< map
->count
; i
++) {
242 unsigned long addr
= (unsigned long)
243 pfn_to_kaddr(page_to_pfn(map
->pages
[i
]));
244 gnttab_set_map_op(&map
->map_ops
[i
], addr
, map
->flags
,
246 map
->grants
[i
].domid
);
247 gnttab_set_unmap_op(&map
->unmap_ops
[i
], addr
,
248 map
->flags
, -1 /* handle */);
252 * Setup the map_ops corresponding to the pte entries pointing
253 * to the kernel linear addresses of the struct pages.
254 * These ptes are completely different from the user ptes dealt
255 * with find_grant_ptes.
257 for (i
= 0; i
< map
->count
; i
++) {
259 unsigned long address
= (unsigned long)
260 pfn_to_kaddr(page_to_pfn(map
->pages
[i
]));
263 BUG_ON(PageHighMem(map
->pages
[i
]));
265 ptep
= lookup_address(address
, &level
);
266 pte_maddr
= arbitrary_virt_to_machine(ptep
).maddr
;
267 gnttab_set_map_op(&map
->kmap_ops
[i
], pte_maddr
,
272 map
->grants
[i
].domid
);
276 pr_debug("map %d+%d\n", map
->index
, map
->count
);
277 err
= gnttab_map_refs(map
->map_ops
, use_ptemod
? map
->kmap_ops
: NULL
,
278 map
->pages
, map
->count
);
282 for (i
= 0; i
< map
->count
; i
++) {
283 if (map
->map_ops
[i
].status
)
286 BUG_ON(map
->map_ops
[i
].handle
== -1);
287 map
->unmap_ops
[i
].handle
= map
->map_ops
[i
].handle
;
288 pr_debug("map handle=%d\n", map
->map_ops
[i
].handle
);
294 static int __unmap_grant_pages(struct grant_map
*map
, int offset
, int pages
)
298 if (map
->notify
.flags
& UNMAP_NOTIFY_CLEAR_BYTE
) {
299 int pgno
= (map
->notify
.addr
>> PAGE_SHIFT
);
300 if (pgno
>= offset
&& pgno
< offset
+ pages
&& use_ptemod
) {
301 void __user
*tmp
= (void __user
*)
302 map
->vma
->vm_start
+ map
->notify
.addr
;
303 err
= copy_to_user(tmp
, &err
, 1);
306 map
->notify
.flags
&= ~UNMAP_NOTIFY_CLEAR_BYTE
;
307 } else if (pgno
>= offset
&& pgno
< offset
+ pages
) {
308 uint8_t *tmp
= kmap(map
->pages
[pgno
]);
309 tmp
[map
->notify
.addr
& (PAGE_SIZE
-1)] = 0;
310 kunmap(map
->pages
[pgno
]);
311 map
->notify
.flags
&= ~UNMAP_NOTIFY_CLEAR_BYTE
;
315 err
= gnttab_unmap_refs(map
->unmap_ops
+ offset
, map
->pages
+ offset
, pages
);
319 for (i
= 0; i
< pages
; i
++) {
320 if (map
->unmap_ops
[offset
+i
].status
)
322 pr_debug("unmap handle=%d st=%d\n",
323 map
->unmap_ops
[offset
+i
].handle
,
324 map
->unmap_ops
[offset
+i
].status
);
325 map
->unmap_ops
[offset
+i
].handle
= -1;
330 static int unmap_grant_pages(struct grant_map
*map
, int offset
, int pages
)
334 pr_debug("unmap %d+%d [%d+%d]\n", map
->index
, map
->count
, offset
, pages
);
336 /* It is possible the requested range will have a "hole" where we
337 * already unmapped some of the grants. Only unmap valid ranges.
