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);
198 evtchn_put(map->notify.event);
203 unmap_grant_pages(map, 0, map->count);
205 free_xenballooned_pages(map->count, map->pages);
210 kfree(map->unmap_ops);
214 /* ------------------------------------------------------------------ */
216 static int find_grant_ptes(pte_t *pte, pgtable_t token,
217 unsigned long addr, void *data)
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;
224 BUG_ON(pgnr >= map->count);
225 pte_maddr = arbitrary_virt_to_machine(pte).maddr;
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,
235 static int map_grant_pages(struct grant_map *map)
240 /* Note: it could already be mapped */
241 if (map->map_ops[0].handle != -1)
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,
248 map->grants[i].domid);
249 gnttab_set_unmap_op(&map->unmap_ops[i], addr,
250 map->flags, -1 /* handle */);
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.
259 for (i = 0; i < map->count; i++) {
261 unsigned long address = (unsigned long)
262 pfn_to_kaddr(page_to_pfn(map->pages[i]));
265 BUG_ON(PageHighMem(map->pages[i]));
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,
274 map->grants[i].domid);
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);
284 for (i = 0; i < map->count; i++) {
285 if (map->map_ops[i].status)
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);
296 static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
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);
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;
317 err = gnttab_unmap_refs(map->unmap_ops + offset, map->pages + offset,
322 for (i = 0; i < pages; i++) {
323 if (map->unmap_ops[offset+i].status)
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;
333 static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
337 pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
339 /* It is possible the requested range will have a "hole" where we
340 * already unmapped some of the grants. Only unmap valid ranges.
342 while (pages && !err) {
343 while (pages && map->unmap_ops[offset].handle == -1) {
348 while (range < pages) {
349 if (map->unmap_ops[offset+range].handle == -1) {
355 err = __unmap_grant_pages(map, offset, range);
363 /* ------------------------------------------------------------------ */
365 static void gntdev_vma_open(struct vm_area_struct *vma)
367 struct grant_map *map = vma->vm_private_data;
369 pr_debug("gntdev_vma_open %p\n", vma);
370 atomic_inc(&map->users);
373 static void gntdev_vma_close(struct vm_area_struct *vma)
375 struct grant_map *map = vma->vm_private_data;
377 pr_debug("gntdev_vma_close %p\n", vma);
379 vma->vm_private_data = NULL;
383 static struct vm_operations_struct gntdev_vmops = {
384 .open = gntdev_vma_open,
385 .close = gntdev_vma_close,
388 /* ------------------------------------------------------------------ */
390 static void mn_invl_range_start(struct mmu_notifier *mn,
391 struct mm_struct *mm,
392 unsigned long start, unsigned long end)
394 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
395 struct grant_map *map;
396 unsigned long mstart, mend;
399 spin_lock(&priv->lock);
400 list_for_each_entry(map, &priv->maps, next) {
403 if (map->vma->vm_start >= end)
405 if (map->vma->vm_end <= start)
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);
418 spin_unlock(&priv->lock);
421 static void mn_invl_page(struct mmu_notifier *mn,
422 struct mm_struct *mm,
423 unsigned long address)
425 mn_invl_range_start(mn, mm, address, address + PAGE_SIZE);
428 static void mn_release(struct mmu_notifier *mn,
429 struct mm_struct *mm)
431 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
432 struct grant_map *map;
435 spin_lock(&priv->lock);
436 list_for_each_entry(map, &priv->maps, next) {
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);
445 spin_unlock(&priv->lock);
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,
454 /* ------------------------------------------------------------------ */
456 static int gntdev_open(struct inode *inode, struct file *flip)
458 struct gntdev_priv *priv;
461 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
465 INIT_LIST_HEAD(&priv->maps);
466 spin_lock_init(&priv->lock);
469 priv->mm = get_task_mm(current);
474 priv->mn.ops = &gntdev_mmu_ops;
475 ret = mmu_notifier_register(&priv->mn, priv->mm);
484 flip->private_data = priv;
485 pr_debug("priv %p\n", priv);
490 static int gntdev_release(struct inode *inode, struct file *flip)
492 struct gntdev_priv *priv = flip->private_data;
493 struct grant_map *map;
495 pr_debug("priv %p\n", priv);
497 while (!list_empty(&priv->maps)) {
498 map = list_entry(priv->maps.next, struct grant_map, next);
499 list_del(&map->next);
504 mmu_notifier_unregister(&priv->mn, priv->mm);
509 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
510 struct ioctl_gntdev_map_grant_ref __user *u)
512 struct ioctl_gntdev_map_grant_ref op;
513 struct grant_map *map;
516 if (copy_from_user(&op, u, sizeof(op)) != 0)
518 pr_debug("priv %p, add %d\n", priv, op.count);
