1 /******************************************************************************
4 * Device for accessing (in user-space) pages that have been granted by other
7 * Copyright (c) 2006-2007, D G Murray.
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/sched/mm.h>
36 #include <linux/spinlock.h>
37 #include <linux/slab.h>
38 #include <linux/highmem.h>
39 #include <linux/refcount.h>
42 #include <xen/grant_table.h>
43 #include <xen/balloon.h>
44 #include <xen/gntdev.h>
45 #include <xen/events.h>
47 #include <asm/xen/hypervisor.h>
48 #include <asm/xen/hypercall.h>
50 MODULE_LICENSE("GPL");
53 MODULE_DESCRIPTION("User-space granted page access driver");
55 static int limit = 1024*1024;
56 module_param(limit, int, 0644);
57 MODULE_PARM_DESC(limit, "Maximum number of grants that may be mapped by "
60 static atomic_t pages_mapped = ATOMIC_INIT(0);
62 static int use_ptemod;
63 #define populate_freeable_maps use_ptemod
66 /* maps with visible offsets in the file descriptor */
67 struct list_head maps;
68 /* maps that are not visible; will be freed on munmap.
69 * Only populated if populate_freeable_maps == 1 */
70 struct list_head freeable_maps;
71 /* lock protects maps and freeable_maps */
74 struct mmu_notifier mn;
79 /* Address relative to the start of the grant_map */
85 struct list_head next;
86 struct vm_area_struct *vma;
91 struct unmap_notify notify;
92 struct ioctl_gntdev_grant_ref *grants;
93 struct gnttab_map_grant_ref *map_ops;
94 struct gnttab_unmap_grant_ref *unmap_ops;
95 struct gnttab_map_grant_ref *kmap_ops;
96 struct gnttab_unmap_grant_ref *kunmap_ops;
98 unsigned long pages_vm_start;
101 static int unmap_grant_pages(struct grant_map *map, int offset, int pages);
103 /* ------------------------------------------------------------------ */
105 static void gntdev_print_maps(struct gntdev_priv *priv,
106 char *text, int text_index)
109 struct grant_map *map;
111 pr_debug("%s: maps list (priv %p)\n", __func__, priv);
112 list_for_each_entry(map, &priv->maps, next)
113 pr_debug(" index %2d, count %2d %s\n",
114 map->index, map->count,
115 map->index == text_index && text ? text : "");
119 static void gntdev_free_map(struct grant_map *map)
125 gnttab_free_pages(map->count, map->pages);
129 kfree(map->unmap_ops);
130 kfree(map->kmap_ops);
131 kfree(map->kunmap_ops);
135 static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count)
137 struct grant_map *add;
140 add = kzalloc(sizeof(struct grant_map), GFP_KERNEL);
144 add->grants = kcalloc(count, sizeof(add->grants[0]), GFP_KERNEL);
145 add->map_ops = kcalloc(count, sizeof(add->map_ops[0]), GFP_KERNEL);
146 add->unmap_ops = kcalloc(count, sizeof(add->unmap_ops[0]), GFP_KERNEL);
147 add->kmap_ops = kcalloc(count, sizeof(add->kmap_ops[0]), GFP_KERNEL);
148 add->kunmap_ops = kcalloc(count, sizeof(add->kunmap_ops[0]), GFP_KERNEL);
149 add->pages = kcalloc(count, sizeof(add->pages[0]), GFP_KERNEL);
150 if (NULL == add->grants ||
151 NULL == add->map_ops ||
152 NULL == add->unmap_ops ||
153 NULL == add->kmap_ops ||
154 NULL == add->kunmap_ops ||
158 if (gnttab_alloc_pages(count, add->pages))
161 for (i = 0; i < count; i++) {
162 add->map_ops[i].handle = -1;
163 add->unmap_ops[i].handle = -1;
164 add->kmap_ops[i].handle = -1;
165 add->kunmap_ops[i].handle = -1;
170 refcount_set(&add->users, 1);
175 gntdev_free_map(add);
179 static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add)
181 struct grant_map *map;
183 list_for_each_entry(map, &priv->maps, next) {
184 if (add->index + add->count < map->index) {
185 list_add_tail(&add->next, &map->next);
188 add->index = map->index + map->count;
190 list_add_tail(&add->next, &priv->maps);
193 gntdev_print_maps(priv, "[new]", add->index);
196 static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
197 int index, int count)
199 struct grant_map *map;
201 list_for_each_entry(map, &priv->maps, next) {
202 if (map->index != index)
204 if (count && map->count != count)
211 static void gntdev_put_map(struct gntdev_priv *priv, struct grant_map *map)
216 if (!refcount_dec_and_test(&map->users))
219 atomic_sub(map->count, &pages_mapped);
221 if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
222 notify_remote_via_evtchn(map->notify.event);
223 evtchn_put(map->notify.event);
226 if (populate_freeable_maps && priv) {
227 mutex_lock(&priv->lock);
228 list_del(&map->next);
229 mutex_unlock(&priv->lock);
232 if (map->pages && !use_ptemod)
233 unmap_grant_pages(map, 0, map->count);
234 gntdev_free_map(map);
237 /* ------------------------------------------------------------------ */
239 static int find_grant_ptes(pte_t *pte, pgtable_t token,
240 unsigned long addr, void *data)
242 struct grant_map *map = data;
243 unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT;
244 int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte;
247 BUG_ON(pgnr >= map->count);
248 pte_maddr = arbitrary_virt_to_machine(pte).maddr;
251 * Set the PTE as special to force get_user_pages_fast() fall
252 * back to the slow path. If this is not supported as part of
253 * the grant map, it will be done afterwards.