339 while (pages
&& !err
) {
340 while (pages
&& map
->unmap_ops
[offset
].handle
== -1) {
345 while (range
< pages
) {
346 if (map
->unmap_ops
[offset
+range
].handle
== -1) {
352 err
= __unmap_grant_pages(map
, offset
, range
);
360 /* ------------------------------------------------------------------ */
362 static void gntdev_vma_open(struct vm_area_struct
*vma
)
364 struct grant_map
*map
= vma
->vm_private_data
;
366 pr_debug("gntdev_vma_open %p\n", vma
);
367 atomic_inc(&map
->users
);
370 static void gntdev_vma_close(struct vm_area_struct
*vma
)
372 struct grant_map
*map
= vma
->vm_private_data
;
374 pr_debug("gntdev_vma_close %p\n", vma
);
376 vma
->vm_private_data
= NULL
;
380 static struct vm_operations_struct gntdev_vmops
= {
381 .open
= gntdev_vma_open
,
382 .close
= gntdev_vma_close
,
385 /* ------------------------------------------------------------------ */
387 static void mn_invl_range_start(struct mmu_notifier
*mn
,
388 struct mm_struct
*mm
,
389 unsigned long start
, unsigned long end
)
391 struct gntdev_priv
*priv
= container_of(mn
, struct gntdev_priv
, mn
);
392 struct grant_map
*map
;
393 unsigned long mstart
, mend
;
396 spin_lock(&priv
->lock
);
397 list_for_each_entry(map
, &priv
->maps
, next
) {
400 if (map
->vma
->vm_start
>= end
)
402 if (map
->vma
->vm_end
<= start
)
404 mstart
= max(start
, map
->vma
->vm_start
);
405 mend
= min(end
, map
->vma
->vm_end
);
406 pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
407 map
->index
, map
->count
,
408 map
->vma
->vm_start
, map
->vma
->vm_end
,
409 start
, end
, mstart
, mend
);
410 err
= unmap_grant_pages(map
,
411 (mstart
- map
->vma
->vm_start
) >> PAGE_SHIFT
,
412 (mend
- mstart
) >> PAGE_SHIFT
);
415 spin_unlock(&priv
->lock
);
418 static void mn_invl_page(struct mmu_notifier
*mn
,
419 struct mm_struct
*mm
,
420 unsigned long address
)
422 mn_invl_range_start(mn
, mm
, address
, address
+ PAGE_SIZE
);
425 static void mn_release(struct mmu_notifier
*mn
,
426 struct mm_struct
*mm
)
428 struct gntdev_priv
*priv
= container_of(mn
, struct gntdev_priv
, mn
);
429 struct grant_map
*map
;
432 spin_lock(&priv
->lock
);
433 list_for_each_entry(map
, &priv
->maps
, next
) {
436 pr_debug("map %d+%d (%lx %lx)\n",
437 map
->index
, map
->count
,
438 map
->vma
->vm_start
, map
->vma
->vm_end
);
439 err
= unmap_grant_pages(map
, /* offset */ 0, map
->count
);
442 spin_unlock(&priv
->lock
);
445 struct mmu_notifier_ops gntdev_mmu_ops
= {
446 .release
= mn_release
,
447 .invalidate_page
= mn_invl_page
,
448 .invalidate_range_start
= mn_invl_range_start
,
451 /* ------------------------------------------------------------------ */
453 static int gntdev_open(struct inode
*inode
, struct file
*flip
)
455 struct gntdev_priv
*priv
;
458 priv
= kzalloc(sizeof(*priv
), GFP_KERNEL
);
462 INIT_LIST_HEAD(&priv
->maps
);
463 spin_lock_init(&priv
->lock
);
466 priv
->mm
= get_task_mm(current
);
471 priv
->mn
.ops
= &gntdev_mmu_ops
;
472 ret
= mmu_notifier_register(&priv
->mn
, priv
->mm
);
