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 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/init.h>
27 #include <linux/miscdevice.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>
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>
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");
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 "
58 static atomic_t pages_mapped = ATOMIC_INIT(0);
60 static int use_ptemod;
61 #define populate_freeable_maps use_ptemod
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 */
72 struct mmu_notifier mn;
77 /* Address relative to the start of the grant_map */
83 struct list_head next;
84 struct vm_area_struct *vma;
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;
97 static int unmap_grant_pages(struct grant_map *map, int offset, int pages);
99 /* ------------------------------------------------------------------ */
101 static void gntdev_print_maps(struct gntdev_priv *priv,
102 char *text, int text_index)
105 struct grant_map *map;
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 : "");
115 static void gntdev_free_map(struct grant_map *map)
121 free_xenballooned_pages(map->count, map->pages);
125 kfree(map->unmap_ops);
126 kfree(map->kmap_ops);
130 static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count)
132 struct grant_map *add;
135 add = kzalloc(sizeof(struct grant_map), GFP_KERNEL);
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 ||
151 if (alloc_xenballooned_pages(count, add->pages, false /* lowmem */))
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;
162 atomic_set(&add->users, 1);
167 gntdev_free_map(add);
171 static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add)
173 struct grant_map *map;
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);
180 add->index = map->index + map->count;
182 list_add_tail(&add->next, &priv->maps);
185 gntdev_print_maps(priv, "[new]", add->index);
188 static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
189 int index, int count)
191 struct grant_map *map;
193 list_for_each_entry(map, &priv->maps, next) {
194 if (map->index != index)
196 if (count && map->count != count)
203 static void gntdev_put_map(struct gntdev_priv *priv, struct grant_map *map)
208 if (!atomic_dec_and_test(&map->users))
211 atomic_sub(map->count, &pages_mapped);
213 if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
214 notify_remote_via_evtchn(map->notify.event);
215 evtchn_put(map->notify.event);
218 if (populate_freeable_maps && priv) {
219 spin_lock(&priv->lock);
220 list_del(&map->next);
221 spin_unlock(&priv->lock);
224 if (map->pages && !use_ptemod)
225 unmap_grant_pages(map, 0, map->count);
226 gntdev_free_map(map);
229 /* ------------------------------------------------------------------ */
231 static int find_grant_ptes(pte_t *pte, pgtable_t token,
232 unsigned long addr, void *data)
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;
239 BUG_ON(pgnr >= map->count);
240 pte_maddr = arbitrary_virt_to_machine(pte).maddr;
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,
250 static int map_grant_pages(struct grant_map *map)
255 /* Note: it could already be mapped */
256 if (map->map_ops[0].handle != -1)
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,
263 map->grants[i].domid);
264 gnttab_set_unmap_op(&map->unmap_ops[i], addr,
265 map->flags, -1 /* handle */);
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.
274 for (i = 0; i < map->count; i++) {
276 unsigned long address = (unsigned long)
277 pfn_to_kaddr(page_to_pfn(map->pages[i]));
280 BUG_ON(PageHighMem(map->pages[i]));
282 ptep = lookup_address(address, &level);
283 pte_maddr = arbitrary_virt_to_machine(ptep).maddr;
284 gnttab_set_map_op(&map->kmap_ops[i], pte_maddr,
289 map->grants[i].domid);
293 pr_debug("map %d+%d\n", map->index, map->count);
294 err = gnttab_map_refs(map->map_ops, use_ptemod ? map->kmap_ops : NULL,
295 map->pages, map->count);
299 for (i = 0; i < map->count; i++) {
300 if (map->map_ops[i].status)
303 BUG_ON(map->map_ops[i].handle == -1);
304 map->unmap_ops[i].handle = map->map_ops[i].handle;
305 pr_debug("map handle=%d\n", map->map_ops[i].handle);
311 static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
315 if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) {
316 int pgno = (map->notify.addr >> PAGE_SHIFT);
317 if (pgno >= offset && pgno < offset + pages) {
318 /* No need for kmap, pages are in lowmem */
319 uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno]));
320 tmp[map->notify.addr & (PAGE_SIZE-1)] = 0;
321 map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
325 err = gnttab_unmap_refs(map->unmap_ops + offset,
326 use_ptemod ? map->kmap_ops + offset : NULL, map->pages + offset,
331 for (i = 0; i < pages; i++) {
332 if (map->unmap_ops[offset+i].status)
334 pr_debug("unmap handle=%d st=%d\n",
335 map->unmap_ops[offset+i].handle,
336 map->unmap_ops[offset+i].status);
337 map->unmap_ops[offset+i].handle = -1;
342 static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
346 pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
348 /* It is possible the requested range will have a "hole" where we
349 * already unmapped some of the grants. Only unmap valid ranges.