255 if (xen_feature(XENFEAT_gnttab_map_avail_bits))
256 flags |= (1 << _GNTMAP_guest_avail0);
258 gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags,
259 map->grants[pgnr].ref,
260 map->grants[pgnr].domid);
261 gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags,
267 static int set_grant_ptes_as_special(pte_t *pte, pgtable_t token,
268 unsigned long addr, void *data)
270 set_pte_at(current->mm, addr, pte, pte_mkspecial(*pte));
275 static int map_grant_pages(struct grant_map *map)
280 /* Note: it could already be mapped */
281 if (map->map_ops[0].handle != -1)
283 for (i = 0; i < map->count; i++) {
284 unsigned long addr = (unsigned long)
285 pfn_to_kaddr(page_to_pfn(map->pages[i]));
286 gnttab_set_map_op(&map->map_ops[i], addr, map->flags,
288 map->grants[i].domid);
289 gnttab_set_unmap_op(&map->unmap_ops[i], addr,
290 map->flags, -1 /* handle */);
294 * Setup the map_ops corresponding to the pte entries pointing
295 * to the kernel linear addresses of the struct pages.
296 * These ptes are completely different from the user ptes dealt
297 * with find_grant_ptes.
299 for (i = 0; i < map->count; i++) {
300 unsigned long address = (unsigned long)
301 pfn_to_kaddr(page_to_pfn(map->pages[i]));
302 BUG_ON(PageHighMem(map->pages[i]));
304 gnttab_set_map_op(&map->kmap_ops[i], address,
305 map->flags | GNTMAP_host_map,
307 map->grants[i].domid);
308 gnttab_set_unmap_op(&map->kunmap_ops[i], address,
309 map->flags | GNTMAP_host_map, -1);
313 pr_debug("map %d+%d\n", map->index, map->count);
314 err = gnttab_map_refs(map->map_ops, use_ptemod ? map->kmap_ops : NULL,
315 map->pages, map->count);
319 for (i = 0; i < map->count; i++) {
320 if (map->map_ops[i].status) {
325 map->unmap_ops[i].handle = map->map_ops[i].handle;
327 map->kunmap_ops[i].handle = map->kmap_ops[i].handle;
332 static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
335 struct gntab_unmap_queue_data unmap_data;
337 if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) {
338 int pgno = (map->notify.addr >> PAGE_SHIFT);
339 if (pgno >= offset && pgno < offset + pages) {
340 /* No need for kmap, pages are in lowmem */
341 uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno]));
342 tmp[map->notify.addr & (PAGE_SIZE-1)] = 0;
343 map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
347 unmap_data.unmap_ops = map->unmap_ops + offset;
348 unmap_data.kunmap_ops = use_ptemod ? map->kunmap_ops + offset : NULL;
349 unmap_data.pages = map->pages + offset;
350 unmap_data.count = pages;
352 err = gnttab_unmap_refs_sync(&unmap_data);
356 for (i = 0; i < pages; i++) {
357 if (map->unmap_ops[offset+i].status)
359 pr_debug("unmap handle=%d st=%d\n",
360 map->unmap_ops[offset+i].handle,
361 map->unmap_ops[offset+i].status);
362 map->unmap_ops[offset+i].handle = -1;
367 static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
371 pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
373 /* It is possible the requested range will have a "hole" where we
374 * already unmapped some of the grants. Only unmap valid ranges.