481 flip
->private_data
= priv
;
482 pr_debug("priv %p\n", priv
);
487 static int gntdev_release(struct inode
*inode
, struct file
*flip
)
489 struct gntdev_priv
*priv
= flip
->private_data
;
490 struct grant_map
*map
;
492 pr_debug("priv %p\n", priv
);
494 while (!list_empty(&priv
->maps
)) {
495 map
= list_entry(priv
->maps
.next
, struct grant_map
, next
);
496 list_del(&map
->next
);
501 mmu_notifier_unregister(&priv
->mn
, priv
->mm
);
506 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv
*priv
,
507 struct ioctl_gntdev_map_grant_ref __user
*u
)
509 struct ioctl_gntdev_map_grant_ref op
;
510 struct grant_map
*map
;
513 if (copy_from_user(&op
, u
, sizeof(op
)) != 0)
515 pr_debug("priv %p, add %d\n", priv
, op
.count
);
516 if (unlikely(op
.count
<= 0))
520 map
= gntdev_alloc_map(priv
, op
.count
);
524 if (unlikely(atomic_add_return(op
.count
, &pages_mapped
) > limit
)) {
525 pr_debug("can't map: over limit\n");
530 if (copy_from_user(map
->grants
, &u
->refs
,
531 sizeof(map
->grants
[0]) * op
.count
) != 0) {
536 spin_lock(&priv
->lock
);
537 gntdev_add_map(priv
, map
);
538 op
.index
= map
->index
<< PAGE_SHIFT
;
539 spin_unlock(&priv
->lock
);
541 if (copy_to_user(u
, &op
, sizeof(op
)) != 0)
547 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv
*priv
,
548 struct ioctl_gntdev_unmap_grant_ref __user
*u
)
550 struct ioctl_gntdev_unmap_grant_ref op
;
551 struct grant_map
*map
;
554 if (copy_from_user(&op
, u
, sizeof(op
)) != 0)
556 pr_debug("priv %p, del %d+%d\n", priv
, (int)op
.index
, (int)op
.count
);
558 spin_lock(&priv
->lock
);
559 map
= gntdev_find_map_index(priv
, op
.index
>> PAGE_SHIFT
, op
.count
);
561 list_del(&map
->next
);
564 spin_unlock(&priv
->lock
);
570 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv
*priv
,
571 struct ioctl_gntdev_get_offset_for_vaddr __user
*u
)
573 struct ioctl_gntdev_get_offset_for_vaddr op
;
574 struct vm_area_struct
*vma
;
575 struct grant_map
*map
;
577 if (copy_from_user(&op
, u
, sizeof(op
)) != 0)
579 pr_debug("priv %p, offset for vaddr %lx\n", priv
, (unsigned long)op
.vaddr
);
581 vma
= find_vma(current
->mm
, op
.vaddr
);
582 if (!vma
|| vma
->vm_ops
!= &gntdev_vmops
)
585 map
= vma
->vm_private_data
;
589 op
.offset
= map
->index
<< PAGE_SHIFT
;
590 op
.count
= map
->count
;
592 if (copy_to_user(u
, &op
, sizeof(op
)) != 0)
597 static long gntdev_ioctl_notify(struct gntdev_priv
*priv
, void __user
*u
)
599 struct ioctl_gntdev_unmap_notify op
;
600 struct grant_map
*map
;
603 if (copy_from_user(&op
, u
, sizeof(op
)))
606 if (op
.action
& ~(UNMAP_NOTIFY_CLEAR_BYTE
|UNMAP_NOTIFY_SEND_EVENT
))
609 spin_lock(&priv
->lock
);
611 list_for_each_entry(map
, &priv
->maps
, next
) {
612 uint64_t begin
= map
->index
<< PAGE_SHIFT
;
613 uint64_t end
= (map
->index
+ map
->count
) << PAGE_SHIFT
;
614 if (op
.index
>= begin
&& op
.index
< end
)
621 if ((op
.action
& UNMAP_NOTIFY_CLEAR_BYTE
) &&
622 (map
->flags
& GNTMAP_readonly
)) {
627 map
->notify
.flags
= op
.action
;
628 map
->notify