351 while (pages && !err) {
352 while (pages && map->unmap_ops[offset].handle == -1) {
357 while (range < pages) {
358 if (map->unmap_ops[offset+range].handle == -1) {
364 err = __unmap_grant_pages(map, offset, range);
372 /* ------------------------------------------------------------------ */
374 static void gntdev_vma_open(struct vm_area_struct *vma)
376 struct grant_map *map = vma->vm_private_data;
378 pr_debug("gntdev_vma_open %p\n", vma);
379 atomic_inc(&map->users);
382 static void gntdev_vma_close(struct vm_area_struct *vma)
384 struct grant_map *map = vma->vm_private_data;
385 struct file *file = vma->vm_file;
386 struct gntdev_priv *priv = file->private_data;
388 pr_debug("gntdev_vma_close %p\n", vma);
390 /* It is possible that an mmu notifier could be running
391 * concurrently, so take priv->lock to ensure that the vma won't
392 * vanishing during the unmap_grant_pages call, since we will
393 * spin here until that completes. Such a concurrent call will
394 * not do any unmapping, since that has been done prior to
395 * closing the vma, but it may still iterate the unmap_ops list.
397 spin_lock(&priv->lock);
399 spin_unlock(&priv->lock);
401 vma->vm_private_data = NULL;
402 gntdev_put_map(priv, map);
405 static struct vm_operations_struct gntdev_vmops = {
406 .open = gntdev_vma_open,
407 .close = gntdev_vma_close,
410 /* ------------------------------------------------------------------ */
412 static void unmap_if_in_range(struct grant_map *map,
413 unsigned long start, unsigned long end)
415 unsigned long mstart, mend;
420 if (map->vma->vm_start >= end)
422 if (map->vma->vm_end <= start)
424 mstart = max(start, map->vma->vm_start);
425 mend = min(end, map->vma->vm_end);
426 pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
427 map->index, map->count,
428 map->vma->vm_start, map->vma->vm_end,
429 start, end, mstart, mend);
430 err = unmap_grant_pages(map,
431 (mstart - map->vma->vm_start) >> PAGE_SHIFT,
432 (mend - mstart) >> PAGE_SHIFT);
436 static void mn_invl_range_start(struct mmu_notifier *mn,
437 struct mm_struct *mm,
438 unsigned long start, unsigned long end)
440 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
441 struct grant_map *map;
443 spin_lock(&priv->lock);
444 list_for_each_entry(map, &priv->maps, next) {
445 unmap_if_in_range(map, start, end);
447 list_for_each_entry(map, &priv->freeable_maps, next) {
448 unmap_if_in_range(map, start, end);
450 spin_unlock(&priv->lock);
453 static void mn_invl_page(struct mmu_notifier *mn,
454 struct mm_struct *mm,
455 unsigned long address)
457 mn_invl_range_start(mn, mm, address, address + PAGE_SIZE);
460 static void mn_release(struct mmu_notifier *mn,
461 struct mm_struct *mm)
463 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
464 struct grant_map *map;
467 spin_lock(&priv->lock);
468 list_for_each_entry(map, &priv->maps, next) {
471 pr_debug("map %d+%d (%lx %lx)\n",
472 map->index, map->count,
473 map->vma->vm_start, map->vma->vm_end);
474 err = unmap_grant_pages(map, /* offset */ 0, map->count);
477 list_for_each_entry(map, &priv->freeable_maps, next) {
480 pr_debug("map %d+%d (%lx %lx)\n",
481 map->index, map->count,
482 map->vma->vm_start, map->vma->vm_end);
483 err = unmap_grant_pages(map, /* offset */ 0, map->count);
486 spin_unlock(&priv->lock);
489 static struct mmu_notifier_ops gntdev_mmu_ops = {
490 .release = mn_release,
491 .invalidate_page = mn_invl_page,
492 .invalidate_range_start = mn_invl_range_start,
495 /* ------------------------------------------------------------------ */