519 if (unlikely(op.count <= 0))
523 map = gntdev_alloc_map(priv, op.count);
527 if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) {
528 pr_debug("can't map: over limit\n");
533 if (copy_from_user(map->grants, &u->refs,
534 sizeof(map->grants[0]) * op.count) != 0) {
539 spin_lock(&priv->lock);
540 gntdev_add_map(priv, map);
541 op.index = map->index << PAGE_SHIFT;
542 spin_unlock(&priv->lock);
544 if (copy_to_user(u, &op, sizeof(op)) != 0)
550 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
551 struct ioctl_gntdev_unmap_grant_ref __user *u)
553 struct ioctl_gntdev_unmap_grant_ref op;
554 struct grant_map *map;
557 if (copy_from_user(&op, u, sizeof(op)) != 0)
559 pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
561 spin_lock(&priv->lock);
562 map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
564 list_del(&map->next);
567 spin_unlock(&priv->lock);
573 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
574 struct ioctl_gntdev_get_offset_for_vaddr __user *u)
576 struct ioctl_gntdev_get_offset_for_vaddr op;
577 struct vm_area_struct *vma;
578 struct grant_map *map;
580 if (copy_from_user(&op, u, sizeof(op)) != 0)
582 pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
584 vma = find_vma(current->mm, op.vaddr);
585 if (!vma || vma->vm_ops != &gntdev_vmops)
588 map = vma->vm_private_data;
592 op.offset = map->index << PAGE_SHIFT;
593 op.count = map->count;
595 if (copy_to_user(u, &op, sizeof(op)) != 0)
600 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
602 struct ioctl_gntdev_unmap_notify op;
603 struct grant_map *map;
606 unsigned int out_event;
608 if (copy_from_user(&op, u, sizeof(op)))
611 if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
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.
621 if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
622 if (evtchn_get(op.event_channel_port))
626 out_flags = op.action;
627 out_event = op.event_channel_port;
629 spin_lock(&priv->lock);
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)
641 if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
642 (map->flags & GNTMAP_readonly)) {
647 out_flags = map->notify.flags;
648 out_event = map->notify.event;
650 map->notify.flags = op.action;
651 map->notify.addr = op.index - (map->index << PAGE_SHIFT);
652 map->notify.event = op.event_channel_port;
657 spin_unlock(&priv->lock);
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);
666 static long gntdev_ioctl(struct file *flip,
667 unsigned int cmd, unsigned long arg)
669 struct gntdev_priv *priv = flip->private_data;
670 void __user *ptr = (void __user *)arg;
673 case IOCTL_GNTDEV_MAP_GRANT_REF:
674 return gntdev_ioctl_map_grant_ref(priv, ptr);
676 case IOCTL_GNTDEV_UNMAP_GRANT_REF:
677 return gntdev_ioctl_unmap_grant_ref(priv, ptr);
679 case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
680 return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
682 case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
683 return gntdev_ioctl_notify(priv, ptr);
686 pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
693 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
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;
701 if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
704 pr_debug("map %d+%d at %lx (pgoff %lx)\n",
705 index, count, vma->vm_start, vma->vm_pgoff);
707 spin_lock(&priv->lock);
708 map = gntdev_find_map_index(priv, index, count);
711 if (use_ptemod && map->vma)
713 if (use_ptemod && priv->mm != vma->vm_mm) {
714 printk(KERN_WARNING "Huh? Other mm?\n");
718 atomic_inc(&map->users);
720 vma->vm_ops = &gntdev_vmops;
722 vma->vm_flags |= VM_RESERVED|VM_DONTEXPAND;
725 vma->vm_flags |= VM_DONTCOPY|VM_PFNMAP;
727 vma->vm_private_data = map;
733 if ((vma->vm_flags & VM_WRITE) &&
734 (map->flags & GNTMAP_readonly))
737 map->flags = GNTMAP_host_map;
738 if (!(vma->vm_flags & VM_WRITE))
739 map->flags |= GNTMAP_readonly;
742 spin_unlock(&priv->lock);
745 err = apply_to_page_range(vma->vm_mm, vma->vm_start,
746 vma->vm_end - vma->vm_start,
747 find_grant_ptes, map);
749 printk(KERN_WARNING "find_grant_ptes() failure.\n");
754 err = map_grant_pages(map);
759 for (i = 0; i < count; i++) {
760 err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
770 spin_unlock(&priv->lock);
774 spin_unlock(&priv->lock);
782 static const struct file_operations gntdev_fops = {
783 .owner = THIS_MODULE,
785 .release = gntdev_release,
787 .unlocked_ioctl = gntdev_ioctl
790 static struct miscdevice gntdev_miscdev = {
791 .minor = MISC_DYNAMIC_MINOR,
792 .name = "xen/gntdev",
793 .fops = &gntdev_fops,
796 /* ------------------------------------------------------------------ */
798 static int __init gntdev_init(void)
805 use_ptemod = xen_pv_domain();
807 err = misc_register(&gntdev_miscdev);
809 printk(KERN_ERR "Could not register gntdev device\n");
815 static void __exit gntdev_exit(void)
817 misc_deregister(&gntdev_miscdev);
820 module_init(gntdev_init);
821 module_exit(gntdev_exit);
823 /* ------------------------------------------------------------------ */