.addr
= op
.index
- (map
->index
<< PAGE_SHIFT
);
629 map
->notify
.event
= op
.event_channel_port
;
632 spin_unlock(&priv
->lock
);
636 static long gntdev_ioctl(struct file
*flip
,
637 unsigned int cmd
, unsigned long arg
)
639 struct gntdev_priv
*priv
= flip
->private_data
;
640 void __user
*ptr
= (void __user
*)arg
;
643 case IOCTL_GNTDEV_MAP_GRANT_REF
:
644 return gntdev_ioctl_map_grant_ref(priv
, ptr
);
646 case IOCTL_GNTDEV_UNMAP_GRANT_REF
:
647 return gntdev_ioctl_unmap_grant_ref(priv
, ptr
);
649 case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR
:
650 return gntdev_ioctl_get_offset_for_vaddr(priv
, ptr
);
652 case IOCTL_GNTDEV_SET_UNMAP_NOTIFY
:
653 return gntdev_ioctl_notify(priv
, ptr
);
656 pr_debug("priv %p, unknown cmd %x\n", priv
, cmd
);
663 static int gntdev_mmap(struct file
*flip
, struct vm_area_struct
*vma
)
665 struct gntdev_priv
*priv
= flip
->private_data
;
666 int index
= vma
->vm_pgoff
;
667 int count
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
668 struct grant_map
*map
;
669 int i
, err
= -EINVAL
;
671 if ((vma
->vm_flags
& VM_WRITE
) && !(vma
->vm_flags
& VM_SHARED
))
674 pr_debug("map %d+%d at %lx (pgoff %lx)\n",
675 index
, count
, vma
->vm_start
, vma
->vm_pgoff
);
677 spin_lock(&priv
->lock
);
678 map
= gntdev_find_map_index(priv
, index
, count
);
681 if (use_ptemod
&& map
->vma
)
683 if (use_ptemod
&& priv
->mm
!= vma
->vm_mm
) {
684 printk(KERN_WARNING
"Huh? Other mm?\n");
688 atomic_inc(&map
->users
);
690 vma
->vm_ops
= &gntdev_vmops
;
692 vma
->vm_flags
|= VM_RESERVED
|VM_DONTEXPAND
;
695 vma
->vm_flags
|= VM_DONTCOPY
|VM_PFNMAP
;
697 vma
->vm_private_data
= map
;
703 if ((vma
->vm_flags
& VM_WRITE
) &&
704 (map
->flags
& GNTMAP_readonly
))
707 map
->flags
= GNTMAP_host_map
;
708 if (!(vma
->vm_flags
& VM_WRITE
))
709 map
->flags
|= GNTMAP_readonly
;
712 spin_unlock(&priv
->lock
);
715 err
= apply_to_page_range(vma
->vm_mm
, vma
->vm_start
,
716 vma
->vm_end
- vma
->vm_start
,
717 find_grant_ptes
, map
);
719 printk(KERN_WARNING
"find_grant_ptes() failure.\n");
724 err
= map_grant_pages(map
);
729 for (i
= 0; i
< count
; i
++) {
730 err
= vm_insert_page(vma
, vma
->vm_start
+ i
*PAGE_SIZE
,
740 spin_unlock(&priv
->lock
);
744 spin_unlock(&priv
->lock
);
752 static const struct file_operations gntdev_fops
= {
753 .owner
= THIS_MODULE
,
755 .release
= gntdev_release
,
757 .unlocked_ioctl
= gntdev_ioctl
760 static struct miscdevice gntdev_miscdev
= {
761 .minor
= MISC_DYNAMIC_MINOR
,
762 .name
= "xen/gntdev",
763 .fops
= &gntdev_fops
,
766 /* ------------------------------------------------------------------ */
768 static int __init
gntdev_init(void)
775 use_ptemod
= xen_pv_domain();
777 err
= misc_register(&gntdev_miscdev
);
779 printk(KERN_ERR
"Could not register gntdev device\n");
785 static void __exit
gntdev_exit(void)
787 misc_deregister(&gntdev_miscdev
);
790 module_init(gntdev_init
);
791 module_exit(gntdev_exit
);
793 /* ------------------------------------------------------------------ */