497 static int gntdev_open(struct inode *inode, struct file *flip)
499 struct gntdev_priv *priv;
502 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
506 INIT_LIST_HEAD(&priv->maps);
507 INIT_LIST_HEAD(&priv->freeable_maps);
508 spin_lock_init(&priv->lock);
511 priv->mm = get_task_mm(current);
516 priv->mn.ops = &gntdev_mmu_ops;
517 ret = mmu_notifier_register(&priv->mn, priv->mm);
526 flip->private_data = priv;
527 pr_debug("priv %p\n", priv);
532 static int gntdev_release(struct inode *inode, struct file *flip)
534 struct gntdev_priv *priv = flip->private_data;
535 struct grant_map *map;
537 pr_debug("priv %p\n", priv);
539 while (!list_empty(&priv->maps)) {
540 map = list_entry(priv->maps.next, struct grant_map, next);
541 list_del(&map->next);
542 gntdev_put_map(NULL /* already removed */, map);
544 WARN_ON(!list_empty(&priv->freeable_maps));
547 mmu_notifier_unregister(&priv->mn, priv->mm);
552 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
553 struct ioctl_gntdev_map_grant_ref __user *u)
555 struct ioctl_gntdev_map_grant_ref op;
556 struct grant_map *map;
559 if (copy_from_user(&op, u, sizeof(op)) != 0)
561 pr_debug("priv %p, add %d\n", priv, op.count);
562 if (unlikely(op.count <= 0))
566 map = gntdev_alloc_map(priv, op.count);
570 if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) {
571 pr_debug("can't map: over limit\n");
572 gntdev_put_map(NULL, map);
576 if (copy_from_user(map->grants, &u->refs,
577 sizeof(map->grants[0]) * op.count) != 0) {
578 gntdev_put_map(NULL, map);
582 spin_lock(&priv->lock);
583 gntdev_add_map(priv, map);
584 op.index = map->index << PAGE_SHIFT;
585 spin_unlock(&priv->lock);
587 if (copy_to_user(u, &op, sizeof(op)) != 0)
593 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
594 struct ioctl_gntdev_unmap_grant_ref __user *u)
596 struct ioctl_gntdev_unmap_grant_ref op;
597 struct grant_map *map;
600 if (copy_from_user(&op, u, sizeof(op)) != 0)
602 pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
604 spin_lock(&priv->lock);
605 map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
607 list_del(&map->next);
608 if (populate_freeable_maps)
609 list_add_tail(&map->next, &priv->freeable_maps);
612 spin_unlock(&priv->lock);
614 gntdev_put_map(priv, map);
618 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
619 struct ioctl_gntdev_get_offset_for_vaddr __user *u)
621 struct ioctl_gntdev_get_offset_for_vaddr op;
622 struct vm_area_struct *vma;
623 struct grant_map *map;
626 if (copy_from_user(&op, u, sizeof(op)) != 0)
628 pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
630 down_read(¤t->mm->mmap_sem);
631 vma = find_vma(current->mm, op.vaddr);
632 if (!vma || vma->vm_ops != &gntdev_vmops)
635 map = vma->vm_private_data;
639 op.offset = map->index << PAGE_SHIFT;
640 op.count = map->count;
644 up_read(¤t->mm->mmap_sem);
646 if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0)
651 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
653 struct ioctl_gntdev_unmap_notify op;
654 struct grant_map *map;
657 unsigned int out_event;
659 if (copy_from_user(&op, u, sizeof(op)))
662 if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
665 /* We need to grab a reference to the event channel we are going to use
666 * to send the notify before releasing the reference we may already have
667 * (if someone has called this ioctl twice). This is required so that
668 * it is possible to change the clear_byte part of the notification
669 * without disturbing the event channel part, which may now be the last
670 * reference to that event channel.