376 while (pages && !err) {
377 while (pages && map->unmap_ops[offset].handle == -1) {
382 while (range < pages) {
383 if (map->unmap_ops[offset+range].handle == -1) {
389 err = __unmap_grant_pages(map, offset, range);
397 /* ------------------------------------------------------------------ */
399 static void gntdev_vma_open(struct vm_area_struct *vma)
401 struct grant_map *map = vma->vm_private_data;
403 pr_debug("gntdev_vma_open %p\n", vma);
404 refcount_inc(&map->users);
407 static void gntdev_vma_close(struct vm_area_struct *vma)
409 struct grant_map *map = vma->vm_private_data;
410 struct file *file = vma->vm_file;
411 struct gntdev_priv *priv = file->private_data;
413 pr_debug("gntdev_vma_close %p\n", vma);
415 /* It is possible that an mmu notifier could be running
416 * concurrently, so take priv->lock to ensure that the vma won't
417 * vanishing during the unmap_grant_pages call, since we will
418 * spin here until that completes. Such a concurrent call will
419 * not do any unmapping, since that has been done prior to
420 * closing the vma, but it may still iterate the unmap_ops list.
422 mutex_lock(&priv->lock);
424 mutex_unlock(&priv->lock);
426 vma->vm_private_data = NULL;
427 gntdev_put_map(priv, map);
430 static struct page *gntdev_vma_find_special_page(struct vm_area_struct *vma,
433 struct grant_map *map = vma->vm_private_data;
435 return map->pages[(addr - map->pages_vm_start) >> PAGE_SHIFT];
438 static const struct vm_operations_struct gntdev_vmops = {
439 .open = gntdev_vma_open,
440 .close = gntdev_vma_close,
441 .find_special_page = gntdev_vma_find_special_page,
444 /* ------------------------------------------------------------------ */
446 static void unmap_if_in_range(struct grant_map *map,
447 unsigned long start, unsigned long end)
449 unsigned long mstart, mend;
454 if (map->vma->vm_start >= end)
456 if (map->vma->vm_end <= start)
458 mstart = max(start, map->vma->vm_start);
459 mend = min(end, map->vma->vm_end);
460 pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
461 map->index, map->count,
462 map->vma->vm_start, map->vma->vm_end,
463 start, end, mstart, mend);
464 err = unmap_grant_pages(map,
465 (mstart - map->vma->vm_start) >> PAGE_SHIFT,
466 (mend - mstart) >> PAGE_SHIFT);
470 static void mn_invl_range_start(struct mmu_notifier *mn,
471 struct mm_struct *mm,
472 unsigned long start, unsigned long end)
474 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
475 struct grant_map *map;
477 mutex_lock(&priv->lock);
478 list_for_each_entry(map, &priv->maps, next) {
479 unmap_if_in_range(map, start, end);
481 list_for_each_entry(map, &priv->freeable_maps, next) {
482 unmap_if_in_range(map, start, end);
484 mutex_unlock(&priv->lock);
487 static void mn_invl_page(struct mmu_notifier *mn,
488 struct mm_struct *mm,
489 unsigned long address)
491 mn_invl_range_start(mn, mm, address, address + PAGE_SIZE);
494 static void mn_release(struct mmu_notifier *mn,
495 struct mm_struct *mm)
497 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
498 struct grant_map *map;
501 mutex_lock(&priv->lock);
502 list_for_each_entry(map, &priv->maps, next) {
505 pr_debug("map %d+%d (%lx %lx)\n",
506 map->index, map->count,
507 map->vma->vm_start, map->vma->vm_end);
508 err = unmap_grant_pages(map, /* offset */ 0, map->count);
511 list_for_each_entry(map, &priv->freeable_maps, next) {
514 pr_debug("map %d+%d (%lx %lx)\n",
515 map->index, map->count,
516 map->vma->vm_start, map->vma->vm_end);
517 err = unmap_grant_pages(map, /* offset */ 0, map->count);
520 mutex_unlock(&priv->lock);
523 static const struct mmu_notifier_ops gntdev_mmu_ops = {
524 .release = mn_release,
525 .invalidate_page = mn_invl_page,
526 .invalidate_range_start = mn_invl_range_start,
529 /* ------------------------------------------------------------------ */
531 static int gntdev_open(struct inode *inode, struct file *flip)
533 struct gntdev_priv *priv;
536 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
540 INIT_LIST_HEAD(&priv->maps);
541 INIT_LIST_HEAD(&priv->freeable_maps);
542 mutex_init(&priv->lock);
545 priv->mm = get_task_mm(current);
550 priv->mn.ops = &gntdev_mmu_ops;
551 ret = mmu_notifier_register(&priv->mn, priv->mm);
560 flip->private_data = priv;
561 pr_debug("priv %p\n", priv);
566 static int gntdev_release(struct inode *inode, struct file *flip)
568 struct gntdev_priv *priv = flip->private_data;
569 struct grant_map *map;
571 pr_debug("priv %p\n", priv);
573 mutex_lock(&priv->lock);
574 while (!list_empty(&priv->maps)) {
575 map = list_entry(priv->maps.next, struct grant_map, next);
576 list_del(&map->next);
577 gntdev_put_map(NULL /* already removed */, map);
579 WARN_ON(!list_empty(&priv->freeable_maps));
580 mutex_unlock(&priv->lock);
583 mmu_notifier_unregister(&priv->mn, priv->mm);
588 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
589 struct ioctl_gntdev_map_grant_ref __user *u)
591 struct ioctl_gntdev_map_grant_ref op;
592 struct grant_map *map;
595 if (copy_from_user(&op, u, sizeof(op)) != 0)
597 pr_debug("priv %p, add %d\n", priv, op.count);
598 if (unlikely(op.count <= 0))
602 map = gntdev_alloc_map(priv, op.count);
606 if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) {
607 pr_debug("can't map: over limit\n");
608 gntdev_put_map(NULL, map);
612 if (copy_from_user(map->grants, &u->refs,
613 sizeof(map->grants[0]) * op.count) != 0) {
614 gntdev_put_map(NULL, map);
618 mutex_lock(&priv->lock);
619 gntdev_add_map(priv, map);
620 op.index = map->index << PAGE_SHIFT;
621 mutex_unlock(&priv->lock);
623 if (copy_to_user(u, &op, sizeof(op)) != 0)
629 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
630 struct ioctl_gntdev_unmap_grant_ref __user *u)
632 struct ioctl_gntdev_unmap_grant_ref op;
633 struct grant_map *map;
636 if (copy_from_user(&op, u, sizeof(op)) != 0)
638 pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
640 mutex_lock(&priv->lock);
641 map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
643 list_del(&map->next);
644 if (populate_freeable_maps)
645 list_add_tail(&map->next, &priv->freeable_maps);
648 mutex_unlock(&priv->lock);
650 gntdev_put_map(priv, map);
654 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
655 struct ioctl_gntdev_get_offset_for_vaddr __user *u)
657 struct ioctl_gntdev_get_offset_for_vaddr op;
658 struct vm_area_struct *vma;
659 struct grant_map *map;
662 if (copy_from_user(&op, u, sizeof(op)) != 0)
664 pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
666 down_read(¤t->mm->mmap_sem);
667 vma = find_vma(current->mm, op.vaddr);
668 if (!vma || vma->vm_ops != &gntdev_vmops)
671 map = vma->vm_private_data;
675 op.offset = map->index << PAGE_SHIFT;
676 op.count = map->count;
680 up_read(¤t->mm->mmap_sem);
682 if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0)
687 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
689 struct ioctl_gntdev_unmap_notify op;
690 struct grant_map *map;
693 unsigned int out_event;
695 if (copy_from_user(&op, u, sizeof(op)))
698 if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
701 /* We need to grab a reference to the event channel we are going to use
702 * to send the notify before releasing the reference we may already have
703 * (if someone has called this ioctl twice). This is required so that
704 * it is possible to change the clear_byte part of the notification
705 * without disturbing the event channel part, which may now be the last
706 * reference to that event channel.