672 if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
673 if (evtchn_get(op.event_channel_port))
677 out_flags = op.action;
678 out_event = op.event_channel_port;
680 spin_lock(&priv->lock);
682 list_for_each_entry(map, &priv->maps, next) {
683 uint64_t begin = map->index << PAGE_SHIFT;
684 uint64_t end = (map->index + map->count) << PAGE_SHIFT;
685 if (op.index >= begin && op.index < end)
692 if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
693 (map->flags & GNTMAP_readonly)) {
698 out_flags = map->notify.flags;
699 out_event = map->notify.event;
701 map->notify.flags = op.action;
702 map->notify.addr = op.index - (map->index << PAGE_SHIFT);
703 map->notify.event = op.event_channel_port;
708 spin_unlock(&priv->lock);
710 /* Drop the reference to the event channel we did not save in the map */
711 if (out_flags & UNMAP_NOTIFY_SEND_EVENT)
712 evtchn_put(out_event);
717 static long gntdev_ioctl(struct file *flip,
718 unsigned int cmd, unsigned long arg)
720 struct gntdev_priv *priv = flip->private_data;
721 void __user *ptr = (void __user *)arg;
724 case IOCTL_GNTDEV_MAP_GRANT_REF:
725 return gntdev_ioctl_map_grant_ref(priv, ptr);
727 case IOCTL_GNTDEV_UNMAP_GRANT_REF:
728 return gntdev_ioctl_unmap_grant_ref(priv, ptr);
730 case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
731 return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
733 case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
734 return gntdev_ioctl_notify(priv, ptr);
737 pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
744 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
746 struct gntdev_priv *priv = flip->private_data;
747 int index = vma->vm_pgoff;
748 int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
749 struct grant_map *map;
750 int i, err = -EINVAL;
752 if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
755 pr_debug("map %d+%d at %lx (pgoff %lx)\n",
756 index, count, vma->vm_start, vma->vm_pgoff);
758 spin_lock(&priv->lock);
759 map = gntdev_find_map_index(priv, index, count);
762 if (use_ptemod && map->vma)
764 if (use_ptemod && priv->mm != vma->vm_mm) {
765 pr_warn("Huh? Other mm?\n");
769 atomic_inc(&map->users);
771 vma->vm_ops = &gntdev_vmops;
773 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
776 vma->vm_flags |= VM_DONTCOPY;
778 vma->vm_private_data = map;
784 if ((vma->vm_flags & VM_WRITE) &&
785 (map->flags & GNTMAP_readonly))
788 map->flags = GNTMAP_host_map;
789 if (!(vma->vm_flags & VM_WRITE))
790 map->flags |= GNTMAP_readonly;
793 spin_unlock(&priv->lock);
796 err = apply_to_page_range(vma->vm_mm, vma->vm_start,
797 vma->vm_end - vma->vm_start,
798 find_grant_ptes, map);
800 pr_warn("find_grant_ptes() failure.\n");
805 err = map_grant_pages(map);
810 for (i = 0; i < count; i++) {
811 err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
821 spin_unlock(&priv->lock);
825 spin_unlock(&priv->lock);
829 gntdev_put_map(priv, map);
833 static const struct file_operations gntdev_fops = {
834 .owner = THIS_MODULE,
836 .release = gntdev_release,
838 .unlocked_ioctl = gntdev_ioctl
841 static struct miscdevice gntdev_miscdev = {
842 .minor = MISC_DYNAMIC_MINOR,
843 .name = "xen/gntdev",
844 .fops = &gntdev_fops,
847 /* ------------------------------------------------------------------ */
849 static int __init gntdev_init(void)
856 use_ptemod = xen_pv_domain();
858 err = misc_register(&gntdev_miscdev);
860 pr_err("Could not register gntdev device\n");
866 static void __exit gntdev_exit(void)
868 misc_deregister(&gntdev_miscdev);
871 module_init(gntdev_init);
872 module_exit(gntdev_exit);
874 /* ------------------------------------------------------------------ */