708 if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
709 if (evtchn_get(op.event_channel_port))
713 out_flags = op.action;
714 out_event = op.event_channel_port;
716 mutex_lock(&priv->lock);
718 list_for_each_entry(map, &priv->maps, next) {
719 uint64_t begin = map->index << PAGE_SHIFT;
720 uint64_t end = (map->index + map->count) << PAGE_SHIFT;
721 if (op.index >= begin && op.index < end)
728 if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
729 (map->flags & GNTMAP_readonly)) {
734 out_flags = map->notify.flags;
735 out_event = map->notify.event;
737 map->notify.flags = op.action;
738 map->notify.addr = op.index - (map->index << PAGE_SHIFT);
739 map->notify.event = op.event_channel_port;
744 mutex_unlock(&priv->lock);
746 /* Drop the reference to the event channel we did not save in the map */
747 if (out_flags & UNMAP_NOTIFY_SEND_EVENT)
748 evtchn_put(out_event);
753 #define GNTDEV_COPY_BATCH 16
755 struct gntdev_copy_batch {
756 struct gnttab_copy ops[GNTDEV_COPY_BATCH];
757 struct page *pages[GNTDEV_COPY_BATCH];
758 s16 __user *status[GNTDEV_COPY_BATCH];
760 unsigned int nr_pages;
763 static int gntdev_get_page(struct gntdev_copy_batch *batch, void __user *virt,
764 bool writeable, unsigned long *gfn)
766 unsigned long addr = (unsigned long)virt;
768 unsigned long xen_pfn;
771 ret = get_user_pages_fast(addr, 1, writeable, &page);
775 batch->pages[batch->nr_pages++] = page;
777 xen_pfn = page_to_xen_pfn(page) + XEN_PFN_DOWN(addr & ~PAGE_MASK);
778 *gfn = pfn_to_gfn(xen_pfn);
783 static void gntdev_put_pages(struct gntdev_copy_batch *batch)
787 for (i = 0; i < batch->nr_pages; i++)
788 put_page(batch->pages[i]);
792 static int gntdev_copy(struct gntdev_copy_batch *batch)
796 gnttab_batch_copy(batch->ops, batch->nr_ops);
797 gntdev_put_pages(batch);
800 * For each completed op, update the status if the op failed
801 * and all previous ops for the segment were successful.
803 for (i = 0; i < batch->nr_ops; i++) {
804 s16 status = batch->ops[i].status;
807 if (status == GNTST_okay)
810 if (__get_user(old_status, batch->status[i]))
813 if (old_status != GNTST_okay)
816 if (__put_user(status, batch->status[i]))
824 static int gntdev_grant_copy_seg(struct gntdev_copy_batch *batch,
825 struct gntdev_grant_copy_segment *seg,
831 * Disallow local -> local copies since there is only space in
832 * batch->pages for one page per-op and this would be a very
833 * expensive memcpy().
835 if (!(seg->flags & (GNTCOPY_source_gref | GNTCOPY_dest_gref)))
838 /* Can't cross page if source/dest is a grant ref. */
839 if (seg->flags & GNTCOPY_source_gref) {
840 if (seg->source.foreign.offset + seg->len > XEN_PAGE_SIZE)
843 if (seg->flags & GNTCOPY_dest_gref) {
844 if (seg->dest.foreign.offset + seg->len > XEN_PAGE_SIZE)
848 if (put_user(GNTST_okay, status))
851 while (copied < seg->len) {
852 struct gnttab_copy *op;
858 if (batch->nr_ops >= GNTDEV_COPY_BATCH) {
859 ret = gntdev_copy(batch);
864 len = seg->len - copied;
866 op = &batch->ops[batch->nr_ops];
869 if (seg->flags & GNTCOPY_source_gref) {
870 op->source.u.ref = seg->source.foreign.ref;
871 op->source.domid = seg->source.foreign.domid;
872 op->source.offset = seg->source.foreign.offset + copied;
873 op->flags |= GNTCOPY_source_gref;
875 virt = seg->source.virt + copied;
876 off = (unsigned long)virt & ~XEN_PAGE_MASK;
877 len = min(len, (size_t)XEN_PAGE_SIZE - off);
879 ret = gntdev_get_page(batch, virt, false, &gfn);
883 op->source.u.gmfn = gfn;
884 op->source.domid = DOMID_SELF;
885 op->source.offset = off;
888 if (seg->flags & GNTCOPY_dest_gref) {
889 op->dest.u.ref = seg->dest.foreign.ref;
890 op->dest.domid = seg->dest.foreign.domid;
891 op->dest.offset = seg->dest.foreign.offset + copied;
892 op->flags |= GNTCOPY_dest_gref;
894 virt = seg->dest.virt + copied;
895 off = (unsigned long)virt & ~XEN_PAGE_MASK;
896 len = min(len, (size_t)XEN_PAGE_SIZE - off);
898 ret = gntdev_get_page(batch, virt, true, &gfn);
902 op->dest.u.gmfn = gfn;
903 op->dest.domid = DOMID_SELF;
904 op->dest.offset = off;
910 batch->status[batch->nr_ops] = status;
917 static long gntdev_ioctl_grant_copy(struct gntdev_priv *priv, void __user *u)
919 struct ioctl_gntdev_grant_copy copy;
920 struct gntdev_copy_batch batch;
924 if (copy_from_user(©, u, sizeof(copy)))
930 for (i = 0; i < copy.count; i++) {
931 struct gntdev_grant_copy_segment seg;
933 if (copy_from_user(&seg, ©.segments[i], sizeof(seg))) {
938 ret = gntdev_grant_copy_seg(&batch, &seg, ©.segments[i].status);
945 ret = gntdev_copy(&batch);
949 gntdev_put_pages(&batch);
953 static long gntdev_ioctl(struct file *flip,
954 unsigned int cmd, unsigned long arg)
956 struct gntdev_priv *priv = flip->private_data;
957 void __user *ptr = (void __user *)arg;
960 case IOCTL_GNTDEV_MAP_GRANT_REF:
961 return gntdev_ioctl_map_grant_ref(priv, ptr);
963 case IOCTL_GNTDEV_UNMAP_GRANT_REF:
964 return gntdev_ioctl_unmap_grant_ref(priv, ptr);
966 case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
967 return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
969 case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
970 return gntdev_ioctl_notify(priv, ptr);
972 case IOCTL_GNTDEV_GRANT_COPY:
973 return gntdev_ioctl_grant_copy(priv, ptr);
976 pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
983 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
985 struct gntdev_priv *priv = flip->private_data;
986 int index = vma->vm_pgoff;
987 int count = vma_pages(vma);
988 struct grant_map *map;
989 int i, err = -EINVAL;
991 if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
994 pr_debug("map %d+%d at %lx (pgoff %lx)\n",
995 index, count, vma->vm_start, vma->vm_pgoff);
997 mutex_lock(&priv->lock);
998 map = gntdev_find_map_index(priv, index, count);
1001 if (use_ptemod && map->vma)
1003 if (use_ptemod && priv->mm != vma->vm_mm) {
1004 pr_warn("Huh? Other mm?\n");
1008 refcount_inc(&map->users);
1010 vma->vm_ops = &gntdev_vmops;
1012 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP | VM_MIXEDMAP;
1015 vma->vm_flags |= VM_DONTCOPY;
1017 vma->vm_private_data = map;
1023 if ((vma->vm_flags & VM_WRITE) &&
1024 (map->flags & GNTMAP_readonly))
1025 goto out_unlock_put;
1027 map->flags = GNTMAP_host_map;
1028 if (!(vma->vm_flags & VM_WRITE))
1029 map->flags |= GNTMAP_readonly;
1032 mutex_unlock(&priv->lock);
1035 err = apply_to_page_range(vma->vm_mm, vma->vm_start,
1036 vma->vm_end - vma->vm_start,
1037 find_grant_ptes, map);
1039 pr_warn("find_grant_ptes() failure.\n");
1044 err = map_grant_pages(map);
1049 for (i = 0; i < count; i++) {
1050 err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
1058 * If the PTEs were not made special by the grant map
1059 * hypercall, do so here.
1061 * This is racy since the mapping is already visible
1062 * to userspace but userspace should be well-behaved
1063 * enough to not touch it until the mmap() call
1066 if (!xen_feature(XENFEAT_gnttab_map_avail_bits)) {
1067 apply_to_page_range(vma->vm_mm, vma->vm_start,
1068 vma->vm_end - vma->vm_start,
1069 set_grant_ptes_as_special, NULL);
1072 map->pages_vm_start = vma->vm_start;
1078 mutex_unlock(&priv->lock);
1082 mutex_unlock(&priv->lock);
1086 gntdev_put_map(priv, map);
1090 static const struct file_operations gntdev_fops = {
1091 .owner = THIS_MODULE,
1092 .open = gntdev_open,
1093 .release = gntdev_release,
1094 .mmap = gntdev_mmap,
1095 .unlocked_ioctl = gntdev_ioctl
1098 static struct miscdevice gntdev_miscdev = {
1099 .minor = MISC_DYNAMIC_MINOR,
1100 .name = "xen/gntdev",
1101 .fops = &gntdev_fops,
1104 /* ------------------------------------------------------------------ */
1106 static int __init gntdev_init(void)
1113 use_ptemod = !xen_feature(XENFEAT_auto_translated_physmap);
1115 err = misc_register(&gntdev_miscdev);
1117 pr_err("Could not register gntdev device\n");
1123 static void __exit gntdev_exit(void)
1125 misc_deregister(&gntdev_miscdev);
1128 module_init(gntdev_init);
1129 module_exit(gntdev_exit);
1131 /* ------------------------------------------------------------------ */